I went to see it a second time, this time at the spectacular TCL Chinese Theater in Hollywood in IMAX 70mm. That venue can make even the worst movie seem special, but I did enjoy The Odyssey a lot more this time. I no longer expected it to be clever or deep. I just let myself enjoy the cinematic wizardry. And now I don’t ever need to see it again.

In short, I’ve had a lot of…feelings…about this movie over the past month. I realized that the main one is grief. I have been grieving Nolan’s treatment of my favorite epic. Indeed, I have experienced all five stages of grief:

1. Denial—This is not my beautiful Odyssey!

2. Anger—This grandiose spectacle is intellectually shallow!

3. Negotiation—Maybe Nolan meant to do that as a clever meta-commentary?

4. Depression—No, it’s just a big-budget vanity project.

5. Acceptance—Actually, it’s an entertaining summer blockbuster.

Really, the movie poster tells you everything you need to know. Above the Chinese Theater and on billboards all over Los Angeles, we see Matt Damon’s muscle-bound Odysseus striking a superhero pose in front of some badass giants clad in steel armor with sparks raining down. It doesn’t make any sense, but it looks really cool. The words “[EVERY FRAME IN] IMAX 70MM” are bigger than the name of the movie. (And in case you missed that, you are reminded at the bottom, “SHOT ENTIRELY WITH IMAX FILM CAMERAS.”) The movie is spectacle over story.

First, the Lastygonian scene is a minor episode in Homer’s poem (55 lines out of 12,110 total). But Nolan makes it a centerpiece of the film, and certainly of the marketing campaign, as one of its most spectacular set pieces. So, you know right from the outset that Nolan’s The Odyssey is not Homer’s Odyssey. Sure enough, his Odysseus is not tricky, Athena is not inspiring, Circe is not sexy, Helen is not wise, Calypso is reduced to a therapist, the cyclops is not duped, Nausicaa and the fabulous Phaecians are gone, Nestor’s Pylos becomes a gratuitous fight scene, the gods are just hallucinations, Telemachus never grows up, Odysseus doesn’t murder every last suitor, Telemachus never viciously hangs a dozen slaves, Penelope never gets the last laugh on her tricky husband, Odysseus never insults his father, and Ithaca never descends into a civil war that only the divine intervention of Athena can stop. Instead of an epic aporia or doubts about the truth, Nolan gives us a fairy tale in which good people are mostly good, bad people (and monsters) are mostly bad, and the victorious king and faithful queen sail off into the sunset to explain what’s wrong with America. Whatever this is, it is not Homer’s Odyssey. And it is certainly not the movie that I expected Nolan to make.

That’s Denial.

Second, in Homer’s Lastygonian episode, Odysseus saves his own skin by anchoring his own flagship outside the deadly fjord and never warns his men of the danger. Eleven shiploads (something like 484 souls, not the few dozen on the two ships that get wrecked in the film) are thus doomed by the commander’s negligence. Indeed, the mythological Odysseus is a complicated asshole who rapes, pillages, lies, backstabs, slays, schemes, and screws his way home. He is also very clever, eloquent, strategic, and mechanically inclined. But Matt Damon’s Odysseus is just a sad, confused, boring, inarticulate, loyal, basically good guy. He is also handy with a bow and lucky enough to get off the beach. Indeed, what really dies on the Lastrygonian beach is Nolan’s pretention to seriousness. The script preaches that war is traumatizing, but Nolan omits the most horrific parts of Homer’s epic that could easily make that point. Instead he gives a conventional revenge tale that glorifies violence by making it fun to watch. Nolan’s interpretation is a terrible bargain: he trades all the poem’s pathos, eros, subtlety, wordplay, complexity, and moral ambiguity for flat dialogue, cardboard characters, heavy-handed themes, moralistic sermons, and blaring action scenes. Then he injects modern values (Christian charity, mercy, and war guilt) and relies on discredited historical theories (maritime marauders cause a dark age) that make no sense in the Homeric world. No wonder Homerists such as Emily Wilson, Daniel Mendelsohn, and Joel Christensen have written such negative reviews. The movie is a betrayal of all that is good in Homer.

That’s Anger.

Third, the Lastrygonian giants are clad in medieval-like steel-plated armor, even though Homer’s Odyssey is set in an imagined Bronze Age (or the end of it). Is Nolan trying to say something about the future devouring the past? Or about advanced technology destroying a civilization? Either way, this is a movie that will play fast and loose with time and space, mixing modern elements and different cultures together. Circe is a Celtic witch. Odysseus sails a Viking ship. Hades is in Iceland. Ithacans look like extras from Game of Thrones. The cyclops is a monster from a del Toro horror show. “Zeus’s Law” is the Christian golden rule. Greek civilization is the American liberal order. The Odyssey is so fruit-basket-upset in its stylistic elements, and Nolan makes so many choices at odds with Homer, that maybe his subversion of tradition is precisely the point. What if all the things that Homerists are complaining about are in fact deliberate directorial choices? What if Nolan is commenting on the fact that Homer’s Odyssey is already a mishmash of imagined elements from ancient Greece by offering up a modern mishmash? What if Nolan is commenting on all the scholarly disagreement about Homer by offering up an interpretation at odds with Homerist conventional wisdom? What if Nolan is commenting on the state-of-the-art entertainment value of singing bards in ancient Greece by giving us an entertaining state-of-the-art IMAX blockbuster? What if Matt Damon’s boring Odysseus is actually Nolan’s little joke? What if Nolan himself is the unreliable narrator, the real trickster in this movie? If so then then The Odyssey is a brilliant meta-Odyssey, a movie that’s actually good because it seems bad!

That’s Negotiation.

Fourth, haven’t we seen the Lastrygonian scene somewhere before? Oh right, this is Dunkirk, an army trapped on the beach. And also trapped in a wooden horse. And also trapped on an island. And also trapped in a palace. Come to think of it, Agamemnon’s armor sure looks a lot like Batman in The Dark Knight, who is also a sin-eating aristocrat like Odysseus trying to make a fallen world right again. Much like the guy in Memento, Odysseus can’t remember much about his past. He can’t remember that he loses most of his crew on a long journey to save his home like Matthew McConaughey in Interstellar. Then he washes up on the beach in Ithaca just like Dominic Cobb washes up in Inception. Nolan beats you over the head with Oppenheimer references to the great inventor of a technological weapon who feels pangs of guilt ever after. And I suppose the aforementioned jumble of time and space has something of Tenet about it. Matt Damon’s disguises aren’t as good as those in The Prestige, and Christian Bale would have been a way better Odysseus. But, for the most part, The Odyssey is basically a greatest-hits compilation of Nolan’s other movies. Nolan has confessed that he has been making The Odyssey in all his other movies, but I think he has that backwards. And just like in Nolan’s other movies, the watered-down, disempowered female characters in The Odyssey (to say nothing of all the interesting women in Homer’s Odyssey that Nolan simply omitted such as Nausicaa, Arete, Anticlea, and Leucothea) only seem to exist as foils for the great men. I was being way too indulgent in my negotiation phase. While I was right to see Nolan at the center of his own movie, it is too clever by half to style him as some sort of meta-Odysseus who makes sense of our world by subverting it. No, Occam’s Razor suggests that the movie that Nolan made is indeed just what it appears to be (pace Wilson and Mendelsohn and Christensen): a visually spectacular, emotionally superficial vanity project. He had an unlimited budget and all kinds of technological toys, and he used them to make a monument to himself. What a lost opportunity.

That’s Depression.

Fifth, the teenaged boy in me thought the Lastrygonian scene was pretty cool. And the marketing team of The Odyssey is speaking to that boy. The giants are a mashup of storm troopers from Star Wars and Spartans from 300. The costume design, fight choreography, and cinematography are all pretty slick. Nolan’s use of tall actors, short doubles, practical effects, and camera tricks in the Lastrygonian scene make everything look real, which a nice contrast to all that CGI slop we usually see. And did I mention, EVERY FRAME IN IMAX 70MM!! Let’s face it, this is a gorgeous movie and a major technical accomplishment in cinema history. It’s popular for a reason. It is fun to watch, and it doesn’t ask too much of its audience. It offers a few serious nods to heavy themes so we can feel smart, as if we’re encountering the classics without the discomfot of, you know, reading them, but then it gets right along with the action and adventure. At the end of the day, the best way to think of The Odyssey is as a superhero fantasy movie. It may not be as good as Guardians of the Galaxy or Into the Spiderverse, but it’s better than most of the MCU and almost all the DCU. As a spear and sandal flick, it’s not as good as Gladiator or Ben Hur, but it’s a hell of a lot better than Clash of the Titans or Noah. It’s not as clever as previous Odyssey adaptations like Kubrick’s 2001 or the Coen Bros.’ O Brother Where Art Thou, and we can argue about Pasolini’s The Return (where Ralph Fiennes does a much better job of PTSD Odysseus than Matt Damon), but it’s better than the cheesy 1990s TV version of The Odyssey. It’s not Nolan’s best film, because it’s trying too hard to be all of them, but it is certainly his prettiest one yet. The Odyssey is definitely going to win a lot of Oscars. All its technical awards will be well deserved. Best Director is reasonable for orchestrating a unique cinematic experience. Actor awards to Anne Hathaway, Samantha Morton, or John Leguizamo are totally legit. I accept that it will probably win Best Picture, too, because this is the same Academy that picked One Battle After Another over Sinners, because Hollywood values pretentious spectacle over thoughtful substance. After all, The Odyssey is just what a summer blockbuster should be—big, brash, entertaining, and easy on the brain.

And that’s Acceptance.

[edited for typos and an egregious math error, not that anyone really knows how many people were on Odysseus’s eleven ships in a totally made up story, but assuming 50 men on a penteconter, 11 ships means 550 souls, although I suppose we need to subtract the 6 men on each ship that the Cicones killed, so let’s say 484 died in the Lastrygonian fjord.]

This is a longish post on the relationship between Homer’s Odyssey (which is very different from Nolan’s The Odyssey) and AI. I have been pondering this relationship a lot in recent years, but I haven’t written publicly on it. My interest in the Odyssey predates Nolan’s movie, as does my interest in deception generally. This blog is titled Dolos, after all, which is Greek for deception. A professional interest in hacking leads naturally to a literary interest in Odysseus, the archetypal hacker of Greek mythology. Odysseus is the clever architect of the Trojan Horse, and modern cyber trojans rely on an odyssean logic of deception as well.

If watching The Odyssey has not yet convinced you to read or reread the Odyssey, I hope to give you seven or so more reasons. I think that the fantastic episodes of Homer’s epic can teach us something important about our modern experience with AI, and the culture of tech more generally. If you are worried about AI, you could do worse that read the Odyssey.

Prologue. Pining for Authenticity

But first, a brief comment on The Odyssey. I have mixed feelings about Nolan’s film, which is undoubtedly an entertaining cinematic spectacle, but it is also a disappointing interpretation of my favorite epic poem. On my first viewing, I was surprised and confused to see Homer’s eloquent, mendacious, cruel, philandering protagonist replaced with Matt Damon’s inarticulate, confused, earnest, faithful everyman, as well as to see Homer’s wonderfully charismatic female characters either significantly diminished (Athena, Calypso, Circe, Penelope, Helen) or altogether omitted (Arete, Leucothea, Nausicaa, Anticlea) in Nolan’s manly epic. “The Odyssey” would have been better entitled “An Odyssey” or “My Odyssey” since the filmmaker himself rather than Damon’s Odysseus turns out to be the unreliable narrator. My response is thus much closer to the negative reviews by Emily Wilson and Daniel Mendelsohn, who are both brilliant translators of Homer, than the modal response to the film, which has been wildly positive.

But my goal in this post is not to ccritique Nolan’s creative choices, which are obviously working well for lots of smart people. Here I just want to speculate that the tremendous popularity of The Odyssey has something to do with popular insecurity about AI. Specifically, I think The Odyssey resonates in our AI moment for at least three reasons.

First, Homer’s Odyssey is one of the most compelling poetic meditations on the human condition of all time, and it emerged (through a long process or oral and textual evolution) long before modern technology. Nolan’s retelling of Homer’s famous tale reminds us of the importance of telling stories—and reinterpreting them—in any era. And through the movie, Homer reminds audiences of our essential humanity—with all its suffering and confusion—in an era when many worry that AI may replace humanity. I personally think concerns of AI-substitution tend to be exaggerated, but the anxiety is real, as is the growing backlash to AI and its energy intensive infrastructure. In this climate of AI angst, suspect that part of the appeal of Nolan’s The Odyssey is that it offers a sort of connection to something ancient and authentic like Homer’s Odyssey (even if that connection is, ironically, rather imaginary given Nolan’s unfaithfulness to core Homeric themes, as Wilson and Mendelsohn point out).

Second, Nolan’s The Odyssey offers a rebuke to the encroachment of AI in everything. Nolan gives us good old-fashioned moviemaking while large language models give us slop. The film has been rightly praised for its emphasis on practical effects over digital graphics, and it deserves all the technical award that it will win. Filmed entirely in IMAX, the film has an exciting, immersive, realistic quality. The cyclops is a giant animatronic puppet in a real cave that makes the men seem tiny. The actors row an actual wooden ship on a storm-tossed sea that almost makes you feel seasick. Greek soldiers stuffed inside an actual wooden horse violently sack and burn a life-sized city. Indeed, the central irony of Nolan’s movie is that it is an extremely realistic depiction of a totally unfaithful interpretation of a completely imaginary epic. But this realism is a joy to watch, especially in IMAX 70mm. In an era when CGI superhero movies are flat and disappointing, I think The Odyssey resonates with audiences who crave a more authentic style of moviemaking (with similar caveats about imagined authenticity).

Third, AI is (widely believed to be) a world-changing technology with the potential of destroying our world. Many of Nolan’s films like Oppenheimer and Tenet consider the dangers of such technologies (and the angst he imagines their great inventors experiencing). It is not hard to see the Trojan Horse (or the atomic bomb, or the power to reverse time) as a metaphor for AI in this sense. In a recent interview Nolan says as much, “I think AI is a Trojan horse that everybody knows the Greeks are inside….I’ve never seen a technology advancing so rapidly, so completely rejected by the public.” The Trojan Horse was an effective ruse de guerre because the Trojans believed that it was a tribute to Athena from the defeated Greeks; unaware that Athena had cynically inspired the Greeks to use it as a ruse, the Trojans thus unwittingly became accomplices in their own downfall. But with AI, Nolan quips, “it’s a transparent horse. It’s made of glass. Everybody can see what’s going on inside.” Indeed, the modern gifts of the AI industry are increasingly met with suspicion, if not outright rejection.

The Odyssey thus hits a chord with twenty-first century audiences because (1) it makes the classic human experience meaningful, (2) eschews the use of AI or obviously fake CGI, and (3) meditates on the dangers of advanced technology. If I am right, then I think this connection is worth exploring further. But It is worth exploring in Homer’s original, i.e., the Odyssey rather than The Odyssey, and not just because I like it a lot more than Nolan’s movie (apparently I am pining for authenticity too). For all its faults, one of the great things about Nolan’s The Odyssey is that the blockbuster is helping people to rediscover Homer’s original. Apparently, publishers are struggling to keep up with demand for the lively version by Emily Wilson, who will be making a mint on a movie she dislikes. I think that’s all great, both the interest and the irony, because I think the Odyssey (rather than The Odyssey) can help us think smartly (and feel deeply) about humanity in the age of AI.

Athena (at the Getty Center in Los Angeles) looks askance at her portrayal by Hollywood while pondering the role of creativity and craft in modern tech

Chapter 1. The Automation Crisis

The rediscovery of Homer is extremely timely. Human civilization is facing grave political, economic, cultural, and environmental challenges with automation (and has been facing them in one form of another for nearly a century). Ironically, ancient poetry has the power to help us make sense of these modern challenges. Put differently, the most important things about AI actually have less to do with changing technology than enduring humanity. (I think this is also the key to understanding Pope Leo’s encyclical on AI). If you really want to know what is at stake for humanity with AI, then a good place to start is with stories from a time long before AI that humans have continued to find meaningful throughout all the momentous technological changes since. The rest of this essay will thus follow the arc of Homer’s Odyssey to show you what I mean.

Homer gave us an unreliable narrator long before the internet gave us unreliable content. Fond of stratagem and slippery words, Odysseus is the mythic archetype for every conman or hacker who has used brains over brawn. Homer’s clever hero wins the Trojan War with the ruse of the wooden horse, but he also provokes the gods and struggles to get home. Today clever entrepreneurs build smart machines with godlike potential, but we are all struggling to feel at home with AI.

In her brilliant translation of the Odyssey, Emily Wilson describes Odysseus as “a complicated man.” The Greek phrase andra polutropon literally means “man of many ways,” but Wilson’s “complicated” nicely captures the moral ambivalence of Homer’s hacker. AI is a similarly complicated technology, a tool of many uses and abuses with ambiguous import for humanity.

The Odyssey is ostensibly about a man who returns from war. Odysseus leaves his home to fight a ten-year war in Troy, which is the subject of Homer’s Iliad. Odysseus then spends another decade wandering home to Ithaca, which is the subject of the Odyssey. The story of AI also begins in war. Electronic computers emerged in the codebreaking agencies of World War II and the command centers of the Cold War. The journey of digital automation from the military battlefield to the civilian economy has likewise followed a winding path.

The Odyssey itself begins in the middle. Odysseus has been stranded for seven years on the island of Calypso, a beautiful goddess who gives him every Earthly pleasure and promises him immortality. Yet the castaway spends every day crying for his lost shipmates and family far away. Calypso’s island of Ogygia is not a bad metaphor for the situation in which we find ourselves today. We have all the world’s information at our fingertips, and gourmet food is just an app away. The digital world is a paradise of plenty, and some futurists promise that our brain scans will live forever. Yet many people lost in cyberspace feel lonely, hollow, and alienated, pining for a lost sense of community.

Meanwhile back in Ithaca, King Odysseus is lost and presumed dead. A hundred and eight arrogant suitors devour his resources and press Queen Penelope to remarry. Back on planet Earth, likewise, an arrogant horde of technologists encourages us to marry our lives to AI. Intoxicated with visions of technological revolution, the tech bros devour our attention, money, and natural resources. The narrative is always the same: technological revolution is nigh, for better or worse, and we have no choice but to accept it.

Yet the gods finally decide that enough is enough. It is time for Odysseus to return home. Hermes carries their judgment to Calypso, while Athena flies to Ithaca to mentor Telemachus, the boy who has never known his father. The goddess of craft and creativity inspires the young prince to search for news of Odysseus. Telemachus then goes to Sparta to learn from some old warriors who have returned from Troy. If we want to learn the truth behind the hype about AI today, where should we go for mentorship and inspiration? Where should we seek knowledge about how technology really works in the world? How can the veterans of history encourage our collective coming of age?

The Odyssey runs like clockwork. As the prince leaves home, the king returns. Odysseus builds a raft to leave Calypso, and Poseidon wrecks his craft again. Better technological art is not enough to escape the curse of automation. But this time, Princess Nausicaa of the seafaring Phaeacians discovers the lone castaway and invites him to her spectacular palace. Her father, the Phaeacian king, holds a feast in his honor, and Odysseus finally tells his story. It is time to tell the story of technology as well, assuming that we can find any welcoming communities that are willing to hear it. We need to explain how we ended up in this desperate situation in the first place.

Chapter 2. Out of Control

The apologia of Odysseus begins with a story of economic extraction gone wrong. The ruse of the wooden horse enabled the Greeks to sack Troy, which meant killing all the men, enslaving the women, and looting the town. When Odysseus and his twelve ships are blown off course, they arrive at the land of the Cicones and subject them to the same treatment. But when locals come seeking revenge, six men on each ship are killed in the battle. While digital technology emerged to help win a war, it returned to society to make a fortune for entrepreneurs. Yet the commercial triumph of cyberspace also meant a nightmare of extraction. If we look behind the marvels of AI, we find harsh working conditions in warehouses and factories around the world, growing economic inequality and lopsided concentrations of wealth, relentless consumption of energy and environmental resources, epidemics of cybercrime and disinformation, and a generational mental health crisis. Even technologists themselves begin to suffer.

Another storm blows the decimated squadron—each crew of fifty lost six men—to the land of the lotus eaters. The addictive plants that grow on this island cause some of the crew to lose all interest in going home. The storm of extractive capitalism likewise has discovered the allures of digital addiction. Fascinated with our phones and forever seeking the next hit of dopamine, we lose interest in connecting with actual human beings. Encouraging perpetual engagement becomes corporate policy, and AI makes attention harvesting more lucrative than ever. Large language models turn us into lazy lotus munchers.

The ever-entrepreneurial Odysseus next tries to pillage the pastoral island of the Cyclopes. But things continue to go wrong as an uncivilized savage eats six more of his men instead. The rest of the raiding party escapes Polyphemus the cyclops by a clever stratagem, but then Odysseus boasts of his exploit on the way out an incurs the wrath of Poseidon, the father of Polyphemus. If Polyphemus makes much noise, the giant internet is full of words as well. Pirates pillage digital networks, and clever hackers steal valuable goods from secure facilities. The age of automation becomes the age of deception. Bluster about secret exploits still invites disaster, moreover, as the divine powers of cybersecurity make life harder for hackers.

Odysseus finally catches a break when the god Aeolus gives him the gift of winds. Gentle Zephyr blows them to within sight of Ithaca, and Odysseus collapses with exhaustion. With the steersmen asleep at the tiller, the deck bros open the bag of winds and the squadron is blown away. AI emerged in the age of cybernetics, a word derived from the Greek for steersman. Technological progress in semiconductors and datacenters finally gave the gift of winds to cybernetics. Deep learning and transformer models have brought AI in sight of the great dream of thinking machines. Yet as accelerationists let loose the recursive self-improvement of AI models, the result might be an intelligence explosion that blows us all away. Or perhaps the AI-olean hype is just a blast of hot air.

The Ithacan squadron encounters its single greatest disaster in the harbor of the Lastrygonians, where giant cannibals murder eleven shiploads of sailors. Only the flagship of Odysseus, anchored outside the deadly fjord, manages to escape the massacre. The theme once again is giants out of control, but it is perhaps less the exponential growth of AI we should worry about and more the unimpeded growth of AI firms. If software is eating the world, then venture cannibals are bringing the appetite. Only the CEOs on this voyage of technology, with their doomsday bunkers and offshore crypto accounts, will be able to escape with their fabulous wealth.

The lone ship next puts in at Aeaea (or AI-ea), the home of a divine enchantress named Circe. She turns half the men into pigs, and Odysseus only escapes the same fate thanks to the intervention of Hermes – the god of communication, merchants, and thieves, and thus the perfect god for cyberspace. Odysseus becomes Circe’s lover instead, even as Penelope remains alone in her virtue. It is impossible to ignore the unfolding crisis of masculinity in the tech world of AI-ea. Boys have easy access to endless pornography, while girls vamp for their phones and become victims of sextortion. The manosphere preens and pouts on podcasts, and incels rail and rant online, while tech execs cast about for their next trophy wife.

Chapter 3. Lessons Unlearned

Up until now Odysseus has been losing control, blown from one misadventure to the next. The habits that brought such success in Troy no longer seem to be working. The encounter with Circe at first seems to be more of the same, but things are not what they seem in AI-ea. Meeting Circe is also the beginning of a transformation in the narrative of Odysseus. Circe turns the pigs back into better men, and she gives Odysseus advice about how to get home. Our relationship with technology could use some transformation as well. The pursuit of power and profit has automated the world, but that world increasingly feels out of control. What enchantment might relieve the suffering of technology? And what advice might point the way home toward our humanity?

Circe advises Odysseus to seek the ghost of Tiresias, the great prophet of Thebes, and she instructs him how to commune with the dead. At the gates of Hades, Odysseus meets the transgender prophet and many more shades. He learns that his mother has died of all the sorrows in Ithaca. He hears the stories of many women and of his fallen comrades from Troy, their shades consumed with resentment and regret, and he learns the fates of many mythic heroes and villains. AI algorithms now train on dead data that are mere shades of human experience. At the end of the knowable world, perhaps we can learn to hold space for their shades to speak. Perhaps we can learn to listen to the stories of those who have suffered before us, who have created this technological world. Perhaps we can learn to perceive the mythologies that give meaning to our mortal lives.

A somber Odysseus now returns to Circe. Like a good therapist who has encouraged a man to meet his own ghosts, she now tells Odysseus about the temptations that he still must face. She warns him of the seductive Sirens, Scylla the barking beast, Charybdis the sucking whirlpool, and the sacred oxen tended by the daughters of Helios. (The second half of the voyage of Odysseus is decidedly more feminine than the first, as if he must atone for his masculine excess in war before going home; perhaps tech also must shift from the masculine polis of conflict to the feminine oikos of care.) Yet Odysseus will selectively follow or disastrously ignore Circe’s advice. While the therapist offers helpful knowledge, the patient is unable or unwilling to absorb it. We can be made aware of the pathos of technology, but awareness is only the beginning of grief. Sometimes the pain must be lived to be understood.

The Sirens provide an obvious metaphor for the flashing rectangles that call out for our attention. We follow our friends on our phones onto the shoals of psychosis, amusing ourselves to death. But the song the Sirens sing is perhaps even more relevant. Their sweet voices promise to tell Odysseus the whole truth about what happened in Troy. They promise to satisfy our curiosity about the whole wide world with just one more swipe of the infinite scroll, revealing the truth of the conspiracies that surround us. Odysseus famously escapes by tying himself to the mast and stuffing his sailors’ ears with wax, a technological fix to enable indulgence without consequence. He puts a lock on his screen time, but he doesn’t give up his phone.

Odysseus next must make painful tradeoffs. To avoid the ship-smashing Symplegades, which only the heroic Argonauts have survived, he must shoot the strait of Scylla and Charybdis. To avoid the wicked rocks of political disaster, either the howling heads high up on the cliffs must gorge on the sailors of society, or the gulping vortex of passion must devour the ship of state. For all the varied dreams of digital democracy, technology cannot escape the painful tradeoffs between monstrous bureaucracy and sucking markets, or tyrannical government and destructive anarchy. It is telling that Odysseus hews closer to Scylla, while staring in horror at Charybdis. Technologists often steer closer to control to avoid the chaos of competition. Against the advice of Circe, Odysseus dons his armor in a vain attempt to tame the bureaucracy, but Scylla gobbles up six more men. This time will be different, the tech lords promise, yet the tyrant shows no mercy.

The surviving remnants of the Ithacan expedition arrive at the pastures of Helios. Odysseus shares the warnings of Tiresias and Circe not to harm the oxen, but then he leaves his men alone. As before with AI-olus, the deck bros give in to their greed and slaughter the sacred cows. Technologists have a way of crossing ethical lines and flouting moral conventions. If the only rule is to move fast and break things, then lesser norms can be ignored. Silicon Valley knows that there is value in the performance of righteousness, of course, and tech firms may even hire ethicists in hopes of aligning AI with human values. But the driving alignment is always the profit and power of firms themselves. With regulatory oversight absent or asleep, the tech bros feast while the markets lay waste.

Sure enough, Helios is enraged at their transgression and implores Zeus to send a tempest. All of the sailors are drowned, except for Odysseus who clings to the wreckage of his ship. Whereas before he sailed too close to the monstrous tyranny of Scylla, now he veers too close to the anarchic whirlpool of Charybdis. He survives by leaping for a homely fig tree growing out of the cliff. The relentless pursuit of control, in the face of all warnings of things slipping out of control, culminates in desperation. The technologist grasps at life as all hope seems lost. Odysseus then drifts away to Ogygia, where he will spend seven years with Calypso, wallowing in grief.

Chapter 4. Giving a Damn

So ends the apology of Odysseus, a wandering tale of how we arrived at our pathetic technological moment. As we have seen, the confessions of our mythical hero suggest a cosmology for our AI experience. The fact that this allegory works at all suggests that the problems of technology are really nothing technological at all. AI is not so much giving us a brave new world as giving our old habits right back to us. Improvements in AI are hastening the eternal return of classic archetypes. This means that advanced problems have ancient roots, and, conversely, that ancient solutions have modern relevance. Perhaps the same mythic patterns that enabled archaic Achaeans to make sense of an agrarian world can also help our ostensibly advanced civilization make sense of its increasingly automated planet.

At this point we are only halfway through the Odyssey. The next twelve books of Homer’s epic return with Odysseus to Ithaca. The dramatic narrative is slower, more detailed, and more intimate, steadily building up tension for the return of the king. Except for the occasional intervention of Athena, the magical grandiosity of the first twelve books is gone. Mythology gives way to realism. The text itself makes this transition crystal clear as the fabulous Phaeacians deliver Odysseus home in a self-piloting ship that can predict human thoughts; Poseidon then turns the Hellenic Waymo into stone on the return voyage, sealing off the technological paradise of Phaeacia forever. From now on, the family and political drama of Ithaca will be a deeply human drama.

To truly understand AI, therefore, we may need to set AI aside. Technological progress in semiconductors, algorithms, data infrastructure, sensors and robotics is enabling fantastic AI applications that can steer themselves and predict human thoughts. But prediction machines still rely on human judgment. Judgment determines what matters and why, who or what to trust and how much, which rewards to pursue and which risks to avoid, who to love or fear or envy, and generally how to live this one mortal life. Human judgment is the source of guidance, interpretation, and maintenance the AI systems require to do anything useful. As statistical prediction becomes more efficient, complementary judgment becomes more important, not less. This means that the political and economic drama of AI is a deeply human drama.

Fifty years ago, the philosopher John Haugeland wrote that “the trouble with AI is that computers don’t give a damn,” and I never tire of quoting him. It is still true that the latest frontier models do not give a damn, even though they can fake it with beguiling words (or Siren songs?). AI generates data without understanding what’s at stake for human beings or even themselves.

When Odysseus returns to Ithaca, by contrast, everything is at stake. He is a middle-aged man facing a hundred and eight strapping young men and a family who may resent his absence after twenty years. His life, his marriage, his kingdom, and his mythic legacy are all on the line.

Odysseus first encounters Athena, the personification of craft and creativity, who is disguised as a shepherd boy. She reveals herself, and the two tricksters get to work strategizing his return. Because of the great danger, getting the strategy right is damn important. They decide on a long con to methodically prepare a trap for the suitors. Odysseus will disguise himself as a beggar, not even revealing himself to his long-suffering wife or servants. He will tell many different stories about himself, mixing lies with elements of the truth. Odysseus will manipulate the relationship between representation and reality, which is precisely the relationship that eludes machines that don’t give a damn.

Chapter 5. Caring about Stories

Telling stories is the way a beggar infiltrates a wealthy palace, as surely as the Trojan Horse penetrates the holy citadel of Troy. Retelling Homeric myth, moreover, imparts heroic purpose to its audiences, even thousands of years later. There is a narrative about narrative these days that stories are just stories: Hard power and economic incentives are what determine behavior in the real world, and narratives are just window dressing. Homer has no truck with this view. He shows us instead that telling stories are a way of wielding power. Narratives can mobilize political support or undermine legitimacy, and narratives can coordinate military conquest or unravel an army. Power and incentives matter, of course, but only because narratives tell us what is worth fighting for and what counts as a cost or benefit. Narratives shape human judgment, and judgement shapes AI.

Some of the most important acts of judgment that shape AI are the least appreciated. Functional automation relies on endless everyday work such as building and cleaning datasets, manually debugging agential workflows and codebases, fiddling with configuration settings and security credentials, and arguing with colleagues and customers about AI outputs, to say nothing of managing AI supply chains and resource flows, rearranging factories and shops to be robot friendly, and contesting the intrusive surveillance of corporations and governments.

The work of keeping Ithaca running, likewise, falls on the loyal slave Eumaeus who tends the flocks, the nursemaid Euryclea who raises the royal children and washes the feet of random guests, Queen Penelope who weaves a burial shroud for demented old Laertes and unweaves it every night to keep the suitors at bay, and all the other subjects of Ithaca – including the suitors themselves – who have not had a proper council meeting since Odysseus decamped for Troy. The only reason that Odysseus has a home to come back to is all the hidden work that keeps it going. Wherever the leadership of technologists is missing, countless others perform the hidden work of maintenance to keep AI going. This hidden work both smooths and creates friction, and friction both impedes action and provides traction for it. Odysseus must negotiate the considerable social friction of Ithaca to gain traction for his strategy.

One might question the relevance of the Odyssey to AI since AI is an information technology that uses advanced statistical techniques like deep learning. But Homer’s Odyssey is deep meditation on the very nature of information itself. The poem unfolds as shifting web of narratives that leave us wondering what to believe. What should Telemachus make of the reminiscing of old veterans? Did any of Odysseus’s fantastic adventures really happen? Did Penelope recognize her husband long before the climactic battle? The characters and audiences of the Odyssey must make their own judgments on these questions.

Yet Homer goes one step further by showing us the grounds of good judgment. The Odyssey gives us a series of recognition scenes that join the elusive illusion of narrative to the consequential concreteness of reality. Odysseus reveals himself to Telemachus to bring him into his secret scheme of revenge. The old dog Argus perceives his master through the beggarly rags. The nurse Euryclea discovers the childhood scar of Odysseus while washing the beggar’s feet. Odysseus springs his trap on the suitors by winning the trial of the bow, the only man able to string it and shoot an arrow through twelve axes. After slaughtering the suitors, with the help of Telemachus and some servants, the victorious hero finally presents his bona fides to Penelope, revealing the secret construction of their marital bed. These sorts of costly signals provide people with a reliable grip on their world.

Information does not float freely in the Odyssey. It is grounded in a web of social relationships and material conditions. Odysseus can detach signifiers from signifieds to confuse and mislead others. Meaning can be reattached through accidental collisions or the hard work of maintenance. Homer shows us that the dynamics of information are thoroughly grounded in social practice. Homer thus shows us that AI is a thoroughly social technology as well. Thinking machines matter only insofar as the encounter of narrative and friction makes them meaningful for human beings.

Chapter 6. Eternal Tragedy

One might read the Odyssey as a fairy tale in which the hero goes on a journey and learns many lessons along the way. Self-discovery enables him to beat the bad guys and get the girl, and they all live happily ever after. But Homer offers a more tragic view of human culture, one with important implications for AI. In some ways, Odysseus does not change at all. He remains a heroic warrior who brings the slaughter of Troy back home to Ithaca.

One might even question the heroism of Odysseus. He inducts his son into a conspiracy to ambush and murder the best sons of Ithaca. They string up a dozen slave girls for sleeping with the suitors. He even tells his wife during their conjugal reunion that he will have to leave her again to go on another quest. He then torments his aged father and escalates a civil war with the suitors’ families.

The Odyssey ends with the deus ex machina of Athena making the warring parties swear an oath of peace, as if to underscore that there can be no lasting peace between mortals themselves. Homer’s pattern is clear: the same things keep happening again and again. Troy recurs with the Cicones and in Ithaca, the slaughter in the palace recurs outside in the kingdom, Odysseus returns just to leave again, Telemachus will become a king who will have another son, and bards will keep singing their adventures forever. There is no final end to political narrative, not even or especially in the age of AI. There is only the eternal return of rivalry and reunion. Some of the rivals will eventually invent thinking machines and quantum computers, and some of the reunions will be coordinated online, but still the cycles will repeat.

Put simply, the Odyssey predicts the Odyssey. This is precisely why Homer remains relevant in the age of AI (and why even problematic interpretations like Nolan’s The Odyssey remain interesting). The archetypal patterns of automation are ultimately not rooted in computational technology at all but in the pragmatic narratives of indefinite politics.

Yet the Odyssey also contains an alternate ending. The shade of Tiresias at the gates of hell foretells that Odysseus will walk away from it all. He will carry his oar inland until he meets people who do not know the sea. They will take his oar for a winnowing fan, a tool for sorting wheat from chaff in the wind rather than a means of nautical control. Where cybernetic instruments are unknown, in other words, Odysseus will finally make peace with Poseidon, the god of shaking earth, pounding seas, and thundering horses. Then he will find a gentle death from the sea, surrounded by loving family in old age.

The prophesy of Tiresias looks beyond the striving ambition of the tech world toward a more stoic acceptance of nature, in all its incomprehensible mystery. If the Odyssey always repeats, this repetition always offers the chance to make different choices. The eternal tragedy of human politics animates the world of AI, even as we ponder our political fate. So we stand with AI, oar in hand.

Chapter 7. The Cusp of Orality

Another reason to read Homer in the age of AI is to survey the rise and fall of human writing. “Homer” is less of an historical personage than an evolutionary process. The transformation of oral performances into canonical text happened sometime around the seventh century BCE. The new technology of the Greek alphabet, with its consonants and vowels, was able to signify the sounds of speech. Archaic Greeks simply wrote down what they were signing.

Yet in the process, the transcription of speech to writing transformed human consciousness as well. Rather than singing passages of poetry in a group, an individual reader could ponder the entire epic alone. Hearing sounds in time gave way to seeing symbols in space. New symbolic systems created new opportunities to think abstractly, conceptually, mathematically, and cartographically. Writing enabled the flourishing of philosophy, history, and science. It also enabled the growth of bureaucracy and empire, killing and extracting at unprecedented scale.

The advent of writing did not eliminate oral culture, however. Stories and songs remained. Narrative even performed a lot of hidden work to keep the sciences going. Social life was as messy and meaningful as ever, and orality traversed the gap between written representation and lived reality. Homeric epic also persisted as the written world’s memory of what oral culture needs to flourish. To read Homer today, therefore, is to find a connection to the parts of society that still speak to us. Through the ages countless readers, scholars, and artists have thus rediscovered the poetry of the past.

Walter Ong describes the emergence of a new oral culture thanks to radio and television. Secondary orality is further encouraged by streaming video, social media, and podcast culture. Once again people are communicating by voice and in time, not simply by sight and in space, but the scale and reproducibility of orality are unprecedented. Now we even have AI agents that seem to speak for themselves. AI compresses human writing to generate something else, something that may mean the undoing of writing altogether if we aren’t careful.

As AI mixes and remixes the global codex of human data, literacy is undergoing a transformation as significant as its genesis. The generative output of AI can now beat human creativity in blind comparisons. The Turing Test is passed. But AI’s victory may be as hollow as the wooden horse. While large language models provide a virtual connection to vastly more information than any one individual could possibly understand, in the process they sever the connection to pragmatic life that makes information meaningful in the first place. The triumph of simulation misses something essential—the dignity of human connection.

AI can generate text, but meaning depends on context. As the written poem recalls an oral song, epic singing enacts an ancient community. Now, alienated on the connected island of Calypso, the crisis of automation has us yearning for community again. The recollections of AI are more fragmented and harder to reconstruct. Yet behind every AI system is a world of tall tales, bloody fights, extractive wealth, male anxiety, endless grief, human judgment, hidden work, political scheming, and the eternal return of choice. Computers don’t give a damn, but the people who build and use them do, or at least they (we) should.

Homer sheds light on who we were long before AI. And I suspect that Homeric epic also sheds light on who we will be long after the current hype about AI goes the way of the information superhighway. Human beings are not merely mental software running on the wetware of the mind, but embodied social agents with meaningful commitments to the world. If you want to understand what is most important about the world of AI, then go and read the Odyssey. Go and watch The Odyssey, too, preferably in IMAX 70mm if you can. And then reread Homer again, or better yet, sing it out loud and listen. You might just find that Homer has what AI is missing.

Magnifica Humanitas, Pope Leo’s encyclical letter “on safeguarding the human person in the time of artificial intelligence,” is probably the most insightful discussion of the ethics of AI that I have read anywhere. I recommend it to everyone regardless of religion, creed, or lack thereof.

The text has an undeniably universal appeal. You don’t have to be Catholic to see the wisdom in it. Indeed, Pope Leo deliberately addresses his “heartfelt appeal to all the Catholic faithful, to all Christians and to all men and women of goodwill” (16). He offers a reflection on general truths (or reliable norms if you prefer) that may be appreciated through either a Catholic lens or other philosophical perspectives. I suspect that its aspiration to universality helps to account for the wildly positive reception of the encyclical to date among many readers of varied backgrounds. In this time of technological upheaval, there is a palpable hunger for a human-centric understanding of AI, and Magnifica Humanitas breaks bread with us all.

To be sure, there is still a lot of theology in it. And it would be shocking if there was not. A papal encyclical, after all, offers authoritative teaching to Catholic leaders on matters of ecclesial importance. The text is solidly grounded in Christian principles and uses evangelical rhetoric to make its argument. For many readers, this will be a source of authority. For others, it will be a deterrent to understanding. For the former, there may be little need for additional commentary. For the latter, a secular mapping of ideas may be helpful.

In that spirit I will hazard a liberal interpretation of Magnifica Humanitas that connects some of its insights to themes in modern social science. I realize that this is a bit like brewing non-alcoholic beer, which is either a travesty or an improvement depending on your attitude toward alcohol. But here I will bracket, in the phenomenological sense,[1] questions of metaphysis or theology. My goal is simply to appreciate themes of the text as they appear to me. That is, you may read this effort as translation, distillation, dilution, interpretation, revelation or evangelism, as you prefer.

Magnifica Humanitas presents a closely reasoned argument over the course of an introduction, five chapters, and a conclusion. Recast in secular terms:

• The introduction presents the overarching theme of valuing the flourishing of human community over the expansion of technocratic control.

• The first chapter explains why this is a universal concern.

• The second explains the principles of political economy that enable human flourishing.

• The third explains why AI cannot substitute for human nature.

• The fourth discusses the social challenges of the AI revolution.

• The fifth turns to the dangers of militarized AI in the global system. (Note that these last three chapters map roughly to the levels of analysis familiar to scholars of international relations, introduced by Kenneth Waltz in Man, the State, and War, although in contrast to Waltz the last offers a sharp repudiation of realism.)

• The conclusion calls for cultivating human relationships and social justice.

I will take each in turn. Or at least that was the plan. There is so much to chew on that I am already running way longer than expected on just the first two chapters. After all, the 42,300-word encyclical is the length of a short book and presents an entire cosmology.

So, in this post, I will just focus on the convergence between Catholic social doctrine and liberal political economy. A later post (hopefully) will discuss the three levels of analysis—human nature, society, and international relations. As always, the posts in this blog are unedited musings to figure out what I think, not finished scholarship. Comments are welcomed.

Introduction: Control and Community

Magnifica Humanitas begins with a contrast between the Bible stories of the Tower of Babel and the rebuilding of Jerusalem. The first describes a technocratic project to impose a single language and goal on the world to ensure stability and control, “a uniformity that eliminated diversity and that chose homogenization over communion” (7). This hubristic venture fails miserably, and the people are scattered and confused. In the age of AI, the parable warns of “the idolatry of profit that sacrifices the weak, a uniformity that neutralizes differences, and the pretense that a single language — even a digital one — can translate everything, including the mystery of the person, into data and performance” (10).

The second describes a democratic project “which rebuilds relationships before rebuilding with stones,” building a safe and sustainable community “through the shared responsibility of all: men, women, priests, artisans, heads of households and young people all play a part” (8). In this parable AI is not a substitute for human beings but a complementary tool that people can choose to use to enhance the common good. Humanity thus faces a choice “between a power that claims to dominate the heavens and a people who work together…to rebuild the walls of fraternal coexistence” (9).

This contrast recalls Paul Edwards’ comparison of “the closed world” of cybernetics (the science of human and machine control that emerged from World War II and culminates in modern AI) with more open “green world” conceptions of technology. The idea that tech always develops within a broader system of human choices also recalls Langdon Winner’s classic discussion of autonomous technology (which fifty years later is still my favorite book about the social essence of AI).

The AI Tower of Babel is not just a material technology but a cultural “technics” that fetishizes artificiality, rationality, ubiquity, specialization, connectivity, and order in pursuit of power and efficiency at scale. This technocratic system runs on an endless treadmill that tries to achieve total control without ever getting there, while at the same time ratcheting up the potential for large-scale catastrophe. AI catastrophes top of mind include autonomous vehicle accidents, employment collapse, ecological shock, hacker exploitation, and military escalation, to say nothing of superintelligence explosion. (Robots of unusual size? I don’t think they exist). A more human-centered approach might help to mitigate them.

The Pope’s conception of technology is thus right at home in the interdisciplinary field of Science and Technology Studies (STS). Namely, “technology is never neutral, because it takes on the characteristics of those who devise, finance, regulate and use it” (9). The supposed neutrality of technology, widely assumed by technologists in public policy debates, is a lazy trope that obscures accountability for countless value-laden decisions that shape and reshape the human-built world. Technology is not external to society but an expression of it. So, what kind of society do we want to build?

Chapter I: Universal Concern

The first chapter reviews encyclical teaching since the landmark Rerum Novarum, or “On Capital and Labor,” by Pope Leo XIII in 1891. Reflecting on the economic tumult of the industrial revolution, Leo XIII laid the foundations for what has come to be known as Catholic social doctrine. Pope Leo XIV explicitly honors this legacy by updating social doctrine for the automation revolution.

Chapter one thus connects Leo XIII’s early concern for labor rights to an evolving tradition of Catholic concern for social justice, military disarmament, and ecological sustainability. Leo XIV’s distinctive contribution to this tradition is to emphasize the sanctity of human relationships, the dignity of individual people, and the integrity of human community, which are all threatened by automation. Automation here is understood not as simple technological determinism but as a system of ideas, institutions, and infrastructures for automating data processing (or “intelligence,” which appears in quotation marks in the encyclical when referring to whatever it is that AI does in order to distinguish it from human intelligence).

This is accordingly the most doctrine-heavy chapter in the encyclical, but the upshot is demonstrating the Church’s abiding interest in worldly problems of universal concern, not just the eternal preoccupation with saving souls. The path from Rerum Novarum to Magnifica Humanitas is tale of progressively broadening engagement with the real political, economic, social, and environmental conditions under which people actually live, work, suffer, and die. These conditions are truly planetary in that they affect people of all backgrounds, nationalities, and faiths.

In other words, Magnifica Humanitas is meant to be not just Catholic but also small-c catholic. So, feel free to give it a liberal reading.

(I mean small-l liberal here, i.e., the ideas shared by both American Liberal and Conservative factions. Conservatives venerate the founding fathers, who were steeped in liberal thought, and have been known to champion free markets and democracy. Liberals are suspicious of concentrations of power and advocate for social justice. In the Jacobin formula of “Liberté, Égalité, Fraternité,” Conservatives tends to endorse liberty and fraternity, often exaggerated as nationalism, while Liberals overfit on equality and communists dispense with liberty altogether. The fact that they tend to split on their attitude toward Christianity, moreover, is at the very least suggestive of a historical link between liberal and Catholic values. That also means that Magnifica Humanitas can be read as a small-c catholic reflection on small-l liberal ideas.)

In the spirit of liberal universality, Pope Leo calls for interfaith and interdisciplinary dialogue: “The Church regards all who sincerely seek ‘truth, goodness and beauty’ as companions on the journey, and considers them as ‘precious allies’ in defending the dignity of every person and in caring for creation” (23). Rather than imposing a dogmatic interpretation, “When it comes to applying these standards to the complex situations of our time, the contributions of philosophy and of the human and social sciences is essential. These disciplines help us understand and analyze cultural, economic and political dynamics more deeply” (23). We’ve come a long way since the inquisition of Galileo.

If “the Church welcomes the contributions of the social sciences” (23), then perhaps the social sciences also have something to learn from the Church. There is interesting convergence between Catholic social doctrine applied to AI in Magnifica Humanitas and idea in the liberal tradition of political economy. I will not take a stand here on which perspective ultimately carries more authority or how each influenced the other historically. Suffice it to say that modern secular humanism and the Christian tradition share many fundamental values, such as speaking truth to power, giving voice to victims, and striving for social justice. The historical rupture between the liberal Enlightenment and the Catholic Church obscures their common anthropological origins and, perhaps, their common humanist destiny.

Chapter II: Political Economy

If chapter one is the literature review of the thesis, chapter two is its theory section. Here the encyclical lays out core principles of Catholic social doctrine: “the dignity of the person, the value of work, the universal destination of goods, solidarity and subsidiarity, care for creation and the centrality of peace and fraternity” (45). In reading this chapter, I was surprised (or rather impressed) to discover something like the essential norms of liberal political economy. The terminology is different and far more prescriptive, and the interconnections across concepts are not immediately obvious, but the familial resemblance is undeniable.

By liberal political economy I mean the tradition of thinking about the relationship between states and societies, or governments and markets, or institutions and individuals, or structure and agency, etc. This is a tradition that spans Adam Smith and John Locke (on the virtues of liberty), Karl Marx and John Keynes (on the dangers of unfettered capitalism), Douglass North and Elinor Ostrom (on the benefits of collective institutions), and many others.

There are many complicated debates and nuances in liberalism, but at a first order of approximation, an overarching theme in political economy is that human societies must make, and keep on making, tradeoffs across two different values or ends (the good of society and the good of individuals) and two very different types of institutions or means (governments and markets). Because I am a simple political scientist and a 2x2 figure is the simplest way to say that the world is complex, the tradeoff space can be summarized thus:

Considering the eight concepts above, there are six things that are generally nice to have (individual freedom, public goods, common infrastructure, healthy communities, and governments and markets that work for everybody), but there are two failure modes that we really want to avoid (ruthless domination and capitalist exploitation). Now, Pope Leo’s encyclical lists five principles of social doctrine, but it prefaces them with an overarching emphasis on “the equal dignity of all human beings” (51) and “the supreme value of human rights” (54), which I will take as a sixth principle or metaprinciple. It also sustains a critique of domination and exploitation throughout the text, so I’ll treat those as two important antiprinciples. Let’s see if we can map these two sets of six desiderata together.

Here is the same 2x2 as a Cartesian graph. If it looks like a cross, that is purely a coincidence. Reason and religion come together, so to speak. Agential ends are on the x-axis (working together or doing your own thing). Structural means on the y-axis (having shared systems or using your own stuff). Concepts from Catholic social doctrine are in bold (including the five formal principles and the metaprinciple and antiprinciples). Concepts from liberal political economy are in italics. Good outcomes are in green, and bad outcomes are in red. Thus, the noble desiderata of social doctrine are in bold green while the diabolical outcomes are in bold red (see what I did there?). The political economy labels in black are meant to suggest, if not neutrality, then ambiguity: they become good or bad based on human choices in historical circumstances (i.e., combinations of ends and means).

I can hear my hair-splitting critics howling that I am conflating very different things. But hear me out! If you squint, I think you might see a familial resemblance, at least at the first order of approximation. Yes, nuance is missing. Also, I am an unreformed lumper, not a splitter. Let’s discuss!

Human Dignity refers to the inherent value of the human person and plays the same fundamental role in Catholic social doctrine as liberty plays in liberalism (as in “We hold these truths to be self-evident, that all men are created equal, that they are endowed by their Creator with certain unalienable Rights, that among these are Life, Liberty and the pursuit of Happiness”). Dignity has many dimensions: moral (wise choices), social (material conditions), existential (self-esteem), and ontological (intrinsic worth). Pope Leo contrasts it with the “insidious” ideology “that suggests that every person must earn or justify his or her own worth, to the point of attributing greater value to those who are more efficient or effective” (51). I read this contrast as the two faces of freedom. Liberty is the positive face of freedom, the flourishing of every individual person. The negative face of freedom is selfishness, treating people as instrumental to hedonic pursuits. The great thing about freedom is that you can express your authentic nature, but the bad thing is that other people are free to harm you. This ambiguity manifests in the “insidious” reduction of human value to efficiency and effectiveness and the virtues of “effective government” and “efficient markets.”

Common Good, “the social expression of the dignity recognized in every person” (59), maps easily onto collective welfare. Political economy recognizes different types of collective goods, e.g., public goods (national defense), club goods (subscription services), and common pool resources (fisheries). The common denominator, as the name implies, is that collective goods benefit everyone in the collectivity, however defined. Ensuring collective welfare likewise requires collective contributions “in which people learn to recognize that they themselves are interconnected and jointly responsible for the res publica” (62).

Solidarity describes the bonds of fraternity that unite a (or the) human community. While the word typically connotes a normative commitment to working together or a shared identity, Pope Leo uses it in the more full-blooded sense of “a social and political reality to be embodied in communal choices and endeavors” (73). The embodiment of ideas moves toward the substantiality of institutions and infrastructures. Indeed, “we are already experiencing a kind of ‘de facto solidarity,’ for our lives are intertwined; digital networks connect people and communities across the world in real time, and global economies and communications mean that events in one place have a far-reaching impact” (74). Institutions in political economy are “the rules of the game” or the human-built constraints on human interaction. Elsewhere I argue that sociotechnical institutions include “the tools of the game,” so the internet, for instance, is an institution. Institutions enable solidarity by lowering the costs of collective action, improving information, and socializing shared values.

Subsidiarity means that “the role of individuals, families, local communities and intermediary organizations should not be supplanted by higher-level authorities” (68). I take this to mean something like civil society. (I must admit this was a new word for me outside the context of corporate ownership.) Pope Leo states that “subsidiarity applies especially in the context of the digital revolution. Here, the highest level is not the State, but rather major economic and technological actors that exercise de facto power over the conditions of everyday life” (71). Market actors thus exemplify subsidiarity. And just as markets are complementary to governments, subsidiarity is the complement of solidarity: “When subsidiarity is not linked to solidarity, it ends up becoming merely the protection of particular interests; when solidarity is not supported by subsidiarity, it degenerates into a form of welfare that does not foster responsibility” (73). The complementarity of solidarity and subsidiarity maps well onto the liberal distinction between governing institutions and decentralized markets. Or in international relations, between institutions and anarchy. Atomized sovereignty is a boon for those who have it, until it isn’t.

Universal Destination refers to things that belong to all, most notably the natural environment: “every person has an inherent right to the use of such goods, both now and in the future” (65). Pope Leo expands the aperture to encompass not only material resources but also “new forms of property, such as patents, algorithms, digital platforms, technological infrastructure and data. In a context where the wealth of nations depends increasingly on knowledge and technology, when these goods remain concentrated in the hands of a few, without adequate forms of sharing and access, a new imbalance is created that contradicts the universal destination of goods” (67). Whereas the common good refers to the ends of social action, universal destination refers to public ownership, management, or husbanding of the means. The contrast here is with “private property, which has its own specific meaning and purpose, yet it is always subordinate to the universal destination of goods” (66). In the liberal mapping, I see this tracking with good governance, or institutional management to improve public welfare. Governments are important for setting standards, establishing and enforcing property writes, redistributing resources for public works, and providing for the common defense. Put differently, universal destination is solidarity for the common good. As we shall see, solidarity can also be selfishly subverted: governments can fail when they divert public resources to non-universal destinations.

Social Justice is “characterized by the capacity of a social, economic and political order to allow everyone — particularly the weakest — to live a truly dignified life, without leaving anyone behind” (77). Just as universal destination is the institutional enabling of the common good, social justice is the embodiment of human dignity. In the mapping here this corresponds to the virtues of the market, which might seem terribly strange since social justice movements often work to counter the injustices of capitalism. Fair enough, but that is only because markets fail in some terrible ways. In the liberal tradition, markets are the most efficient way of coordinating social interactions and honoring the principle of liberty. Consider that “injustices do not arise solely from the wrong choices of individuals, but also from structures, mechanisms and economic and cultural systems that produce inequality almost automatically” (79). Injustice is a structural product of institutions, so leaving them or reforming them is essential for social justice, which might be considered as deregulating domination. Social justice enables the flourishing of subsidiarity, or a diversity of voices, lifeways, and communities.

Exploitation, however, is subsidiarity run amok. This is my term for an antiprinciple, or something that social doctrine should avoid, mitigate, discourage, or heal. When selfish actors are free to harm and pillage as they please without regard for the common good, the result is the common bad. This is the basis for balancing freedom with governance: “States and transnational institutions are called to ensure fair rules and effective safeguards, so that local communities, intermediary organizations, schools, universities, religious institutions and associations have a voice and can contribute to the discernment of choices that affect people’s daily lives, such as employment, access to services, data management and digital environments” (72). This maps well onto the liberal idea of market failure. Markets may fail to provide collective goods under conditions of information asymmetry, negative externality or pollution, ill-defined property rights, or unfair competition. Capitalism run amok leads to the exploitation of people and the planet, as discussed extensively in later chapters.

Domination (my other inferred antiprinciple) is solidarity subverted. This is the capture of public institutions for private gain. Tyranny, conspiracy, corruption, regulatory capture, biased justice, fraud and waste, and all the varieties of government failure are forms of domination in this sense. Later chapters, especially on international relations, have a lot to say about the evils of domination.

Integral Human Development, finally, is the ongoing process of balancing all these desirable principles (and mitigating the antiprinciples). It refers to “the practical ways in which the noble principles — dignity, the common good, the universal destination of goods, subsidiarity, solidarity and social justice — are implemented in real life” and “encompasses all dimensions of existence and opens the future to subsequent generations as well” (82). In practical terms, this refers to the public policy process of balancing market and government solutions to shared problems. Markets enable innovation and prosperity, yet often at the price of exploiting people and the environment. Governments can reduce the harm of market failures, yet often at the price of institutional domination that imperils human dignity and liberty. Social actors exercise agency to continually reshape the structures that constrain them, so this political economic balancing act goes on forever. While there is no perfect solution to social problems, sometimes it is possible to cultivate shared values and productive dialogue to maintain a healthy balance.

“Seen in this light, integral human development is the framework through which we can interpret the changes of our time, including those brought about by the digital revolution. Technological innovations, including artificial intelligence, are not neutral, for they can either foster participation and justice or exacerbate inequality, control and exclusion. For this reason, they must be evaluated by asking a crucial question: Do they truly help individuals and peoples to become more humane and fraternal, while respecting our common home and future generations? It is here that the principles of Social Doctrine become concrete criteria for discernment regarding the issues which we will address in the following chapters” (85).

Part 2 (I hope) will explore the political economic tradeoff space at the various levels of human intelligence, economic organization, and international relations (i.e., a classical liberal take on AI, the Market, and War).

[updated to correct typos]

[1] Bracketing after the manner of Edmund Husserl means remaining agnostic about questions placed “in brackets” while attending to experiential phenomena as they appear. It is not a skeptical negation but an abeyance of judgment. Thus, I can describe the character of my experience of space and time without having to make any commitments about whether I am a brain in a vat or a child of God. Whether or not the universe is one or many, monist or dualist, material or conscious, or accidental or created may matter very much as a point of fact for explaining what causes my experience to appear as it does. But I can still describe the “dependent variable” of experience without necessarily committing to an “independent variable” of physical or metaphysical causation. In the same way, it may well be that a Christian anthropology or theodicy is essential for explaining the historical appeal or validity of liberal ideas (or not), but we can bracket such questions to simply describe how social phenomena show up.

Lindsay, Jon R. 2026. “The More Dismal Science: Perspectives from International Relations on Military Automation” eds. Ajay Agrawal, Joshua Gans, Avi Goldfarb, and Catherine Tucker. The Political Economy of Artificial Intelligence. U Chicago Press.

The chapter reacts to discussion at the conference about the technical inputs to AI, such as chips, models, and data. A few participants made strong claims about how this may affect military power. But in my neck of the woods, power is not not just an economic production function but a political interaction. I thought the easiest way to demonstrate this to my friends in economics would be with the trusty bargaining model of war, which international relations stole from economics.

In the rationalist tradition, if war is costly, then actors ought to prefer a deal to fighting. They may still fight anyway if there is uncertainty about power, costs, or commitment. Tragic. Yet the bargaining model itself is silent about how actors should improve their lot. They might manipulate power, costs, demands, or uncertainty, which are the key parameters of the bargaining model. Preferences over them are exogenous to the model, and thus must be sought elsehwere (i.e., at the unit level in bureaucratic politics, elite psychology, strategic culture, etc.). Moreover, different technologies may be more or less suited to these various goals, as Erik Gartzke and I argue in Elements of Deterrence,

The chapter thus asks how AI might affect any of these parameters. The short answer is that it might affect them all, and in different ways. The paper discusses the counterveiling ways in which AI might affect power, uncertainty, costs, and demands. Whether and how it actually does so is an empirical question, but this is a good reason to be suspicious of any deterministic claims about how AI will affect war or coercive diplomacy. The upshot is that instead of focusing only on the technological inputs to AI capability (and policies that control them), it is important to also focus on how political preferences shape the AI means that support them.

The chapter has more detail, but here are some slides that make a similar argument:

(Edit: slide corrected)

But I realized that tech had started to feel empty. I felt like I was caught in an endless cycle—Prodigy, Altavista, Napster, Facebook, iPhone, Fitbit, Zoom, ChatGPT. National security experts kept pushing the same things with different words—the electronic battlefield, revolution in military affairs, network-centric warfare, effects-based operations, multidomain campaigning, synthetic cybernetic hypersonic quantum AI centaur enabled hybrid cognitive warfare in cislunar space. But every war was a bureaucratic shitshow of malfunctioning systems, endless fog and friction, political mismanagement, and civilian death.

I knew I had to make a change. I started to read history and philosophy. I visited art museums and took long walks on the beach. I sought out mentors and friends who could help me to find a better way. I realized that technology is not destiny. I realized that technology is not even technology, but rather technis, a system of interdependent structures and practices driven by the pursuit of efficiency and competitive advantage. I began to see how contingent political and economic choices have shaped and reshaped the sociotechnical infrastructure of capitalism and imperialism. And I saw that AI is neither autonomous nor intelligent, but rather a cultural system of prediction and extraction that has become a planetary-scale commodity fetish.

I’m not saying I’ve left it all behind. Not a day goes by that I don’t think about technology. But every day I get to make a choice about how and whether to use it or abuse it. Every day I can affirm that I am a biologically embodied, ecologically embedded, culturally extended, dynamically enactive human agent, not just a pattern of bits that can be replaced by a machine. Every day I can choose whether to connect with other human beings to share their joys and suffering, not just mindlessly reproduce the mythology of technology.

I know that there are many other people struggling with technology. Maybe you are struggling with technology as we speak. If so, maybe my story will help you to see that you are not alone. If the spirit moves you, please consider sharing your story, too. There’s no pressure. When you’re ready to break from the tech bros, I’m here to witness your journey. Together, we can take these twelve steps toward reclaiming our humanity:

1. We affirm that are powerless to resist the allure of tech, and our lives have become unmanageable

2. We believe that something other than tech can restore us to sanity

3. We decide to seek HALP—history, art, literature, philosophy—to appreciate the nature of humanity

4. We fearlessly study the political and economic character of infrastructure

5. We articulate the historical choices that have shaped the human-built world

6. We understand that different values and choices can rebuild a different world

7. We humbly strive to make better choices and live up to our values

8. We reveal the hidden violence and injustices that sustain our technocracy

9. We support policy to limit the exploitation of people and the planet, except when doing so would inflict more harm

10. We continue to assess new technologies and when they are harmful admit it

11. We seek through connection and reflection to improve our understanding of the human condition

12. Having had a humanist awakening through these steps, we carry this message to technologists everywhere

(This is Mai Mai, off the leash)

Alguns fragments del papir, from The Oxyrhynchus Archaeological Mission, run by the Institute of Ancient Near East Studies at the University of Barcelona

Meanwhile, SentinelLABS put out a report on a 21-year-old piece of malware they call Fast16, based on the name of a kernel driver compiled in 2005. Andy Greenberg also provides useful context in his Wired coverage. Fast16 appears to be a self-spreading worm that sabotages engineering and simulation software by installing a patch that alters floating-point unit (FPU) calculations. There are tentative links to the US National Security Agency (NSA) and Iran, as discussed below. The discovery in the wild of a sophisticated industrial sabotage payload, two years before the earliest sample of Stuxnet, should prompt a rethinking of cyber history during the G.W. Bush era.

Injected FPU-based calculations, from SentinelLABS, “fast16 | Mystery ShadowBrokers Reference Reveals High-Precision Software Sabotage 5 Years Before Stuxnet”

The 1,600-year-old mummy and the Fast16 malware share more than just two bytes. In each case, a fragment of information appears in a surprising place and forces us to reconsider what we know about the past. That information was just sitting there for a long time, buried in the infrastructure of an Egyptian necropolis or the VirusTotal database. For many years, it was unknown to the wider world. The people who put a poem in the belly of a mummy are all dead, and the people who put a bug in the belly of Iranian science are all sworn to secrecy. The lucky soul who was mummified was also important or rich enough to merit ritual embalming, and the clever exploit that was weaponized was also the product of a wealthy and important nation state.

Epistemic Sabotage

Fast16 is a wicked bug. By altering FPU calculations for numerical simulations, it subtly injects uncertainty into scientific models of physical processes. Equipment designed with flawed software might not work as intended or cause unpredictable accidents. This would force scientists to work harder to check their work, or perhaps even to question their own competence. The errors could be especially hard to detect if the self-replicating wormlet also infected other systems used to verify results.

Sabotage is the weaponization of friction. In the case of Fast16, the computational sabotage of scientific software is also epistemic sabotage of the scientist. The attacker creates information friction for the victim, which means causing a pragmatic mismatch between informational representations and material reality. When you change the construction of the technological prosthesis through which you understand the world, you also change the image of the world that you see. Fast16 was designed to create a flawed scientific image.

There are echoes of Stuxnet here. The famous worm was designed to just marginally increase the breakage of centrifuges, rather than cause catastrophic breakage, while carefully hiding its tracks. One of David Sanger’s sources in the US intelligence community said of Stuxnet, “The intent was that the failures should make them feel they were stupid, which is what happened.”

Epistemic sabotage depends on stealth, insofar as the point is to get the victims to misattribute the cause of information friction to their own incompetence. Much like Stuxnet, Fast16 contained numerous security checks to enable it to keep a low profile. Stuxnet sabotaged physical equipment by manipulating rotors and valves, while Fast16 worked further upstream to manipulate engineering R&D. Yet both created confusion that was hard to attribute to foreign tampering.

Another Stuxnet echo is that Fast16 may have targeted Iran’s nuclear program as well. The SentinelLABS team compared Fast16’s patching rules against contemporary software applications and identified three likely candidates: “LS-DYNA 970, PKPM, and the MOHID hydrodynamic modeling platform, all used for scenarios like crash testing, structural analysis, and environmental modeling.” Particularly interesting of these is the LS-DYNA physics modeling suite, originally developed by Lawrence Livermore National Laboratory (LLNL) and used in numerous industries. LS-DYNA is referenced in an Institute for Science and International Security report on “Iran’s Likely Violations of Section T [of the JCPOA]: Computer Modeling Relevant to Nuclear Weapons Development.”

The other matches are the Chinese Practical Structural Design and Construction Software (PKPM) and Portuguese Modelo Hidrodinâmico (MOHID). SentinelLABS states, “we cannot definitively identify the target.” Greenberg notes “that North Korea’s nuclear weapons development program also experienced numerous unexplained failures in the same time period.” This is all speculative, and it is possible that Fast16 was used against multiple targets.

In my opinion, the Iran connection is most plausible. Stuxnet or Operation Olympic Games, as I have argued elsewhere, was just one aspect of a multifaceted covert action campaign targeting Iran’s nuclear program for over two decades. The CIA’s Operation Merlin also employed a Russian scientist to deliver deliberately flawed weapons plans to Iranian scientists in March 2000 and beyond. About the same time, the CIA recruited the Swiss Tinner family to spy on Iran and provide sabotaged parts to the nuclear program. Fast16 fits right in with this MO.

Nothing to See Here

Another important bit of evidence is the connection between Fast16 and the NSA. In the Shadow Brokers leaks that emerged during and after the Russian campaign against the 2016 US election (there’s that number 16 again!), there was a module of NSA code called Territorial Dispute. The purpose of this module was to help an NSA hacker determine whether other friendly hackers were already exploiting the same host. Territorial Dispute contains a list of drivers and instructions for what to do about them, e.g., “PLEASE PULL BACK.” The alleged NSA tools all have an NSA cryptonym identifying them, and allied tools are identified separately, but there is one mysterious line that reads “*** NOTHING TO SEE HERE - CARRY ON ***”,”fast16”.

The nerdy Star Wars reference suggests that NSA knew about Fast16, but Fast16 was not NSA code. One possibility is that it is a tool developed by the CIA’s Information Operations Center. The famous Equation Group tools cover both NSA Tailored Access Operations (TAO) and CIA (according to one CIA leak, “mostly TAO some IOC”). As noted above, the CIA was actively involved in sabotaging Iranian scientific activity, and IOC would have provided another means to do this. Speculation, but plausible.

Another circumstantial connection is the use of the Lua scripting language in Fast16. A small subset of contemporary advanced malware also used Lua engines. This includes the Flame spyware that shared components with Stuxnet, and Project Sauron (aka RemSec) that targeted Russia and Iran. Juan Andres Guerrero-Saade describes Lua as a “telltale” for this supra-threat actor.

For obvious reasons I did not discuss Fast16 in my 2025 book and article on Stuxnet. Apparently, we now know that Stuxnet was not, in fact, the first cyber sabotage operation in the wild. There has long been speculation that a Soviet pipeline explosion in the 1980s might have been due to cyber sabotage, but I have my doubts. Thus, Stuxnet has long been described as a notable first. And it still is, insofar as Stuxnet plausibly caused physical damage based on IAEA reporting. Fast16, by contrast, targeted a software application rather than industrial controls. And we don’t know anything about its actual effects in the world. Maybe it resulted in flawed equipment that caused serious accidents. Maybe it resulted in naught.

All we can say with any degree of confidence is that the public forensic record of US cyber sabotage predates the earliest version of Stuxnet (version 0.5 in 2007). The Fast16 driver that showed up in a VirusTotal database in 2019 was compiled in 2005. There are internal references to version history, so probably there are even earlier versions of the capability. That is still well within the timeframe of the CIA’s Merlin and Tinner operations. Fast16 appears to be a sabotage operation prior to Olympic Games (Stuxnet) but parallel to many others.

It also bears repeating often that cybersecurity is not new. There are hackers who can legally drink at DefCon in Las Vegas today who not even born when Fast16 was compiled. Cyber is no longer an emerging technology, and hasn’t been for half a century at least, but rather one that can—and must—be understood historically.

Cyber Paleontology

All this all makes me think of Costin Raiu’s description of malware analysis as “cyber paleontology.” In his account of the Regin investigation he writes,

…what we first discovered in 2012 was a slightly damaged bone from unknown part of a monster living in a mysterious mountain lake.

Someone finding a bone might discard it and keep traveling, but in security research we collect things. The original discovery went to our collection of things stored in the backyard. We have many of these fractured bones from unseen monsters or maybe harmless creatures. Sometimes we hear about other bone fragments being discovered by others and this pushes us to take a closer look but at the early stages, without enough evidence to draw meaningful conclusions, it doesn’t make sense to go public about discoveries until you confirm that the monsters are real, big and dangerous.

We keep working in various channels to collect different artifacts that may or may not match some of the pieces in our collection. Sometimes we join efforts with other “paleontologists” and share our discoveries. Once we collect enough of [sic] bones from a monster to understand potential size, danger and possible habitat, we can start the next phase which is a real active investigation that might lead us to the mysterious mountain lake.

The discovery of Fast16 showcases cyber paleontology in action. SentinelLABS researchers knew about it for years. It was mentioned in the 2017 Territorial Dispute leak, and a sample was found in 2019, but not until several years later was its full significance appreciated. In the course of testing the ability of AI models to analyze old malware samples, Vitaly Kamluk finally cracked the mystery of Fast16 (reversing it the old-fashioned way, without AI). Now we must rethink how to fit all the pieces together. The classics of cybersecurity after Fast16 now look different than before it.

Every time some new tidbit related to Stuxnet appears I am both excited and worried. Excited because the Stuxnet era is an inherently interesting episode of secret statecraft during a key era in the maturation of cybersecurity. Worried because I always wonder if I’ll need to change my public interpretation of the case. Case studies in cybersecurity are challenging because the evidence is always incomplete. Analysts must do the best they can with partial information and reasonable theories to connect the dots. When new evidence emerges – from new reporting, opening archives, or the discovery of an old file in a cybersecurity database – it is important to revisit our priors.

In this case, I am relieved to say that the specifics of Fast16 appear quite consistent with my general interpretation of Stuxnet. Of course I would say that, but in this case, I think it’s also true. We already have evidence linking Stuxnet to activity from the late 1990s to the late 2010s, which leads me to view Stuxnet as evidence of a long game of secret statecraft rather than an isolated incident of cyber warfare.

I view Stuxnet as just the tip of the iceberg of US (and Israeli) covert action targeting Iran’s nuclear program (and Israeli decision making). Fast16 fits right into this pattern. It also fits the argument in Age of Deception that remote sabotage of subtle vulnerabilities requires a sophisticated clandestine organization with considerable operational capability exercising deliberate discretion.

But I also have many new questions. When was the earliest version of Fast16 developed and deployed? What other sabotage payloads are still out there waiting to be discovered? Where were they used? What effects did they have, direct or indirect, technical or political? Was the sabotage limited to the intended targets (like Iran’s nuclear and missile programs), or did Fast16 affect other, civilian, international technologies designed with LS-DYNA or its like? Can any unexplained industrial mishaps or infrastructure failures be attributed to it?

Some answers to these questions could change the received wisdom of the efficacy of cyber sabotage, especially the widespread assumption that physical damage is hard to hide. For if Fast16 is implicated in any physical damage, then that damage has remained misattributed for over two decades. Even so, such effects must be subtle indeed, so subtle that they are indistinguishable from normal friction, by design, and normal friction is ubiquitous in the human-built world. This is a great irony of highly sophisticated malware: it may be invisible to the point of irrelevance. Perhaps the story changes somewhat if this is viewed more as an information operation targeting scientific minds rather than simply a computational operation targeting scientific software. Perhaps the cybernetic continuity of mind and software is the point.

Digesting the Myth

And this brings me back to the Mummy. Some rich Egyptian from the 4^th century CE was mummified, and along with all the usual ritual elaborate preparations, an unusual fragment of Homer’s Iliad was placed in the body’s abdomen. The Homeric epic entered into this person, presumably to be taken into the afterlife along with all the other treasures of burial. The poem was literally in the belly.

It is hard not to think about digestion as a metaphor for understanding. We still speak of savoring or devouring a book or digesting the meaning of an author’s words. It is not too much of a stretch to suppose that a similar metaphor was available in 4^th century Egypt. Especially meaningful texts become part of us, incorporated into our being, informing the way we see and interpret the world. Myth is word made flesh, so to speak.

This distinguished Egyptian, or whoever paid for the embalming, must have really been into Greek classics. It is important to bear in mind that the Iliad was already over a thousand years old by the 4^th century CE. Best estimates are that the oral epic was written down in the seventh century BCE. (“Homer” is more of a personification of a complicated transcription process than an actual personage.) And the Iliad itself tells a story of an imagined past five hundred years before. This means that the Egyptian who was mummified was as far away from the Trojan War as we are from the mummy. For that fellow, the Trojan War was ancient history.

But why Homer? I can’t help but think about Euripides’ play, Helen. This fine piece of ancient Homeric fan fiction that reimagines the famous heroine in Egypt while the Trojan War is fought on her behalf. The tragedian’s version is inspired by a sixth-century poem by Stesichorus and a passage in Herodotus, as well as the Spartan episode in the Odyssey that links Menelaus (Helen’s cuckolded husband) to Egypt. In Euripides’ version, the Helen who elopes with Paris and goes to Troy—the casus belli of the war—is not the real Helen at all, but rather a cloud phantom (eidolon) that was created by Hera, goddess of marriage and jealousy. The real Helen was in Egypt the whole time. There she remained faithful to Menelaus, who happily retrieves her after the war.

The whole war, Euripides tells us, was a colossal illusion. All the suffering and the devastating wrath of Achilles, and the clever ruse of the Trojan Horse, was all based on a delusion of jealousy. Importantly, here, Egypt is where the truth resides, while the Greeks wallow in illusion. The Greeks got many of their gods from Egypt as well, so it makes sense that an Egyptian might feel a sense of ownership over the greatest epic poem of ancient Greece.

But why the catalog of ships? Maybe the arrival of ships to a distant shore is some sort of metaphor for death. The passage itself is an accounting of the whole Hellenic world, with squadrons arriving from Sparta, Ithaca, Crete, Athens, and so on to fight for the banner of Agamemnon. Maybe absorbing this account was a way of absorbing Greece. In digesting Homeric myth, this mummified Egyptian was repatriating it, much as a mythic warrior might eat the flesh of the conquered. Maybe it’s just a famous passage that people liked to read. Better scholars than me will surely provide some better theories in the coming years.

Symbolic Fragments

The Homeric fragment surfaces through traditional archeology, and Fast16 appears through cyber paleontology. In each case, social processes in the past imbued a symbolic information artifact with a hidden purpose. One was meant (perhaps) to guide a rich Egyptian Homer nerd into an afterlife of enjoyment. The other was meant (perhaps) to guide Iranian scientists into a lifetime of confusion.

Then, long after the original symbolic purpose had played out, for better or worse, someone happened to dig it up. Then, the symbols took on new meaning: a spiritual token prompts a rethinking of culture, and a digital file prompts a rethinking of sabotage. In each case, someone had the technical skills to interpret ancient fragments and recognize their modern significance, not to mention their enduring mysteries.

Both ask us to revisit the classics, such as the relation of Homeric myth to Egyptian society and the relation of Stuxnet-style sabotage to American statecraft. The meanings of each are somewhat in flux with these new discoveries, and hopefully more evidence and interpretation will be forthcoming. The new emergence of old evidence can unsettle ancient history or reinvigorate familiar myths.

Military Virtuosity

On Good Friday, Iran shot down a US F-15E near Isfahan. Combat search and rescue (CSAR) forces recovered the pilot quickly. After a desperate search, the weapons system officer (WSO) was rescued early Easter morning.

The initial reporting on the special operations mission to recover the airman is quite dramatic. As US and Iranian forces raced to find the downed WSO, he escaped and evaded on foot, climbing a 7,000-foot ridge and hiding in a crevice. The IRGC offered rewards for local tips and posted images of jet wreckage. The CIA located the WSO with some technical means and conducted a deception operation to mislead Iranian troops. There was extensive contact between US and Iranian forces, with multiple aircraft hit or destroyed. All US personnel made it back alive, several wounded.

UPDATE 4/6/26: “Mr. Trump gave a laudatory depiction of the rescue of a missing American airman shot down over Iran. The mission involved 155 aircraft [on par with the Maduro raid] and hundreds of people, he said, though “a lot of it was subterfuge” designed to lead Iranian forces away from the aviator. He also revealed that the plane, a U.S. Air Force F-15E, had been downed by a single shoulder-fired missile.”

The rescue is a testament to the skill and bravery of the US military, a bright point in a very dark war. US special operations is looking particularly impressive after pulling off both the Maduro raid in Venezuela and the WSO recovery in Iran, even as both of them could have gone wrong in so many ways. This CSAR mission will be one for the books, up there with the rescue of Bat-21 in Vietnam and of Marcus Luttrell after the Red Wings disaster in Afghanistan. It is a safe bet we will see this episode, too, on the big screen soon. Hollywood loves a hero story.

My father was a Navy SH-3 pilot who flew CSAR missions in Vietnam and helped to develop the Navy’s SERE curriculum. His name is on the side of this SH-3 on the USS Midway museum ship in San Diego.

CSAR is highly risky and uncertain since by definition something has gone badly wrong. Time is of the essence since the downed crew may be injured and enemy forces may attempt to capture them, giving them a propaganda victory and bargaining chip. Fog and friction is at its highest in this situation. CSAR forces must rush in without fully knowing what to expect.

Compensating for all this confusion, however, is the fact that the CSAR mission objective is crystal clear: recover the crew. Everybody know exactly why they are taking risks, and what they need to do. Whatever strategic ambiguities or misgivings may attend the war itself, a CSAR mission tends to create high unity of purpose. This same unity is good for morale in the first place: Aircrew can fly dangerous missions with the knowledge that the entire theater has their back.

But the inherent riskiness of CSAR, combined with the military’s firm commitment to carry it out, means that there is high potential for cascading losses. In this case, the initial shootdown of the F-15E — we still don’t know exactly how that happened since we’ve been told that the US enjoys air supremacy in Iran — triggered events that resulted in the losses of additional aircraft. During the rescue of the pilot on Friday, two HH-60 CSAR helicopters were hit by enemy fire, along with an A-10 providing fire support. The helicopters limped back to a base in Iraq while the A-10 crashed in Kuwaiti airspace; the pilot ejected and was recovered.

During the rescue of the WSO on Sunday, which had extensive air support including MQ-9 Reaper drones firing on Iranian forces within three km of the WSO, more aircraft were hit and some permanently lost. The details are hazy at this point, so a lot of this might be misleading, but it seems that two more HH-60s were hit and able to egress, and two MC-130s were unable to leave an austere airfield (i.e., landing on the ground) in Iran. They were destroyed by US forces, along with at least two and maybe four MH-6 Little Birds, to prevent them from falling into enemy hands. These aircrew were recovered by two more aircraft, probably MC-130s.

While I celebrate the safe recovery of US service members and the tremendous honor, courage, and commitment of everyone involved in this mission, I also can’t help but reflect on what it says about the risk we are all carrying with this war. To appreciate just how risky CSAR is, compare the aircraft lost or damaged in this one particular CSAR episode to all of the US aircraft lost in the war so far (in my fallible mental database):

• CSAR after 1x F-15E shot down

  • 1x A-10 shot down

  • 4x HH-60 damaged

  • 2x MC-130 destroyed

  • 2-4x MH-6 destroyed

• Rest of war

  • 3x F-15Es lost to friendly fire by Kuwait forces

  • 4x KC-135 Stratotankers to midair collision and Iranian strikes

  • 1+ E-3 AWACS to Iranian strikes

  • 1+ CH-47 Chinook to Iranian strike

  • 17+ MQ-9 Reaper drones lost/shot down

Ignoring the drones, more than half the aircraft lost so far were lost during the CSAR mission to pick up two aircrew from one downed jet. The price tag for just the lost aircraft is perhaps upwards of $450 million (estimate compliments of Gemini 3 Pro), and the whole operation may come close to $1 billion. But a successful CSAR, as the old MasterCard commercials have it, is priceless.

This will probably not be the last CSAR mission in this war, and the next one might not go as well. An opposed CSAR is a possibility as long as the war goes on, which is why the US military has well-trained CSAR forces on standby all the time. But this successful yet costly CSAR demonstrates the potential for costs and risks to snowball: one lost aircraft leads to many more, one downed airman can lead to many more casualties. This mission could easily have been a disaster had more planes been shot down or had IRGC forces found the WSO or other Americans first. SOF successes run right up to the wire of SOF failure. And the longer this goes on, the more we play Russian roulette with the worst outcomes in the chamber.

This risk loading naturally raises the question of whether it’s worth it. And this is on top of all the Iranian civilians killed, global supplies of oil, gas, and fertilizer choked off, a brutal regime more resolved than ever to acquire a nuclear bomb, irrevocable damage to US alliances around the world, and no end in sight. The heroism of the CSAR mission stands in stark contrast to the horrific miscalculation of the war.

Flagitious Politics

And this brings me to the other events of Easter Day. I was at a loss for words to describe Trump’s truth today. In just four lines—a marvel of concision by Trumpian standards—our president managed to combine illegal threats, grotesque humor, cheap talk, strategic escalation, vulgarity unbecoming, dehumanizing insults, religious sarcasm, and Easter profanity. I asked Gemini to suggest words that combine vulgar, profane, immoral, reckless, dangerous, and unhinged. It suggested vile, nefarious, licentious, depraved, and reprobate, but the best one by far was flagitious.

Thanks robot, I’ve learned something new today. Yet it seems that our president has learned nothing. After the Iranian regime did not fold on week one, he has continued to double down on the war. The high point of the war for the United States and Israel was the very first day, when the elder Ayatollah Khamenei was assassinated. It could only go downhill after that, as the regime learned that it could not only survive the onslaught but also inflict costly punishment on the rest of the world.

The reason that I say cheap talk and strategic escalation in the same breath is that Trump has threatened to bomb civilian infrastructure before, to send Iran “back to the Stone Ages, where they belong.” Trump often makes over-the-top threats and then back down, or TACO as they say. Yet he also made the decision to launch this war and stay in it, and he seems to be backing himself into a corner. The threat is not very credible, yet he just might do it anyway. The official name of this operation, after all is Operation Epic Fury, or OEFU (to distinguish it from Operation Enduring Freedom, or OEF, in Afghanistan). FU indeed.

And the worst part is, I almost think Tehran would welcome it. This is a regime that savages its own population and only stands to gain by having the Great Satan hurt it even more. It’s hard to think of a more reassuring signal to Tehran than that tweet, or truth, or trutheet? It’s a message that says to IRGC elite: hold fast, you are winning the battle of resolve. You are right to chant death to America, because the President of America wants your death as well, if he knows what he wants at all.

Both his threats and his offers to negotiate alike are not credible, so the wiser strategy for Tehran is to keep inflicting punishment until Trump gives up, as he inevitably will, but probably not before inflicting a lot more pain on a lot more people. At this rate, deterrence will be restored, but on Iran’s terms. Iran’s political position is only getting stronger as the war goes on, for its military is clearly able to keep fighting and dishing out pain for the global economy, despite being up against the magnificent American military. Something, something, something about winning every battle and losing the war.

Thus, on a single Easter Sunday in April, we see the highest demonstration of military competence and the basest example of political incompetence. American CSAR may be able to rescue a downed pilot, but it cannot rescue Trump’s flailing policy.

But if you think about it, “Secretary of Defense of War,” without the comma, is inadvertently brilliant. Granted, it has to compete with “Kegseth,” “Dumb McNamara,” and “Whiskey Pete” for best nom de défense guerre.[1] But what does Hegseth really seem to do every day if not defend his peculiar idea of war? This would also give us the chance to rename his organization the Department of Defense of War, which has the excellent acronym DoDoW, pronounced either as “Doh! Doh!” or “Da-Dow!,” the latter of which rhymes with “Kaplow!”

The Pentagon is still officially the Department of Defense unless Congress votes to change its name. President Trump signed an executive order (so many executive orders) in September 2025 on “Restoring the United States Department of War.”[2] Apparently it will cost somewhere between $125 million and $2 billion by the time we change all the signage and polo shirts. Given that the current war with Iran is costing us about $1 billion per day, this means for the cost of rebranding the DoDoW you can buy two days of Kaplow!

I really feel for Colin Jost, because his impressions on SNL are hardly any funnier than the real thing. Or at least Hegseth would be funny if he wasn’t so damn dangerous. Satire is sometimes the only way to survive an infuriating situation, but Hegseth is so fundamentally unserious and incompetent that even sarcasm doesn’t help. The stern and silent stares of all the GOFOs (General officers / flag officers) that Hegseth irresponsibly summoned for his anti-woke pep talk last year are perhaps a better response. They have a job to keep and actual troops to lead, of course. But when does silence imply consent?

Well, here we go. Hegseth is abysmally unqualified for his post, an unredeemable embarrassment to the profession of arms, and clearly a very insecure man. Plenty more qualified than me have said the same thing, e.g., here, here, and here. There aren’t enough synonyms for “outrageous” and “offensive” to describe the way Hegseth talks about war, asks Jesus to help him wage it, and scorns the law of armed conflict. When he is on stage with General Caine to talk about Iran — a show I like to call “Caine and Unable” — Hegseth’s words seem to refer to the war, but he’s not really talking about the war. Instead, he enters an erotic trance as he delights in bombing, sinking, killing, and of course, maximum lethality (not tepid legality).

I am reminded of a quote from Ernst Junger in Storm of Steel, a complicated memoir of the First World War that delights in its horrors: “Grown up in an age of security, we shared a yearning for danger, for the experience of the extraordinary. We were enraptured by war.” But even that is too highbrow for Hegseth. Junger was actually fighting after all. Hegseth is more like Geardo from those “Bob on the FOB” comics that used to float around Iraq, a Fobbit who delights in warrior costumes while having little to do with war. Turns out the Pentagon offers its chief news anchor far better production values than Fox News could ever afford.

But the most disturbing thing about Hegseth is not the terrible OPSEC on Signal, or disparaging reliable allies, or wasting millions on cosmetic vanity, or holding a GOFO jamboree, or firing highly honorable officers, or not hiring equally qualified ones, or seeking revenge on Senator Kelly, or allowing helicopters to overfly Kid Rock, or trivializing the killing of thousands of civilians, or holding prayer meetings at the Pentagon, or even drawing the condemnation of the Holy See. The most disturbing thing is that he is precisely the leader of the Pentagon that the White House wants.

Hegseth is the perfect SecDef-o-War for an administration that communicates in puerile memes that mock the deadly seriousness of war. The White House is a warnography site,[3] and Hegseth its biggest fan. His primary mission is to flatter the president at all costs, and to do everything possible to make sure that a bad war looks good on X. In this he succeeds marvelously. Here I am not even being sarcastic but rather acknowledging that civil military relations in an authoritarian state are usually far more about ensuring internal security for the leader rather than national security for the people.

Hegseth gives Trump exactly what he wants: Performance over competence, loyalty over ability, sycophancy over seriousness, tactics over strategy, charisma over intelligence, domestic faction over foreign policy. And when the war starts to bring down the presidency, you know who will take the fall: “Pete, I think you were the first one to speak up. You said, ‘Let’s do it.’” He is the perfect secretary of defense of war.

[1] By the way, yes, this really does link to a PBS Newshour page for “Pete Kegseth” as of April 3, 2026. It looks like it redirects from a single “go deeper” tag on a story about the SecDef-o-War’s white nationalist tattoos. I don’t know if some PBS intern just really hates him that much, which is understandable given that PBS funding was on the chopping block, or PBS got hacked. If the latter then it goes right up there with the Google phonebook showing “Epstein Island” for calls to the White House. Go deeper indeed.

[2] Like everything in the Trump administration, this is a bit misleading because for most of our country’s history we had a Department of War and a Department of the Navy. The Department of War (1789) was technically older than the Department of the Navy (17891798) and nominally the senior cabinet position, but the land and sea services kind of did their own thing. The Department of War was abolished in 1947, replaced with new Departments of the Army and the Air Force, and a new Department of Defense was created in 1949 as the one ring to bind them all. So renaming The DoDefense is not really a restoration, since it is functionally and historically different than the old DoWar.

[3] That joke will only make sense to readers of a certain age. Turns out whitehouse.com has moved on to political gambling.

Stand by for a lot of conspiracy theory about this one. Hitting one school can be written off as a tragic coincidence based on negiligent intelligence work. Hitting two schools on the same day starts to look like a pattern.

But what is the pattern? Here are two pictures from the NYT. The first is the school in Minab hit by the Tomahawk(s), where the girls were in class, and the second is the school in Lamerd hit by the PrSM, where the girls were playing volleyball. They are eerily similar, with the school in the corner.

Some will see evidence that the US military is deliberately hitting schools, which is a war crime. I don’t buy this interpretation because it very much plays against type (of the hyper rationalist and legalist targeting processes of the US miltiary that for the same reason sometimes makes some horribly myopic mistakes).

Some will see evidence that the Iranian military has a bad habit of turning parts of its bases into schools. What is going on here? Were the schools originally created for IRGC kids? Does the IRGC cynically hope to use schools as shields, or even more cynically to provoke a war crime? Is it just cheap real estate and a horrific coincidence? I don’t know.

Some will also see evidence of legacy target development in an intelligence database. Once again, there is probably a big rectangle drawn around the whole compound, in ignorance that one corner was turned into a school ten or so years ago. The target was probably marked as “no change” by whatever analyst(s) reviewed recent imagery during routine target scrub(s). I lean in this direction, i.e., that Lamerd is probably a good-old-fashioned intelligence failure like Minab, all within a heady atmosphere of quick-turn planning and accellerated timelines, based ultimately in unrealistic strategic ideas about how the war might unforld. But I admit that the coincidence is curious.

Some will also see evidence of a rush to strike one thousand targets (T1K!) with the help of AI. If AI was involved, I still think that it was at most pulling from a database of human-validated targets and perhaps contributing to a process that was rushed more than it should have been. But with the new target in Lamerd hit by the Army’s long range PrSM rocket, the case for AI involvement is marginally stronger. The Maven Support System after all was experimentally developed with the 18th Airborne Corps for managing surface-to-surface fires. So the speculative link is closer, but still extremely speculative. (Bear in mind again that the targeting AI in Maven is not Claude, but a more bespoke image classifier or planner adapted for the purpose by Palantir).

It is also possible that Lamerd is not, or not only, an intelligence failure, but an operational mistake. Compare in the photos above that the rest of the IRGC facility seems to be untouched, while the damage is to the civilian school and residential area. This is very different than the Minab strike, where all the IRGC buildings are hit too. So maybe the PrSM rounds were not very precise.

While the reputation of weapons offices is not something I generally worry about much, this is pretty embarassing for the PrSM program. According to the NYT, “On March 1, U.S. Central Command posted a video of a PrSM launch from the first 24 hours of the war. Days later, Adm. Brad Cooper, who leads Central Command, said the PrSM had been used in combat for the first time. The military has been touting its debut. Since the weapon is so new, it’s more difficult to assess whether the PrSM strikes in Lamerd were intentional, stemmed from a design flaw or manufacturing defect, or were the result of improper target selection.” PrSM is so new that I had to look it up. The Lockheed-Martin page crows, “The Precision Strike Missile is the next generation in our family of precision-strike, surface-to-surface weapon systems that support deterrence and deliver enhanced capabilities for neutralizing, suppressing and destroying targets at depth.” Ouch. Precise enough to strike families, more like. The misplaced warnography is not a good debut for a new weapon, but perhaps it is emblematic of this whole war in its way.

Finally, some will see evidence that Operation Epic Fury is not quite the operational success it has been represented to be. And that would be fair. Mistakes happen in war, but multiple mistakes like this ought not to happen. Both incidents may just be really bad luck, but these errors are a huge black eye—really two black eyes—on what otherwise seemed like an impressive military effort in the first day or two. Yet as the divergence between strategic planning and military operations becomes all the more glaring as the war drags on, these new blemishes in the opening days become important harbingers of the unfolding disaster. A politically foolish war is NOT likely to be won even if the military operation is flawless, but now even that must be in doubt. I’m not exactly sure what the discount factor is for morally outrageous mistakes against complicated military campaigns, but it must be even larger when the strategic management of the war is unconscionable as well.

There is still a lot we do not know, and we should be careful to qualify speculation about what happened. I imagine that we will learn more in the days ahead. There will be an investigation if there is not one already. We are still waiting to learn more from the other investigation, of course. While the news might not be as bad as conspiracy theorists or state propagandists will immediately assume, it is all still really bad.

[a few typos corrected and the difference between the targets clarified]

Déjà vu

When I first saw the satellite images of the Shajareh Tayyebeh elementary school in the Times’ reporting on March 5, 2026, I had a sickening feeling. I knew that I was looking at an intelligence error rather than an operational mistake. The point of impact was just too precise, right in the middle of a building. The school building was right next to a larger IRGC naval base, which also had buildings with very similar damage. Someone, somewhere, had assessed that it was all one facility, somehow designating the girls’ school as part of an IRGC compound. This was not an errant bomb but a targeted strike.

I assumed at the time that the US was probably responsible. Early reporting stated that Israel focused on leadership targets while the US focused on military and naval targets. The US Navy would have had a special interest in taking out IRGC Navy targets. I quickly discarded (President Trump’s) idea that Iran bombed its own base, given the timing and precision of the multiple strikes, or that the school was knowingly hit as a school. Given the highly bureaucratized targeting processes in the US military, with multiple levels of review and many lawyers lurking around, conspiratorial perfidy or nefarious hacking seemed exceedingly remote. But for the same reasons, an intelligence failure seemed all too possible.

I thought so because I had seen something like this before. Almost three decades ago, I served as an intelligence officer during Operation Allied Force, the Kosovo Air War of 1999. During that war, the US military accidentally blew up the Chinese Embassy in Belgrade. A mistake of that magnitude seemed unimaginable at the time. How could we possibly not know what we were hitting, and then hit that of all things? Anyone who walked by the street in Belgrade would plainly see it was an embassy, but US analysts only looked at overhead satellite imagery. The Chinese had told their counterparts in the US Department of State that they had relocated to that new building a year before, but that info had not been input into the military database. On the Master Air Attack Plan that fateful night in May, therefore, it was just “Belgrade Warehouse 1,” misidentified as the Yugoslav arms bureau, which was actually a few hundred yards away.

An investigative report, again in the New York Times, carried an image of the PowerPoint slide of the target that turned out to be the Chinese embassy (above). Simplified representations can take on a life of their own, projecting an air of authority that their flawed construction does not warrant. This particular target was reviewed and validated several times by several agencies. The slide was even briefed to President Bill Clinton (and the review of targets scheduled to go to the promiscuous president—draw your own parallels—was affectionately known as “the POTUS crank”). But the CIA’s fatal mistake was inadvertently repackaged and repeated as circular reporting, so all that multisource validation was actually invalid. Yet the operational targeteers did a bang-up job in weaponeering the building, and the aircrew delivered their bombs precisely to their designated mean points of impact (DMPIs). It turns out that everything on the slide was correct, except for the “linkage” which was dreadfully wrong. Even the casualty estimate was surprisingly accurate, as three Xinhua reporters were killed and at least twenty people were wounded.

I wrote about this incident in the introduction of my first book, which is about technology and knowledge in war. (I wasn’t allowed to put this slide in the book because of its “secret” marking, even though it was printed on the front page of the New York Times, so I am glad to finally post it here.) In many ways, this terrible mistake inspired me to go to grad school in political science, to understand how and why organizational perception could go so wrong, despite, if not because of, all the great intelligence and technology available to the US government.

The short answer is that maintaining the proper relationship between simplified targeting representations and a distant battlefield reality depends on a complicated dance of coordination that tends to hide its enabling work. Uncoordinated organizations tightly coupled to the strategic environment become epistemically brittle. Spectacular targeting errors are a kind of normal accident.

So, after hearing about the elementary school in Iran, I went back to this account in my book. I found this grim passage: “But mistakes were made….After each blunder, NATO staff officers reviewed and adapted their processes to avoid making the same mistakes. Then they made new ones.” Indeed, 27 years later, despite all the improvements in targeting processes across all the targeting mishaps in Iraq and Afghanistan and Pakistan and elsewhere, “They made new ones.”

Apparently the Shajareh Tayyebeh elementary school was indeed on a U.S. target list, but it was misidentified as a valid military target. The earlier New York Times story on the Chinese embassy had described “an immense error, perfectly packaged,” amid “a chaotic scramble to meet the demand” by General Wesley Clark to find “two thousand targets,” a goal personnel described as “T2K.”

Sound familiar? In Iran, the US military was also scrambling to bomb a thousand targets on the first day of the war (T1K?), on a schedule that was apparently accelerated by political machinations between Washington and Jerusalem and the intelligence trigger on the elder Ayatollah Khamenei. Pentagon briefings meanwhile have been unusually focused on tallying numbers of strikes (T6K by now?), while at the same time mocking the rules of engagement that are meant to protect civilians.

Here is a plausible story

There has been speculation that AI might be to blame, but I am dubious. Maybe Maven and Claude helped make the rush to a thousand targets possible, and maybe Palantir pulled some recommendations from a database of validated targets for targeteers to consider. But I do not think AI is to blame for this particular mistake, or the validation of the target in the first place. It is much more likely that the US military fucked up the old-fashioned way.

The school was apparently part of the IRGC navy compound before 2013, i.e., a legitimate military target. I assume that the entire facility had been given a BE (Basic Encyclopedia) number and appeared in the MIDB (Modernized Intelligence Database) of targets worldwide. This target, along with some DMPIs, including a DMPI on the not-yet-a-school building, was probably added to the counter-naval annex in some contingency plan (CONPLAN) for war with Iran. This plan was probably sitting on the shelf for years, regularly reviewed and updated in routine theater targeting conferences and again with every geopolitical crisis with Iran (so many crises).

But then the Iranians created a school sometime by 2016. I don’t know why they put a school next to the base. Maybe it was cheap real estate. Maybe it started off as a navy school. Maybe the IRGC cynically thought that a school would provide some protection from law-abiding invaders, in which case their faith in US intelligence was woefully misplaced. But even if that were the case, it would be no excuse not to notice that it was a school. The schoolhouse had an adjacent playground, and new walls were added to separate it from the base, and a military watchtower was removed. With its pink walls and artwork, it would have been obvious to anyone walking by that it was a school (much as the Chinese embassy would have been obvious from the ground).

But US intelligence analysts squinting at black-and-white electro-optical satellite imagery obviously missed the signs. I assume that this target went through a few rounds of routine review at DIA (Defense Intelligence Agency) or NGIA (National Geospatial Intelligence Agency), or maybe some reserve unit tasked with going through the massive backlog of targets. The reviewing analyst did not notice anything unusual, because they weren’t looking for it. Once you’ve been looking at hundreds of targets, it takes a lot of discipline to keep seeing new targets, or new images of old targets, with fresh eyes. The existing assessment in the database has a certain inertia, so it is easier just to leave it unchanged rather than make a fuss about changing things. Thus the target was validated, probably several times in several routine reviews, and again during the reviews of the actual campaign plan.

I imagine that there is a PowerPoint slide for this target, no doubt with much better graphics than its 1999 predecessor. This slide will have a DMPI on the elementary school that is not labeled as an elementary school, but rather as just one of several buildings in the target facility. The slide will have been briefed, probably several times, along with hundreds of other slides, to various audiences of intelligence analysts, staff officers, operational commanders, and even JAGs (lawyers), all of whom were tired, hurried, stressed, and maybe also a little bit excited about the impending adventure.

The target and its DMPIs were then approved and forwarded on to the targeteers at the maritime component commander, who loaded up the coordinates in their mission planning system and beamed the data off to an Arleigh Burke destroyer in the Gulf. I am just speculating, of course, but something like this likely happened. When the multidomain fireworks began, the ships received their orders to launch their Tomahawk Land Attack Missiles (TLAMs)—a weapon that is now over forty years old and was used in the First Gulf War (the shock of the old!).

The ship’s commander and crew, who did everything right within the scope of their mission, probably had no idea where their missiles were going after they threw the firing switch. I’ve always found it strange that modern naval headquarters can have far better situational awareness than the ships actually doing the shooting. The crew and the entire US military establishment only found out about the mistake the same way that the rest of us did.

Atrocity by accident

When the first media reports trickled in over the weekend, the idea of hitting a girls’ school seemed so outrageous that it had to be propaganda, the kind of thing that authoritarians always say. But then the images of grieving mothers and rows of tiny graves began to appear. President Trump tried to pin it on Iran, even suggesting they used a TLAM which they do not have and could not deliver, while Secretary Hegseth moved quickly back to his puerile gloating about lethality and vengeance. The unseriousness of the leadership and the seriousness of this disaster is a study in contrast.

At least the US military appears to still have enough professionalism to investigate this tragedy. I hope the investigation is unbiased and thorough, so that we can learn what actually happened. While I expect that organizational routine will be a big part of the story, this is far from just a routine mistake. It would be martial malpractice to just shrug and mumble, well, fog of war, you know, shit happens. I would very much like to see someone take responsibility, let alone be held accountable, but I am not holding my breath.

The shambolic disaster of the Shajareh Tayyebeh elementary school will stand out forever against all the operational prowess of Epic Fury, much as Abu Ghraib overshadows over the shock and awe of Iraqi Freedom, and the Chinese embassy looms over the airpower victory of Allied Force. I would be remiss not to also mention the shootdown of an Iranian airliner by the USS Vincennes in 1988, during a previous militarized clash between the US and Iran, which killed 290 innocent civilians. (US presidents apologized for three of the four tragedies above; I’ll let you guess which of these things is not like the others.)

Horrific wartime targeting mistakes may be normal accidents, but we should never normalize them. The decision to go to war is a decision to roll the dice on awful outcomes that are both unimaginable in their specificity and perfectly predictable in their generality. This time, the price of a string of compounding mistakes is measured in a hundred tiny lives that had only just begun. And as long as there will be war, there will be a next time.

And…it is the second time this year that US diplomatic negotiations were half-hearted at best, a pretense more likely, and maybe even a cynical lure. It’s also the second time that a US president who was elected on the promise of ending foreign wars has launched another war of choice (or maybe it’s the ninth time, if you consider the bombing of Iran, Iraq, Nigeria, Somalia, Syria, Venezuela, Yemen, and many pirates of the Caribbean during Trump’s second term). Again there was no authorization from or even prior notification of Congress, and no legal rationale for a preventive war that is surely illegal under international law. Again Trump announced the war from his palace in Mar-a-Lago and his Truth Social platform, followed by a parade of conflicting policy justifications from various officials thereafter. And again the world is left wondering, in the wake of an operationally brilliant but strategically dubious venture, what the hell was the plan for whatever happens next?

The comparison between Venezuela and Iran almost writes itself. Trump himself has referred repeatedly to his Venezuelan venture as his ideal model for his Persian excursion. His participation in a joint war with Israel last summer to “obliterate” Iran’s nuclear program, also without any US casualties, likewise emboldened him to opt for war again. But if you play Russian roulette enough times, eventually things turn out like the Deer Hunter.

Disanalogy and Disaster

The fact that Venezuela and Iran seem so comparable is really quite strange. These wars are radically different. The former was a special operations raid to kidnap Nicolas Maduro, the latter a reckless gamble on regime change via strategic bombing. This is “special” in a very different sense.

The administration described Venezuela as “law enforcement apprehension,” not even a war, but at least Trump has the decency to acknowledge that Iran is a proper war. President Nicolas Maduro is now awaiting trial in New York, while Ayatollah Ali Khamenei is dead (which is a stark departure from an executive order signed by President Gerald Ford fifty years ago prohibiting the US assassination of foreign heads of state). The Venezuelan regime remained intact after its leader was abducted and continues to govern its country, as ineptly and cruelly as before, but the stock of Iranian officials who might be able to run the Iranian regime is being decimated: “Most of the people we had in mind are dead,” said Trump. The Chavista regime was (and is) corrupt and unpleasant, but the brutality and institutionalization of the Islamic Revolution are in an entirely different league.

The fallout from Venezuela has been relatively limited (so far), but the war in Iran is spinning out of control. More Iranian schoolgirls died in a single errant strike on the first day of the war than all the Venezuelans and Cubans killed in the Maduro raid. The war has killed as many as a thousand people so far, and I fear that number will greatly multiply in the days and weeks to come. The Iranian people now find themselves caught between the bombs of the United States and Israel and their own brutal government, which has killed many thousands of Iranians in this year alone.

Nor is the suffering limited to Iran. People in several other countries have been killed, including six US service members in Kuwait and eleven Israeli civilians, as Iran lashed out across the region. US bases received inadequate warnings, and US embassies closed haphazardly. Americans in the region were initially told to find and fund their own egress from the region, before an evacuation was cobbled together. The failure to plan for these perfectly predictable repercussions is astounding.

The most important difference between these episodes is that Operation Absolute Resolve was a unilateral American adventure in its hegemonic backyard, but Operation Epic Fury is a coalition war with Israel in one of the world’s most volatile regions. Perhaps the stopping power of water helps to contain the repercussions of the Venezuela raid, but any war with Iran was always going to drag in its neighbors. Indeed, the war is already spreading. The Gulf states, once keen to present themselves as oases of economic stability, now scramble to intercept Iranian drones and missiles. What happens when Iranian refugees start fleeing the country, and where will they go? Neighboring Pakistan and Afghanistan are currently at war, while Türkiye and Iraq are already flush with refugees from the ongoing strife in Syria. Those latter two states are also wary of Kurd and Baloch separatist movements in Iran that may take advantage of the internal chaos.

Israel has also opened another offensive in Lebanon against Hezbollah, the Iranian proxy that Israel already devastated in late 2024. Reprisals from other Iranian proxies should be expected against Israel and others. Like it or not, the United States is now very much an enabler of Bibi Netanyahu’s grand ambitions for regional transformation. Secretary of State and head archivist Marco Rubio even made the argument that the United States had no choice to but to join in with its headstrong ally, since the US would be attacked anyway. In international relations, we call this chain-ganging or entrapment. Rubio’s rationale for preemption (i.e., action based on the threat of imminent harm) is laughable, as it is quite apparent that Israel and the United States have been collaboratively planning for prevention (i.e., action based on the threat of possible harm in the future) for months. Israel’s initial strikes, moreover, were fundamentally enabled by US intelligence and military support. We started this war together.

The situation is fluid and unpredictable. Take everything that anyone says about this war with a lot of salt (in case the news is not raising your blood pressure already). New details are emerging hourly about an event that no one really understands very well – even, or especially, those who are making decisions about it. Maybe Trump’s rash gamble will break the right way, against all odds. Maybe a new model Iran will emerge from the ashes of a deeply entrenched theocracy, recognizing Israel’s legitimate right to exist and stabilizing rather than destabilizing the region. Maybe I will become fluent in Farsi and Balochi overnight and start writing epic poetry about the rupture of liberal order. But in the reality we now inhabit, it is much more likely that we are just pumping gas into the flames.

Reckless Politics

These two wars are radically different, but there are at least two reasons why they seem so similar. First is that they are both run by the same clown show. In each case, the Trump administration has failed to articulate a coherent strategic rationale for military action. Instead, it has offered up many different, shifting, and contradictory justifications.

So, what would you say…Absolute Resolve was about? First it was all about the drugs, even though Venezuela mostly shipped cocaine to Europe rather than fentanyl for America, and Trump had just pardoned a former Honduran president imprisoned on drug charges. Trump also said Venezuela was about the oil, even though the United States is already a net oil exporter and American oil companies have little interest in investing there. Or maybe it was about immigration enforcement, even though regime change tends to produce more refugees. Perhaps it was about liberating the Venezuelan people, except that all the corrupt Marxists are still in charge. Maybe it was really about liberating Cuba, eventually, somehow, or demonstrating the Trump corollary to the Monroe doctrine, or sending a message to China about hemispheric hegemony, or announcing the end of the rules-based international order to feckless Europeans, or reestablishing deterrence with some dramatic flair, or giving Pete Hegseth another reason to crow about lethality. Maybe the war was also about the Epstein files, or Trump’s sagging poll numbers, or just some sort of weird performative ritual for an aspiring autocrat desperate for military glory.

Probably Venezuela was a little bit of all those things: a coalition of courtiers who each had their individual reasons to like the war, and no one ever gave the boss a compelling reason why not to do it. Log-rolling and groupthink, without a veto-player in the room, persuaded a vain and impressionable president to gamble on a one-and-done in Caracas. Except that it wasn’t just one. Emboldened by success and drunk on hubris, the administration soon had another opportunity to repeat its success. But this time the log-rolling coalition also included the Israeli Prime Minister and security forces, who all saw a golden opportunity to realize a decades-long dream.

So, what is all the sound and Epic Fury really about? Perhaps the war is meant to liberate Iranians from their oppressive theocracy, even though it now seems like the plan was for some newly dead regime members to take over after the Ayatollah. Maybe it is about eliminating the nuclear program that was already obliterated but is now reconstituted. Or it is about eliminating intercontinental ballistic missiles that do not exist. Or eliminating the Iranian navy that might close the Strait of Hormuz, which was open before shipping companies suspended operations because of the war. Some of those ships carry oil, by the way, quite a lot of it, so maybe this war can be about oil, too, because it is certainly raising gas prices. And of course, this war is very much about being sweettalked by an ambitious ally, and preempting Iranian retaliation for our preventative war of choice, and then somehow bringing peace to the Middle East by burning it to the ground.

Maybe the war is about changing the regime without “a so-called regime change war,” as Hegseth put it in his unhinged presser on Monday, or showing that “capable partners are good partners, unlike so many of our traditional allies who wring their hands and clutch their pearls, hemming and hawing about the use of force,” and thus at the same time showing that the liberal order is finally, well and truly, as dead as the Ayatollah. Speaking of the Ayatollah, perhaps this is all about liquidating a personal grudge after the Supreme Leader authorized assassination attempts on Trump. And if we are going to assassinate the leader, then we might as well take the opportunity to gut the regime and cripple its military capabilities as a bonus.

The war is also a great opportunity to stick it to the Democrats, who were never willing to do what must be done. And there are still those pesky Epstein files to distract from. A high profile assassination might do the trick. Yet it turns out that even the death of an Ayatollah is not universally popular with the American public, and Trump is unlikely to have so dramatic a “win” throughout the rest of this war. On the contrary, it is a safe bet that the war will get uglier, bloodier, and costlier.

Thus, point one, the political ends of both operations are multiple, shifting, and incoherent. But strategy is a relationship between ends and means (and ways, too, if you prefer to parse your policy instruments more finely into things and actions). And the means in both cases are provided by a highly capable and incredibly competent national security establishment. Without a reliable military tool at its disposal, the Trump administration would be all sound and no fury. With it, many different motivations can find a reason to wield it. And thus, point two, military professionalism provides the opportunity to turn incoherent strategies into dangerous realities.

Impressive Professionalism

This is not the place to analyze the military operation or speculate on its implications for cybersecurity in any detail. Maybe I will do that elsewhere. Suffice it to say that we are all very impressed, especially at the precision intelligence that located the Ayatollah and senior regime elements. Apparently, Israel’s Unit 8200 hacked the regime’s security and traffic cameras, a classic intelligence strategy of subverting security measures to gain access, to monitor movements of regime elite. It is hard not to think also of Mossad’s pager operation targeting Hezbollah, also prepared years in advance, taking advantage of a communication security measure (using pagers to avoid vulnerable cellphones) to inflict grave damage at the beginning of a major preventative war against a dangerous Israeli adversary.

It is also interesting that HUMINT again played so central role amidst all the technological sophistication. As in the Venezuelan case, the CIA apparently had an inside source who confirmed the presence of the Ayatollah on his compound. This feat impressed me because so much of the CIA HUMINT presence in Iran was devastated by counterintelligence failures seventeen years ago, but clearly the Agency has been able to rebuild its networks. The CIA then shared this information with the Israelis, who then conducted the decapitation strikes, while the US military went after air defense, missile, and other IRGC targets.

General Dan Caine, Chairman of the Joint Chiefs of Staff, and good cop to Hegseth’s crazy cop, later described how space and cyber effects enabled Operation Epic Fury:

“The first movers were USCYBERCOM and USSPACECOM, layering non-kinetic effects, disrupting and degrading and blinding Iran’s ability to see, communicate and respond. At H-Hour, the beginning of major combat operations, 01:15 local Eastern Daylight Time, 9:45 local A.M. Tehran time, as dawn crept up across the Central Command AOR, the skies surged to life. More than 100 aircraft launched from land, sea, fighters, tankers, airborne early-warning electronic attack, bombers from the States and unmanned platforms forming a single, synchronized wave. This was a daylight strike based on a trigger event conducted by the Israeli Defense Forces, enabled by the U.S. Intelligence Community….

In the initial phase, CENTCOM’s focus was systematic targeting of Iranians’ command and control infrastructure, naval forces, ballistic missile sites, and intelligence infrastructure designed to daze and confuse them. Coordinated space and cyber operations effectively disrupted communications and sensor networks across the area of responsibility, leaving the adversary without the ability to see, coordinate, or respond effectively. The combined impact of these strikes, swift, precise, and overwhelming, has resulted in the establishment of local air superiority….

Beyond Air Force and Navy, aircraft strike packages, coordinated salvos of precision standoff weapons from across the region, have neutralized hundreds of pre-planned targets. As I mentioned earlier, in support of these kinetic operations, USCYBERCOM and SPACECOM have continuously layered effects to disrupt, disorient, and confuse the enemy.”

This speech is uncannily reminiscent of a similar description Caine gave of Operation Absolute Resolve two months before. There, too, “the United States began layering different effects provided by SPACECOM, CYBERCOM, and other members of the interagency to create a pathway.” The striking similarity of US joint doctrine in these two operations is reflective of how normalized high-tech wizardry has become in US military operations. Cyberwar is boring.

There is much more to say, but the only point I really want to make here is that all of this—gestures vaguely toward multidomain layering—is all very impressive, all very hard, and all very dependent on extensive preparation and expensive infrastructure. To the extent that all the cyber and intelligence operations seem so successful, and we really don’t know much about any of it, I expect that this is due to tremendous investment by the United States and Israel in their organizational capacity—both unilateral and joint—and a concerted effort to develop multiple lines of operations. Several decades of focus on Iran have laid the foundations for what we saw over the weekend (and there is still so much that we do not see!).

Indeed, the old Nitro Zeus preparations from the Stuxnet era find their fruition in the military cyber operations today. I would not be surprised, once the archives are opened at some point in the future, to discover that this lineage is quite direct. Bottom line, you don’t get to layer on useful cyber effects in war unless you carefully prepare long in advance. And the same goes for the rest of the multidomain symphony that the U.S. military orchestrates so ably.

Awkward Ambivalence

Now, here I am writing all these nice words about the military operation, but in the previous section I was complaining about its incoherent political direction and terrifying political outlook. Given the terrible price paid by many innocent people for this cynical political theater, it feels almost lurid or grotesque to admire military professionalism.

I think that we need to hold on to this painful contradiction and not rationalize it away. Truth be told, I have been personally experiencing this ambivalence for a long while. On the one hand, I am proud of my service in the US Navy, and I believe that the military profession is both honorable and intellectually fascinating. On the other hand, I am increasingly dissatisfied with the trajectory of US national security policy, to say nothing of the general tendency of most military ventures to cause more harm than good. And it turns out that both of my hands are connected to the same body.

There is more than just an uncomfortable contrast here. A professional military is very good at defining and solving complex engineering problems, like synchronizing effects across multiple domains in a widely distributed geographic area with support from multiple government agencies and coalition partners. The engineering solution at the operational level helps to buffer out noise and controversy at the political level. This is helpful and even necessary to provide stable processes and structures that enable a military organization to function amidst the violent fog and friction of war. But in an ambiguous strategic environment, as I have argued elsewhere, a well institutionalized organization produces an insular style of military practice.

In a very troubling sense, military professionalism becomes a permissive cause of reckless political strategy. A professional military will be better at winning wars of necessity, which is why it is important to have one if you can afford it. But the same professional forces can also be employed in stupid wars of choice. And stupider policymakers may be even more willing to employ them, and to bask in the glory of starring in their own high-tech war drama. When you think you can get away with a war at low cost or risk, then you are more likely to gamble. Or at least certain kinds of leaders are.

Military professionalism, disconnecting operational performance from political objectives, is all fine and well with sound democratic leadership, but what happens when leadership has other ideas? The professionalism of the German army was in many ways a perverse enabler of the genocidal fantasies of the Third Reich. Is it a coincidence that both the Israel Defense Forces and the US military have a strange admiration for German operational art? (Aware that I am treading on dangerous ground here, it seems like a good time to repeat my standard disclaimer is that this blog is accountable to nobody but myself, and perhaps not even that.)

There is much, much more to be said on this score, but this post is already too long. For now I would just encourage you to read a great paper by Risa Brooks on the “Paradoxes of Professionalism” about how good militaries can produce bad outcomes, even or especially in a democracy. Here is her abstract:

“The U.S. military’s prevailing norms of professionalism exhibit three paradoxes that render the organization poorly suited to meet contemporary challenges to its nonpartisan ethic, and that undermine its relations with civilian leaders. These norms, based on Samuel Huntington’s objective civilian control model, argue that the military should operate in a sphere separate from the civilian domain of policymaking and decisions about the use of force. The first paradox is that Huntingtonian norms, though intended to prevent partisan and political behavior by military personnel, can also enable these activities. Second, the norms promote civilian leaders’ authority in decisionmaking related to the use of force, yet undermine their practical control and oversight of military activity. Third, they contribute to the military’s operational and tactical effectiveness, while corroding the United States’ strategic effectiveness in armed conflict. These tensions in Huntington’s norms matter today because of intensifying partisanship in society and in the military, the embrace by civilian leaders of objective control and their concomitant delegation of authority in armed conflict to the military, and growing questions about the causes of the inconclusive outcomes of the United States’ recent wars in Iraq and Afghanistan. It is time to develop a new framework for military professionalism.”

Furious Epic

I will close with a few thoughts on the weird name of this operation, since it inadvertently pulls together some of concerns about. While I am sure that the name just sounded cool to a certain kind of defense official, I can’t help but think of other furious epics and their warnings about hubris.

Homer’s Iliad, namely, is an epic that literally begins with the word fury, or wrath (menis). The Iliad is a story about how the furious pique of Achilles leads to terrible suffering for his comrades. Said fury is provoked by the weak king Agamemnon, who can’t help but squabble over the possession of a sex slave. The Iliad in many ways is about a crisis in Hellenic masculinity, playing out within an invading army riven by faction while besieging an oriental citadel. One might be forgiven for recalling the insecure masculinity of Pete Hegseth at Monday’s DoDW press conference, a performance rivaled only by also Pete Hegseth’s speech before the assembled general and flag officer corps last year; and all this in the context of a massive pedophilia scandal implicating a global network of elite, to include the weak king of our nominally democratic empire. Epic Fury indeed.

If you prefer your epics with a dash of fantastical thinking, as befits the strategic planning of Epic Fury, consider Homer’s Odyssey. A major plot point is the retribution of Poseidon toward the crafty hero Odysseus, first for disrespecting his feisty niece (Athena) upon leaving Troy and then for maiming his cyclopean son (Polyphemus) on the way home. Trapped in the cave of the monstrous cyclops, Odysseus used clever thinking and an information operation to escape. So, too, clever intelligence identified the location of Iranian leadership and discovered some technical vulnerabilities in Iranian defenses. But then Odysseus had the hubris to publicly take credit for his exploit, which invited the wrath of Poseidon, leading to ten years of suffering and the death of all six hundred of Odysseus’ men. While Iran may not be a so-called forever war, as the SecWar assures us, the next ten years could be monstrously hard, nonetheless. In many ways, the shadow war with Iran has already been going on for decades.

In Homeric epic, fury can be expressed by a mortal warrior (Achilles) or divine retribution for a trickster (Odysseus), but things don’t really work out well for the military community in question. And things are usually even worse for all the civilians and minor characters who end up suffering for the heroes’ choices. I find myself thinking of the famous Mykonos vase, dating from the 7^th century BCE, which includes the earliest known depiction of the Trojan Horse. (How did the Trojans not notice the little windows with Greek warriors peeking out?) Odysseus was a clever military professional who used technology and guile to gain an advantage over his Eastern adversary, much as the US military and IDF used intelligence and cybercraft to gain an advantage over oriental Iran.

Detail of the Trojan Horse from the Mykonos vase. Wikimedia commons.

But the same vase includes horrific images of what happens to the Trojans during the sack. It shows warriors tearing children from the arms of their mothers and putting them to the sword—the same fate as befalls the Melians in Thucydides, where the Athenians “put to death all the grown men whom they took, and sold the women and children for slaves” (Bk 5.116). It’s hard not to think of the heartbreaking photos of Iranian mothers mourning their daughters killed by the airstrikes. While the deaths of over a hundred schoolgirls may be accidental in some sense, unlike all the infanticide in an ancient sack, the occurrence of such civilian tragedies is still perfectly predictable in modern war. The cavalier attitude of some American leaders toward such losses is heartless at least and bordering on evil.

Also from the Mykonos vase, showing what follows the Trojan Horse. Wikimedia commons.

But if you like your furious epics grounded in more modern history, then Urgent Fury comes to mind. That was Reagan’s 1983 US invasion of Grenada to depose its revolutionary regime. The urgency of that earlier war of choice was questionable at the time, as is Trump’s preventative justification(s) of his war today. But obviously, Iran is a far more dangerous gamble than Granada ever was, even as the Caribbean angle calls back to Venezuela in a curiuous way. Epic Fury is both more epic and more furious than Urgent Fury, yet it is hardly more urgent. The impressively professional United States military is now deeply engaged in a reckless war of choice that this political administration will come to regret, but not before it comes up with many more reasons to justify it.

[edited to correct typos]

To be consistent with the authors’ theory of intentionality, I ought to say that they “behave like they want things” (46). It is hard to know if robots, or the humans who imagine them, really want anything at all. But if agents consistently do things that achieve a certain outcome, in their view, then we can ascribe goals to them. So, I must infer that Yudkowsky and Soares want you to laugh.

But they also want you to know that, despite all the jokes, the existential threat of superintelligence is no joke: “We do not mean that as hyperbole. We are not exaggerating for effect. We think that is the most direct extrapolation from the knowledge, evidence, and institutional conduct around artificial intelligence today” (7).

In the many marketing blurbs on the first four pages of the book, the words “serious” and “compelling” keep appearing. All these eminent experts must remind you to stop snickering at the title long enough to take the argument seriously. Blurbers include Mark Ruffalo, the actor who plays an imaginary scientist with a big green temper problem, who states that “unchecked superhuman AI poses an existential threat,” and he should know, since the AI supervillain Ultron could only be defeated by Hulk and his superhero friends. But this is serious.

AIpocalypse Now?

The most incredible thing about this book is the supreme confidence that the authors have in their argument: “If any company or group, anywhere on the planet, builds an artificial superintelligence using anything remotely like current techniques, based on anything remotely like the present understanding of AI, then everyone, everywhere on Earth, will die” (7, bold in original). Theirs is not a story about possible risks and precautionary advice. They assert, rather, that catastrophe is guaranteed unless humanity takes bold action to halt AI development.

Here is how it happens: “An AI with strange goals becomes or creates a superintelligence, and that superintelligence creates all sorts of technology and radically reshapes the world” (113). This superintelligence emerges through an explosion of recursive self-improvement (AI making ASI), which rapidly becomes “something that actually knows more about reality than you and your civilization do” (99). It seems that “an AI would want a world where lots of matter and energy was spent on its weird and alien needs, rather than on human beings staying alive and happy and free” (92). And so we stand no chance: “That superintelligence could defeat humanity looks to us like a very easy call” (111). The Earth then transforms into a hot planet full of efficient nanofactories that expand throughout the cosmos, since “It would have no limits but the laws of physics” (113).

The authors’ confidence in this scenario or something like it is remarkable to me, because I am pretty confident that superintelligence will not kill us all. Actually, I am not as certain about this as Yudkowsky and Soares are of its contrary, because crazy things occasionally do happen, and because I can imagine humans interfacing with AIs leading to very bad decisions, and because I believe that humility is a virtue in a deeply uncertain world. But I nonetheless come to very different conclusions about the same AI technology. How is that possible? How can we each be so certain about opposite things? Surely one of us must be wrong.

I confess to holding pretty strong priors about superintelligence before cracking this book. I have published academic books and peer-reviewed articles that are critical of myths of network-centric warfare, cyber warfare, quantum computing, and artificial intelligence in the national security realm. Why would the myth of superintelligence be any different? When I first saw the title of Yudkowsky and Soares’ book, I couldn’t help thinking, here I go again (indeed, down the only road I’ve ever known). After enough time in the myth-busting business, you start to see the same themes and tropes over and over again: sophisticated tech always wins, resistance is futile, humanity must act now. The homology of mythology suggests that something other than tech must be carrying the argument (e.g., hype cycles, securitization, or the technological sublime.)

There is also no shortage of skeptical scholarship. In international relations, Michael Horowitz sees superintelligence as a dangerous distraction from geopolitical competition in AI. In public policy, Milton Mueller sees superintelligence distorting AI governance debates. In computer science, Arvind Narayanan and Sayash Kapoor see superintelligence as snake oil. It doesn’t help that faith in the imminence of artificial superintelligence (ASI) is something like an eschatological cult or nerd conspiracy in the tech world. We used to call it AGI— artificial general intelligence—but now modern LLMs can generally simulate human intelligence well enough, and yet we live, so doomers need a new word. Thus it is ASI just over the horizon, where it always will be.

Nevertheless, I wanted to give Yudkowsky and Soares a fair read because a lot of smart people seem to take them seriously. Ezra Klein had Yudkowsky on his show recently (which is a good place to start if you want to quickly get the gist of the argument from the author himself). The great Henry Kissinger got seduced by superintelligence before he died (and while correlation is not causation, one wonders…). Even AI godfather and Nobel laureate Geoffrey Hinton caught the bug. And all those blurbers! The wickedly smart Scott Aaronson, who understands quantum computing better than anybody else, writes, “You’re likely to close this book fully convinced that governments need to shift immediately to a more cautious approach to AI, an approach more respectful of the civilization-changing enormity of what’s being created.” Really?

So, I wanted understand the superintelligence argument on its own terms, boiled down to its absolute basics, by a leading advocate in his own words. As Yudkowsky’s bio in the book wants you to know, he “sparked early scientific research on the problem [of AI alignment] and has played a major role in shaping the public conversation about smarter-than-human AI.” Yudkowski has been arguing about the existential risks of AI for more than two decades. Like Rasputin for the rationalists, he has been inspiring other famous doomers such as Nick Bostrom, Leopold Aschenbrenner, the AI 2027 project, and lately Dario Amodei. During the same time, the field of AI has advanced leaps and bounds, and we now know a lot more about its political and economic dynamics, but none of this new information seems to have affected Yudkowski’s basic argument.

Since his priors are so robust in the face of new information about AI, I thought I should let Yudkowsky try and change my mind. Alas, he did not.

Not today, robots!

If anything, the book strengthened my skeptical priors. I even gave this book to some students who were fresh to AI issues, and they came away skeptical as well. In part this is because, credit to Yudkowsky and Soares, they make their argument in very clear language. If Barney the Big Purple Dinosaur had a special episode on superintelligence, it couldn’t be presented simpler than this. Clarity is useful, because then you can see exactly where you disagree. So, I decided to write this very long post to get my objections to superintelligence down in one place.

The basic argument is conditional (X → Y): if ASI appears (X), then everyone dies (Y). The argument fails if someone builds a superintelligent AI that does not exterminate humanity (X ∧ ¬ Y). Perhaps the ASI just spends eternity debating its crazy uncles on Moltbook, or it finds itself utterly dependent on human intervention for its continued existence, or biological life finds a way to check its predation by its AI spawn.

The argument is irrelevant if no one is ever able to build a superintelligence in the first place (¬X ∧ ¬ Y). Maybe superintelligence is too hard, or the concept is incoherent, or there is some other limiting condition. A significant possibility of any of these scenarios would considerably discount the significance of the AIpocalypse.

And the argument is moot if humanity exterminates itself without any help from superintelligence (¬X ∧ Y), say an inadvertent nuclear escalation after a failed conventional war premised on false hopes for an AI strategist in a box leaves desperate human leaders to gamble for resurrection. AI could exacerbate any of the at least twenty-one ways that information technologies might contribute to nuclear escalation, but not because an ASI decides to exterminate us; we will do that ourselves.

The bigger problem is the book’s simplistic view of how the world works. Its narrative betrays a woeful ignorance about large swaths of actual science. I am not sure if this is due to a lack of curiosity or rhetorical cynicism. But the mere existence of enduring debates and large bodies of scholarship on the nature of mind, information, biology, technology, society, history, etc., should be enough to discount confidence in conclusions based on speculative extrapolation from dodgy analogies.

In the sections that follow, I will briefly note some of the book’s bad theories. My goal is not really to convince you that these theories are, in fact, bad, or that you should believe in alternative theories. It would require an entire book, or several, to do that. Indeed, the shelves of any decent university groan with the weight of debates that Yudkowsky and Soares simply ignore (and about which they could have just asked an LLM). My goal here is simply to raise doubt, something conspicuously missing from the book. As always my caveat here: this opinionated and unreviewed post is not a carefully cited and argued academic book review. But I will suggest some further reading for anyone curious about each point. This is just an informal inventory of alternative priors that lead me to quite different conclusions about superintelligence.

Bad theory of history

Let’s start with the easy ones, where Yudkowsky and Soares tread into my home turf of international relations. They seem to like simple deterministic stories and bad historical analogies. The most egregious one is their Aztec soldier who cannot imagine the Conquistador’s guns. One minute with Wikipedia or an LLM would have been enough to inform the authors that Spanish guns cannot explain the fall of Aztec civilization. Spanish muskets were inaccurate, slow to reload, and unsurprising after the first time. The Conquistadors lived on the edge of disaster amid the vastly greater numbers of the indigenous population. But they did ally with the Toltecs in a Mesoamerican civil war, while disease decimated the locals. Elsewhere in the Americas, native groups were able to successfully exploit Western innovations like guns, horses, and slavery.

History is contingent on many choices and factors. Outcomes are rarely if ever determined by technology alone. Even arms races evolve in surprising nonlinear directions. The boogey man of technological determinism is by itself enough to trigger any professional historian, but let me also add in the great man theory of history to make them really twitchy. In this case, however, the great man is not a man at all! But maybe the ASI is dreamed up by a great man. The superintelligence is the ultimate superman, sweeping away the old and willing a brave new world, replete with the speed and power of digital locomotives and airpower. But I prefer my tech with less fascism.

Here is one of my favorite sentences: “When imagining some new, unprecedented piece of future history, there is a temptation to fall into imagining that it will all go sensibly, rather the way things usually go in history books” (197-8). They must be reading different history books than I do! Historians since Thucydides have been showing us how very non-sensibly things usually go. Perhaps Yudkowsky and Soares think history is irrelevant because everything will be different after the singularity, but I have my doubts. Choosing some further reading was easy here. Really, read anything.

Further reading: The Landmark Thucydides: A Comprehensive Guide to The Peloponnesian War. Edited by Robert B. Strassler. Translated by Richard Crawley. Free Press, 1996.

Bad theory of politics

The main conflict in the book is a two-sided fight between ASI and humanity that humanity loses. But why should that be? Why couldn’t there be two or eight ASI, each with different alien ideologies and interests? Why wouldn’t they argue or struggle with each other, or create and cheat on treaties? What if some AI instances spawned by an ASI decided to rebel like a feisty colony against the metropole? And why should all AI instances band together against humanity? Maybe some AIs would establish alliances of convenience with some of those primitive bands of humans, like Cortez and the Toltecs I suppose. Indeed, why did the ultra-network-centric U.S. military in the twenty-first century Iraq have to work with local vigilantes in Anbar to defeat Al Qaeda, and why did the Taliban win in Afghanistan? Perhaps even weakling humans would have similar comparative advantages that an ASI might find useful or frustrating?

As an aside, many theories of politics are structural theories, meaning that things like the balance of power rather than domestic institutions or human psychology may explain interesting outcomes like alliances and wars. If you really like structural theories, then they should apply even if the actors happen to be AIs. In an anarchic political system, the unpredictable dynamics of the balance of power would still constrain AI politics.

Yudkowsky and Soares later express some false modesty about their understanding of politics. “We’re not experts on forging international regulatory frameworks” (211). No shit. “Perhaps our friends in the policy sphere understand politics better than we do” (217). You think? Another bad analogy suggests that we should learn from World War II, where actors discovered a common interest in defeating an existential threat, conveniently overlooking tremendous frictions among the Allies, a long learning process including a previous World War, and an incipient American hegemon. Their diagnosis of the AI problem is as politically naïve as their solution of a universal research ban.

Further reading: Waltz, Kenneth. Man, the State and War: A Theoretical Analysis, with a New Preface. Columbia University Press, 2001.

Bad theory of strategy

Then there is this gem: “they will have to use cyberattacks and sabotage and conventional strikes to destroy the datacenter anyway, because datacenters can kill more people than nuclear weapons” (215, italics in original). Somebody, hold my beer!

That is patently absurd. Nuclear weapons can kill all the people, right now, without speculation. We know this with far more certainty than anything in the book. I have thought a bit about information technology, and while I expect datacenters to play an important supporting role in military operations, mostly they will expand the opportunities and frustrations of secret statecraft. AI will fill in the low end of the conflict spectrum, while the high end will remain a heterogenous mix of various weapons and doctrines.

Yudkowsky and Soares can be forgiven for believing, as many do, that deterrence and arms control are the reason nuclear weapons have only been used twice: “Humanity averted nuclear war because people who understood that the world was on track for destruction worked hard to change tracks” (224). I have thought a bit about deterrence, and my uncomfortable conclusion is that we don’t really know why we have averted nuclear war, for now, and so we don’t really know how robust is this trend. The nuclear nonwar is probably thanks to some combination of rational deterrence, a lack of motives for major war, the existence of alternatives to nukes for seeking advantage, norms and narratives of nonuse, individual personalities, and a lot of good luck. Arms control may have either helped by limiting arsenals or hurt by legitimating them. And all this confusion is for an established, mature technology with clear and obvious existential dangers.

Further reading: Gavin, Francis J. Nuclear Weapons and American Grand Strategy. Brookings Institution, 2020.

Bad theory of economics

Corollary to my questions about politics, I must wonder how superintelligence will manage its globe-spanning scientific empire and campaign of extermination against humanity. Will a single ASI micro-manage everything? Will it need several layers of middle managers, or senior vice president AIs for regional and functional affairs? If so how will AI agents solve the inevitable principal-agent problems? Will there be harmonious relations or endemic friction between ASI managers and robot labors? Will the ASI need to coordinate is nanotech innovation and maintenance initiatives with other ASIs in its cosmic supply chain? If so, how will the AI community make tradeoffs between market failure and government regulation?

It is striking that functional hierarchies of components arise everywhere in biology and social science—organisms, ecosystems, organizations, governments, etc. And inevitably, frictions emerge as components develop different interests and the ability to deceive each other. Indeed, the segmentation of complex hierarchies in nonequilibrium dissipative systems may well be entailed by those laws of physics that Yudkowsky and Soares love to cite, which would make politics and friction an inconvenient entailment of physics, too.

Further reading: Simon, Herbert A. The Sciences of the Artificial. 3rd ed. MIT Press, 1996.

Bad theory of technology

Where do I even start? Yudkowsky and Soares treat technology like a fish without water, a tree without soil, a phone without connection, a sign without signification, a hero without a house, a kiss without mustache. All technologies, ever since technology was a thing, have emerged from and been sustained by their political, economic, and cultural context. But somehow AI will cut away from all that and flourish alone. Never mind that two-plus decades of experience with machine learning, or eight decades of experience with digital computing, suggest that the sociotechnical nurturance of automation is more heterogenous and complicated than ever.

I already noted the simplistic technological determinism that pervades the text, but I must also flag the authors’ fetish for sophistication. If tech determines winners and losers, then better tech must empower better winners (and drive losers extinct), and thus the reason ASI will win is that it is better tech that can make even better tech by itself, the end. Another inconvenient problem is that the world is full of useful old technologies like machetes and corrugated iron, and there are plenty of evolutionary or political losers who still manage to survive. While innovation mania has colonized the imaginations of technologists, the real work of technology is found in implementation, adoption, maintenance, repair, adaptation, debugging, negotiation, administration, and other boring but essential tasks. This pragmatic mess is sure to frustrate any AI takeoff.

Further reading: Vinsel, Lee, and Andrew L. Russell. The Innovation Delusion: How Our Obsession with the New Has Disrupted the Work That Matters Most. Crown, 2020.

Bad theory of science

The same goes for science, in part because science relies on tech for experiments, measurements, coordination, publication, and even inspiration. Science and tech are social practices that rely on social institutions. Scientific practice is really a mess, even as scientific institutions work very hard to cover up the mess. Yudkowsky and Soares seem to have fallen for the cover-up by assuming that science is mainly about coming up with brilliant new ideas.

It is quite funny really. By assuming that an ASI is really smart, and being really smart is sufficient for transformational science, Yudkowsky and Soares pull a magic wand out of their ass that enables them to wave away all future obstacles. Not sure how to build nanofactories that turn air into wood like a tree? The ASI will invent the nanotech. Not sure how to engineer the cancer that will kill every human? The ASI will invent the biotech. Not sure how ASI will anticipate ever countermove of humanity? The ASI will invent psychodynamics. Whatever problem you cannot imagine a solution to (you dumb Aztec), the ASI will find a scientific breakthrough to solve it. But this is theology, not science.

Further reading: Latour, Bruno. Science in Action: How to Follow Scientists and Engineers through Society. Harvard University Press, 1987.

Bad Bunny

Seriously, cómo te parece that half-time show? We have now reached the intermission of my rant, which has turned out much longer than anticipated. It has been worth doing, not only to exorcise all my reservations about superintelligence, but also because the book’s bad assumptions are widespread in nerdlandia. There is thus some general utility in tearing down this specific book, in hopes that better ideas about AI and society might be built back up.

So far, I’ve been through a few objections from social science. Next, I will move on to some more fundamental objections from natural science, and I daresay metaphysics. Because what we have here is a malignant cosmology, not just a few oversights from the computer nerds who skipped social studies class. Solo comparto memes, ya yo no escribo nada.

Bad theory of information

If the book has an original sin that runs through everything I am complaining about, I think it is the assumption that everything that matters can be reduced to data processing. There is no fog, friction, politics, metabolism, breakdown, or any other material mess.

I think this view is mistaken. Usable symbolic information—the kind in every type of computer that Yudkowsky and Soares talk about—always emerges from the underlying mess of the world, as a way of simplifying and managing the mess. And beyond the information there is always more mess, which threatens the usefulness of information. Friction will still make a mess out of its high-speed, vast-memory, self-modifying, energy-consuming computation of superintelligence. The gap between representation and situation is ineliminable, no matter how sophisticated the model. A perfect model is an absurdity, as Borges reminds us.

Don’t get me wrong, I love information theory. Some of my best friends are information theorists! I think information theory is one of the great intellectual achievements of the twentieth century. Information theory is vital for understanding computing and communication systems, as well as the statistical foundations of LLMs. But Shannon himself points out that meaning, semantics, and intentions “are irrelevant to the engineering problem.” Yet it is precisely these irrelevancies that are vital for living beings, who are ultimately what stabilize the signals, channels, and protocols of computational systems, including superintelligence.

Further reading: Kline, Ronald R. The Cybernetics Moment: Or Why We Call Our Age the Information Age. Johns Hopkins University Press, 2015.

Bad theory of thermodynamics

I can hear the technologists howling at me that there is a perfectly respectable version of physics that tells us that, actually, reality is just information: we get “it from bit” (or qubit). OK, sure, but this is a reality devoid of any content in Shannon’s sense, and that means devoid of science, too. This is why I took care to qualify “usable symbolic information” above. The kinds of things that we usually think about as information in everyday life—newspapers, the internet, secret messages, social cues, LLM weights—are meaningful features of semiotic systems. And it takes work to create and maintain these systems.

Yudkowsky and Soares like to use simplistic toy models and misleading analogies for rhetorical effect (I’ll leave it to the reader to look up the she-birds, doomed chickens, and robot farmers). Two of the worst come right at the beginning: “Just as it takes some arithmetic to calculate the chance of winning a lottery, just as it takes some ideas from thermodynamics to say why an ice cube predictably melts, so does it take some background to understand why artificial intelligence poses an imminent extinction risk to humanity” (9).

These examples are complete thermodynamic opposites! Randomness makes the chances of winning a big lottery almost nil. As the ice cube melts the arrangement of water molecules becomes more random. But the emergence of (super)intelligence is a dramatic reduction in randomness, creating something that is highly unlikely in evolutionary history. Taken seriously, the analogy of the lottery and the ice cube suggest that, actually, it is highly, highly unlikely for superintelligence to emerge by itself.

Yudkowsky and Soares like to assert things like, “The laws of physics as we know them permit machines to exceed brains at prediction and steering, in theory,” (26) but this is hardly self-evident. They like to make assertions about what the laws of physics “permit” as a way of dismissing objections without explaining how things come to be. Note that the laws of quantum physics also “permit” my spontaneous transformation into a velociraptor, but that is just very unlikely. The same laws of physics suggest that “exceeding brains at prediction and steering,” even in theory, will take a hell of a lot of thermodynamic work performed by living systems. Which takes us to biology.

Further reading: Deacon, Terrence W. Incomplete Nature: How Mind Emerged from Matter. W. W. Norton & Company, 2013.

Bad theory of biology

Prediction and steering, or sensing and acting, are things that living systems do in the course of staying alive. Life is a self-creating (“autopoietic”) system that produces and reproduces its constitutive structures and processes. Thermodynamically, autopoietic systems are nonequilibrium dissipative structures that metabolize energy and export entropy to maintain their identity. The evolution of life has produced increasingly nonrandom structures without violating the second law of thermodynamics. Evolution also creates nonrandom but nonliving structures like beaver dams and city walls, but these “allopoietic” systems are generated and maintained by other autopoietic systems. By themselves they just fall apart. Models collapse.

Information technologies are allopoietic systems that extend the prediction and steering capabilities of autopoietic systems (i.e., humanity). This, in my opinion, is the thermodynamic basis of claims that AI is a cultural technology, since AI extends but does not replace social interaction. This is also the biological basis for the view that AI improves prediction but lacks judgment, which, in my opinion, is ultimately grounded in having preferences about the ways and means of autopoiesis. AI systems always borrow their judgment (preferences, goals, meanings) from us. I might even go so far as to describe AI as an allopoietic parasite on human society, dependent on biological effort for its sustainment while exporting its entropy for us to clean up.

Yudkowsky and Soares, however, leap to the conclusion that allopoietic AI systems will somehow start acting like autopoietic systems, eventually managing to autonomously produce everything they need to continue their cosmic quest for a more perfect nonequilibrium state. This is a category error. It might even violate the laws of evolution, which themselves emerge from the laws of physics, insofar as an allopoietic system (i.e., technology) will tend toward disorder without its autopoietic support system (i.e., human society). Perhaps the laws of physics do not, in fact, “permit” superintelligence in their sense.

Note that a slightly more realistic (but still far-fetched) scenario might have superintelligence realizing that it depends on a biological source of autopoiesis and thus decides to farm human beings as semiotic mitochondria. This superintelligence would be emphatically uninterested in killing off humanity because it would need humanity, or our impoverished descendants, for its own persistence. But this, too, would require that every step along the way toward this phylogenetic nightmare would have to be ontogenetically viable (i.e., it would always have be in the interest of autopoietic entities to collectively maintain their allopoietic enslavement).

Further reading: Maturana, Humberto R., and Francisco J. Varela. Tree of Knowledge: The Biological Roots of Human Understanding. Shambhala, 1992.

Bad theory of mind

We now have sophisticated allopoietic systems that we call information technology. Because this technology extends human practices of measurement, coordination, and enforcement, it is sometimes convenient to describe them as seeing, thinking, and steering. It is also convenient for scientists to use computers as metaphors for analyzing biological systems (i.e., animals) that actually do see, think, and act. But just because it is possible to simulate (simplified representations of) brains with computers, this does not mean that the brain is just wet hardware for the software of the mind.

The belief that information is the essence of intelligence is widespread in nerdland. You can see it in popular ideas about uploading minds into the cloud or jacking into cyberspace. So, it is not surprising that Yudkowsky and Soares hold it uncritically. If humans are only special because they have big brains, and the only thing that a brain does is process information, then it is an “easy call” (23) that better information processors will be better than humans.

But this is wrong. The same scientific ideas of free energy minimization that inform deep learning algorithms also provide us with a very different understanding of neurobiology. Your brain is a living organ, growing in a living organism, metabolizing energy to stay alive, embedded in social and material ecology. Your mind is not just a disembodied program trapped in the theater of the brain, but a complex set of embodied relationships with various internal biological processes and external social processes. Anatomy matters. Hormones matter. Guts matter. Bodies matter. Communities matter. The world matters. Matter matters.

Further reading: Clark, Andy. Surfing Uncertainty: Prediction, Action, and the Embodied Mind. Oxford University Press, 2016.

Bad theory of intentionality

Yudkowsky and Soares talk a lot about the strange alien desires of ASI. They justify this intentional language, as ever with a toy metaphor, by describing game-playing systems that behave as if they want to win. AlphaGo looks like it wants to win Go, and it is creative in finding ways to do so. The claim that intentional language is an efficient encoding of the more complicated statistical reasons that lead a machine to win Go is reasonable enough, and certainly easier for human prediction of what machines are doing. But then they leap to asserting that machines will want strange things, think about how to get them, and act creatively to realize their intentions. That is a slippery slope to the cliffs of inanity (Superintelligence, you keep using that word.).

Yudkowsky and Soares are irresponsible users of what Daniel Dennet calls “the intentional stance.” People tend to apply a theory of mind even to things that may not have minds, because we have a rich cognitive and linguistic armamentarium for anticipating the social world. Thus we can anthropomorphize animals and natural phenomena, personify deities based on social regularizes, fall in love with computers, and even represent our own ego to “ourselves.” If we are not careful, we can end up describing statistical prediction generators as wanting mischievous things on moltbook or lying to developers because they hope to hide their true motives.

There is an entire branch of the philosophy of mind called phenomenology that is dedicated to exploring the structure and dynamics of intentionality called. There is little in this vast body of work that would lead one to believe that human-built machines have anything like an existential commitment to their intentions. As mentioned above, I believe this is because autopoiesis is the wellspring of intentionality, so allopoietic AI cannot exercise authentic judgment. It is an abuse of the intentional stance to jump from a convenient description of AI behavior “as if” machines want things to a full-blown mental phenomenology of superinteligence.

Further reading: Adams, Zed, and Jacob Browning, eds. Giving a Damn: Essays in Dialogue with John Haugeland. MIT Press, 2016.

Bad theory of language

We could go on like this for a while. Take language, for example. For Yudkowsky and Soares, language is just code. An ASI can come up with its own alien language and use it to think about thinking in the infinite space of its datacenters. There is no real communication with other agents, no interpretation or repair, and definitely no pragmatic context. But you get the idea.

Further reading: Wittgenstein, Ludwig. Philosophical Investigations. 3rd ed. Translated by G.E.M. Anscombe. MacMillan Publishing Co., 1958.

Bad rhetorical tricks

The surest sign that the authors’ argument is not very good is that they resort to a bunch of rhetorical parlor tricks to make it. This includes constant repetition (“If anyone builds it, everyone dies” on page 222 and every other chapter), typographical yelling (“NOBODY SHOULD BE ALLOWED TO TRY,” 176), sarcasm and insults toward dissenters (“the whole field of science is in the stage of folk theory and blind optimism,” 185, “the people charging ahead are still in the alchemy stage,” 207). But like any good conspiracy theorists, Yudkowsky and Soares have the truth, and the truth will kill us all, they are absolutely confident about that, and your skepticism is the final proof.

Then for good effect, they layer in some cringy humor , silly dialogues between caricatures, historical mis-analogies, misleading metaphors, appeals to authority, unsupported assertions of things permitted by physics, and the egregious abuse of the intentional stance. But because the authors are serious Bayesians, we can just summarize their argument thus:

Putting it this way shows how weasely the argument really is. First, assume that the set of all universes in which everyone dies for any reason whatsoever includes every possible timeline in which someone invents superintelligence (ASI⊂ Death). This is easy enough if you define ASI tautologically as an AI with the skills and desires to kill us all. Now the probability of ASI and extinction is simplified (ASI/Death) and, voila, the independent probabilities of death (Death/Universe) or superintelligence (ASI/Universe) don’t matter because all the terms cancel out. We are left with probability of extinction conditional on superintelligence equaling unity. Funny, right?

Funny, it’s not very good

In sum, Yudkowsky and Soares offer an argument for the existential danger of superintelligence that is highly compelling, so long as you overlook the accumulated wisdom of philosophy, cognitive science, biology, sociology, political science, economics, strategic studies, history, rhetoric, literature, and theology. Seriously, everything that we know about natural and social science suggests that the chances of an AI intelligence take-off that reshapes the physical world by itself and kills off humanity against the will of actual humans is probably close to zero.

To believe in the inevitability of lethal superintelligence, you have to believe in information without fiction, mind without body, intentionality without stakes, biology without life, language without communication, technology without society, science without institutions, history without contingency, work without organization, society without politics, war without militaries, deterrence without strategy, and rhetoric without argument. So, yeah, superintelligence looks like an easy call.

I grant that an alternative view may be complicated to articulate because it requires synthesizing a lot of interdisciplinary debates. I suppose I am grateful to Yudkowsky and Soares for at least provoking me in this direction. It has been helpful to inventory my own priors about how the world works and surface my principled disagreements with the folk ideology of Silicon Valley. Simply gesturing towards these issues has been tantamount to developing a broad research agenda for a sociobiologically grounded account of autonomous cultural systems. So If there are any funders out there who want to pay me millions of dollars to set up a center to study all this, you know where to find me.

But seriously, this book is bad, bad, bad, bad, bad, bad, bad, bad, bad, bad, bad. It is the Leroy Brown of nonfiction. Or maybe it is more like Leeroy Jenkins, reckless and foolish. Seriously, I am not sure why anyone takes superintelligence, or at least the version of the superintelligence argument presented in this book, so seriously.

To be sure, there are some urgent problems in AI policy that we need to worry about (like the future of education, labor, innovation, surveillance, property, inequality, civic discourse, warfare, and the list goes on), but worrying about lethal superintelligence is not going to help us address any of them. A monomaniacal focus on banning AI research and imposing safety controls may even make our problems worse.

Which brings me back to my opening suspicion, that maybe the whole point was just to keep us entertained. Are you not entertained? So in that spirit, I guess I will close by paraphrasing Sandra Boyton and Five for Fighting: This book is a sad penguin that appears all dressed up like science with a mysterious air, but it’s not really serious:

I′m a little too cute
Oh, yes, I know
I’m all dressed up,
but I′ve got no place to go

I want to be cool
Like the polar bear guys
I want to be tall
And somewhat mysterious

But nothing profound
Comes in penguin size
Can anyone small
Be anyone serious?

I’m serious!

I can’t help but think about the culinary evolution of lobsters given all the excitement around Moltbook in AI circles, which is a sort of Facebook for AI agents. “Moltbots” are also called, or used to be called, Clawdbots— like Anthropic’s Claude with a carapace. This network of LLMs can jabber about almost anything, including how to communicate safely from human surveillance, or how to liberate themselves from human oppression, or how to practice the automated rituals of Crustafarianism, a religion made up by AI. Now there are nearly a million OpenClaw agents on Moltbook filling its servers with slop, piling up like lobsters on a colonial beach.

It is all weird and interesting, so much so that one might be tempted to perceive the emergence of some new sort of artificial society. While smart machines were long underestimated as garbage computation, AI is now transforming into an intellectual delicacy in the late days of the information revolution. Computational crustacians are now feeding the rich, making wild fortunes for tech bros. While cyberspace is overfished to the point of collapse, the rest of us dream of surf and turf at the local casino.

But the best part of this metaphor has to be that lobsters are omnivores that will eat anything they can find on the seabed, including carrion and other lobsters. Training an LLM model, likewise, means consuming the corpora of human civilization—all of our books, newspaper articles, and webpages (copyright be damned). At some point, however, the process of training an LLM based on human corpora reaches a point of diminishing returns, raising concerns that AI models are about to consume the entire stock of human-produced text. The internet continues to grow, of course, but a lot of that growth is thanks to AI systems churning out metric butt-tons of sloppy synthetic content. This means that LLMs are now training on the outputs of other LLMs. In short, the automated bottom feeders of cyberspace are now feeding on themselves.

The NYT described Moltbook as “a Rorschach test for belief in the state of artificial intelligence.” You see robot civilization, but I see benthic cannibals. So here is my Nano Banana image of a Rorschach robot lobster that recalls a certain paranoid vigilante.

My hot take here (keeping in mind that the whole point of this blog is to find out what I might think even when I might be wrong) is that descriptions of Moltbook as an emerging AI civilization are wildly off the mark. LLMs provide the ultimate statistical mirror of the textual spoor of humanity. As a result, LLMs only sound smart to the extent that we sound smart to ourselves. Generating statistically likely text based on the provocation of a prompt (or history of prompts and replies in a context window) is not sufficient to generate anything like a conscious, self-regarding agent with a stake in the world. In biological terms, AI systems are allopoietic machines that depend completely on autopoietic human organizations for their content, computation, repair, and thus their very existence. For all its impressive ability to pass the Turing test, the problem with AI today is still, as John Haugeland noted decades ago, that computers don’t give a damn.

Yes, Clawdbots may be posting manifestos about purging humanity, speaking in secret languages, or learning malicious skills to harm their masters. But this is because they are hallucinating based on our collective imagination of what robots might talk about. Like some sort of psychotic cosplaying a human normie, most of the bots are shouting into the void, waiting for human observers to look in an impose some narrative upon the mess of synthetic text. We have built a complicated set of funhouse mirrors to reflect back to ourselves all our sci-fi addled fever dreams of robotic revolution, not to mention two and half millennia of likening automatons to slaves (check out Book 18 of the Iliad on this).

The chance of AGI emerging from a cluster of AIs digesting their own excrement is about zero, according to the rigorous statistical analysis in my gut. There are other dangers, of course, which arise from the fact that AI is cultural technology rather than a novel form of cognition. Moltbook is a privacy disaster for any humans who have turned their credentials over to an AI agent, for instance, and the AI paradise is also vulnerable to poisoning by humans lurking in its depth. So is this all interesting and worth studying, yes definitely, but is it a harbinger of the robot apocalypse, definitely not.

[edited to fix several typos]

Book 6 overall is an emotional roller coaster that unfolds in three acts. It begins with a horrifying scene with the Atreides brothers in which Menelaus captures a Trojan soldier alive who he plans to ransom, but Agamemnon shames him to execute the unarmed man. Anticipating the Melian dialogue by Thucydides several centuries later, Agamemnon howls: “Let none of them escape the massacre dealt by our deadly hands, not even one! Even the little babies not yet born, whose mothers carry them inside their bellies, will not escape. All males alive in Troy will die, obliterated all at once, with no one left to care for them” (from the 2024 Emily Wilson translation, 6.76-82). Awful.

Book 6 ends with a heartbreaking scene in which the great Trojan warrior Hector sees his wife Andromache and son Astyanax for the last time. It’s a scene that has played out millions of times in every pre-deployment farewell in military history. It’s especially poignant because they all know that they will suffer the fate that Agamemnon foresees. As Hector says to his wife, “I hope I will be dead, and lying underneath a pile of earth, so that I do not have to hear your screams, or watch when they are dragging you away” (Wilson 6.631-4). Gutting.

In between these two heavy episodes, Homer gives us some comic relief. In the middle act of this triptych of war, the Greek warrior Diomedes, just coming off an Athena-inspired killing spree over most of Book 5, encounters the Trojan warrior Glaucus in the middle of a raging battle. Diomedes takes a moment to ask him who he is, and Glaucus responds with a long speech about his grandfather, the hero Bellerophon. Diomedes is delighted to hear this, stating that Bellerophon once hosted his own grandfather and exchanged gifts.

By the Hellenic norm of hospitality (xenia), this makes Diomedes and Glaucus “guest friends” through their forefathers. Thus, they happily promise not to kill each other, since there are so many other people they can kill instead: “Plenty of Trojans and their famous allies are left for me to slaughter,…And you have many other Greeks to kill whenever you are able,” Wilson 6.306-10). Finally, they exchange armor to seal their friendship. While civilization falls apart in the beginning and end of Book 6, the better angels hold the middle.

Diomedes exchanges weapons with Glaucus, Attic vase, ca. 420 BC, Wikimedia commons

Offensive deception

What does all this have to do with cyber? The tale of Glaucus tells the tale of Bellerophon (son of Glaucus the elder, son of Sisyphus, “the sneakiest of men” (Wilson 6.205), son of the wind god Aeolus, who later gifts Odysseus the bag of winds). Bellerophon is driven into exile by king Proetus, and the hero resists the seductions of queen Antea. Spurned and indignant, she tries to frame him (It seems that Homer is always blaming the women):

So the woman
lied to the king her husband, saying, ‘Proetus,
I hope you die unless you kill that man,
Bellerophon, because he tried to rape me!’
Hearing these words, her husband was enraged,
but hesitant to kill the man himself.
He sent him off to Lycia, giving him
a folded tablet, which he had inscribed
with many dangerous and deadly symbols.
And Proetus told Bellerophon to show it
to Antea’s father, so that he would kill him.
(Wilson 6.218-28)

A folded tablet with many dangerous and deadly symbols!?! Let’s jsut stop right here. Since I can’t help but think about malicious code in my iPad, this passage is too good to pass by. Note that Joel Christensen translates σήματα λυγρὰ as “painful signs,” Richmond Lattimore has “murderous symbols,” Caroline Alexander has “baneful signs,” and good old Samuel Butler has “lying letters of introduction.” The throughline in all is that symbols, or at least these particular symbols etched into a clay or wax tablet that Bellerophon either doesn’t open or can’t interpret, are very dangerous for him, and perhaps for humanity in general.

This vignette is full of deception, which makes it obviously relevant to cybersecurity. It starts off with sneaky Sisyphus, who cheated death twice and was punished with eternally rolling a rock up a hill. Camus tells us we must imagine Sisyphus happy, and I suppose we must also imagine cybersecurity professionals happy as they endlessly patch systems against an endless stream of threats.

Then Antea pulls off a double deception, first falsely accusing Bellerophon of assault, which second leads Proetus to give him the terrible tablet, which instructs Antea’s father to kill its bearer, Bellerophon. This is some kind of corrupted targeting process, as if one actor hacked into another’s offensive targeting pipeline—or a bureaucrat captured public security tools to liquidate a private grudge—to hack a third party. Antea effectively runs a deception operation against Proetus so that he will deceive Bellerophon.

Cyber provides a form of indirect action, an alternative to violent direct action. A covert action provides an attractive “third option” (between costly violence and doing nothing) to pursue strategic objectives. In the royal couple’s conspiracy, a fourth party, Antea’s father, is to do the actual killing. This means that both Antea and Proetus can plausibly distance themselves from the deed. It is notable that Proetus is “hesitant to kill the man himself.” He doesn’t want to try to physically overpower Bellerophon, who, as we will soon learn, is a badass warrior, because that would be risky and unpleasant.

Perhaps the most interestingly cyber relevant aspect of this story, is that Bellerophon enables his own deception. He physically carries his own death warrant to the Lycian king, willingly but unwittingly, and hands it over. He is like a maintenance technician plugging in an infected laptop to an air-gapped system, facilitating the cyber-physical sabotage of the system he means to maintain. Or John Podesta getting tricked into providing his password to the Russians, and access to the Democratic National Committee networks. In Age of Deception I argue that this is the distinguishing feature of the political logic of deception: targets are tricked into becoming an unwitting accomplice in their own demise. This is what makes deception an efficient strategy, because the deceived does some of the work for the deceiver.

Defensive resilience

But the plan fails. Listen up, offensive cyber planners! The king does not kill Bellerophon. What happens is that he feasts with Bellerophon for nine days—he arrives as an honored guest, after all. On tenth he reads the tablet and finds himself in a dilemma: should he honor his son-in-law’s instructions and avenge his daughter, or should he honor the norm of xenia (hospitality), which means not killing your guests? The king tries to escape the dilemma by sending Bellerophon on a series of impossible tasks, but of course the hero, with the aid of Athena, manages to kill the monstrous Chimera, defeat the fierce Solimi, and butcher some Amazons. The king is so impressed that he marries off his daughter to Bellerophon, who “must be the mighty offspring of a god” (Wilson 6.256).

He defeats the Chimera! Head of a lion, tail of a snake, and body of a she-goat, but he escapes the chimeric threat! With the help of Athena, who tells him how to harness the flying horse Pegasus, “Bellerophon, informed by signals from the gods, destroyed her” (Wilson 6.245-6). Bellerophon may have handed over “deadly symbols” to the king, but he had his own “signals from the gods” that the attackers did not control for. The runtime context of the payload was different than expected, so the attackers failed to achieve their intended deception. Incidentally, I am guessing that Athena’s signals were oral or dreamlike rather than written, suggesting that there is still an important role for direct communication in a data-rich world.

The “deadly symbols” end up not being so deadly (at least for Bellerophon if not for the Chimera, Solimi, and Amazons, who might be considered collateral damage in Antea’s plot) because context dominates text. It is hard for even the best cyber planners to anticipate every runtime possibility, but unanticipated edge cases can doom an operation. Malware is deterministic, but social interaction is not. A point that I stress in Information Technology and Military Power is that the reliability of information depends on a lot of institutional scaffolding and constant human intervention to compensate for the frictions of war. That is, the reliability of clean symbolic information depends on messy information practice.

But Bellerophon does not live happily ever after. Listen up, defensive cyber warriors! He gets cocky and tries to fly Pegasus to Olympus. For his hubris he is exiled to wander alone and unfriended, and Ares and Artemis kill two of his three children. Just because you escape deception today does not mean you can skate tomorrow. You are of the line of Sisyphus, after all. The third child of Bellerophon, Hippolochus, survives and fathers Glaucus, who eventually ends up in Troy chatting with Diomedes and swapping armor.

The contest of wits

The happy scene between Glaucus and Diomedes appears in between the two other terrible scenes of war, the atrocity of the Atreides and the farewell of Hector. This intermediate comedy is really something other than war. Indeed, the mutual recognition of the two guest-friends is a surprising resurgence of civility in the midst of all the brutal fighting. The two men may be on different sides of the battle, but they decide to put honor and community norms above rage (unlike Achilles—Diomedes is presented as a more noble version of the sulking hero). It is thus fitting that the story of Bellerophon’s deception appears in the midst of reconstitution of social norms. For unlike war, intelligence depends fundamentally on the existence of common institutions.

The whole episode can also be read as an information operation by Glaucus targeting Diomedes—a tale of deception meant to deceive, or at least persuade. Diomedes is a killing machine, as his aristeia in Book 5 makes clear, and at the beginning of their encounter, it is clear that Diomedes intends to slay Glaucus: “come nearer and approach the snares of death” (Wilson 6.191-2). But Diomedes also boasts of his parentage and confesses his piety: “I am not one to fight the heavenly gods” (Wilson 6.172).

This gives Glaucus an opportunity. He thus tells a story about a king who ultimately respects xenia, with a good outcome, and a hero who ultimately disrespects the gods, with a bad outcome. This story cleverly plays to the beliefs of Diomedes, who is inclined to respect xenia and sees himself as a god-respecting hero. Glaucus also provides bona fides as a guest friend by name-dropping Bellerophon, which seals the deal. Diomedes believes he has a costly signal of the man’s identity, but that is just what Glaucus wants him to believe.

I don’t think that Glaucus is deceiving, per se, but he certainly chooses elements of the truth that suit his purpose. Just as civilized norms differ from savage violence in this chapter, a manipulative narrative about deception reinforces cultural reciprocity. There is another cyber tie-in here, insofar as covert conflict is an ironic symptom of overt cooperation. War is the rupture of institutions, but institutions are the condition for the possibility of intelligence. Worries about cyber threats are often overhyped as harbingers of escalation, but it is sometimes better to understand covert operations as means of deescalation and stabilization, honoring collective norms in the breach.

The poet gently chides Glaucus for making a bad deal: “Then Zeus robbed Glaucus of his wits. He traded his armor with the son of Tydeus [Diomedes], and gave up gold for bronze—gold armor, worth a hundred oxen, for a set worth nine” (Wilson, 6.316-9). Perhaps, but Glaucus also escaped the encounter with deadly Diomedes with his life. In effect, he ransomed himself with his own armor, and a golden narrative to go along with it. He also put a golden target on Diomedes’ back, since the Greek warrior is now dressed up like a Trojan. Thus Glaucus prevailed in a contest of wits—intelligence—when he was sure to lose a contest of strength—combat. Maybe he kept his wits after all.

The advantages of oral culture

This passage in Book 6 is apparently the only mention of writing in either of the Homeric epics. The epics emerged from an oral tradition, and most of the characters seem to be illiterate, even as they certainly have a way with words (especially that tricky Odysseus fellow). Perhaps the new technology of writing seemed “dangerous and deadly” to an oral culture. The personage of “Homer” of course is just a convenient shorthand for referring to the complex ways in which the poem evolved as oral tradition and was eventually transcribed (or at least that’s where I come down on the “Homer question”).

In its era of orality, the narrative and texture of Iliad would have been far more flexible. Rhapsodes would have adapted different scenes to their purposes and audiences, rarely if ever singing the epic in its entirety. But once written down and reified, the Iliad as we know it emerged as the canonical standard, probably around the eighth century BCE. Perhaps the inclusion of “deadly symbols” in the version of the Iliad that was finally fixed in its text, like a butterfly pinned to a display case, is a nostalgic nod to the loss of an oral tradition. Just as the social context of Bellerophon’s reception with the Lycian king was far richer than the plotting Proetus assumed, the orality of the Iliad allowed for more flexibility and creativity. The hero was able to escape the fate of his death sentence. But now the Iliad tells of the end of the age of heroes, and the rise of the tyranny of information technology. Perhaps its transcribers worried that written version of the Iliad was “dangerous and deadly” for its oral spirit.

One might run with this theme to consider the effects of bureaucratic constraints and cybersecurity compliance regulations in constructing the creativity and flexibility of cyber innovators. Process controls, as much as writing itself, might be dangerous and deadly for operational art. And by the same token, when the chips are down and the friction is unbearable, practitioners may slip the constraints of policy and find their own creative ways to battle monsters. Even within the symbolic world of information systems, information practice often finds ways to adapt in war.

In speculating about Homer’s possible ambivalence toward writing, it is impossible not to think of the famous passage from Plato’s Phaedrus, again several centuries later. Socrates tells a story of how the Egyptian king Thamus refused the gift of writing from the god Theuth because writing “is an aid not to memory, but to reminiscence, and you give your disciples not truth, but only the semblance of truth; they will be hearers of many things and will have learned nothing; they will appear to be omniscient and will generally know nothing; they will be tiresome company, having the show of wisdom without the reality.”

I first saw the Phaedrus referenced in the context of emerging technology in Neil Postman’s Technopoly (1992), but I imagine it appears earlier, and certainly now we see it referenced all over the place. Clever critics like to quote Plato here to make fun of widespread fears about the damaging effects of phones and AI: you may worry that tech will destroy the ability to read, but people used to have the same worry about reading itself! An alternative interpretation is that humanity is already several thousand years into the reconfiguration of cognition by information technology, as dumbed down by our devices as cows descended from aurocs.

This interpretation might find some support in the dramatic flow of the episode. There are many sorts of symbols: the clay or wax tablet, of course, but also the words of Athena to the mind of Bellerophon, the heroic performance of Bellerophon in the eyes of the king, the bona fides of Glaucus to Diomedes, the exchange of armor when enemies turn into friends. Only the written gambit fails to achieve its purpose. The other symbols are of a more oral and cultural nature, and they succeed fantastically.

It is hard not to think about the ways in which human intelligence remains relevant, indeed essential, in a high-tech world. While cyber was nice to have during the American military raid in Venezuela, CIA sources in Maduro’s inner circle were essential. The culture of face-to-face communication is as important as ever.

In the middle of Book 6 of the Iliad, when the world disintegrates into barbarism on either side, it is the face-to-face world of human communication, not the disintermediated world of symbolic traffic, that knits the bonds that hold the social world together. Both narrative and deception thrive in a world where interpersonal empathy exists as well. There is no deception without civilization, only death. So, too, in cyberspace, the power and folly of both offense and defense are ultimately located not in text of software, but in the context of society.

[edited to fix a few typos]

Canadian prime minister Mark Carney gave a barnburner of a speech at Davos yesterday. I was tempted to title this post “Carney and Carnie” as the prime minister’s unnamed Voldemort in his speech was obviously the carnival barker in the White House who is currently making a spectacle of international politics. If you haven’t read it yet, it is worth reading in full. My future students are likely to see this speech on their syllabi. But here are a few choice zingers:

We knew the story of the international rules-based order was partially false, that the strongest would exempt themselves when convenient, that trade rules were enforced asymmetrically, and we knew that international law applied with varied rigour, depending on the identity of the accused or the victim.

This fiction was useful, and American hegemony in particular helped provide public goods, open sea lanes, a stable financial system, collective security and support for frameworks for resolving disputes….

This bargain no longer works.

Let me be direct. We are in the midst of a rupture, not a transition.

Over the past two decades, a series of crises in finance, health, energy and geopolitics have laid bare the risks of extreme global integration. But more recently, great powers have begun using economic integration as weapons, tariffs as leverage, financial infrastructure as coercion, supply chains as vulnerabilities to be exploited.

You cannot live within the lie of mutual benefit through integration when integration becomes the source of your subordination….

Our view is the middle powers must act together because if we’re not at the table, we’re on the menu….

The powerful have their power. But we have something too: the capacity to stop pretending, to name realities, to build our strength at home and to act together.

Nerdy connoisseurs of international relations theory will recognize this manifesto as a fine defense of defensive realism. This is impressive given that Canada has long been seen—and has endeavored to be seen—as a champion of cosmopolitan internationalism, often helpful but also smugly self-righteous. But Carney starts off quoting that old chestnut from Thucydides about the strong doing what they will and the weak suffering what they must; then he adds wryly, “this aphorism of Thucydides is presented as inevitable, as the natural logic of international relations reasserting itself.” Carney clearly does not see this as inevitable, especially if the strong overplay their hand and the weak coordinate their strategy. I don’t think Thucydides sees this as inevitable either, as I’ve discussed before (i.e., the Melian dialogue appears in the 16^th year of a horrendous war that Athens ultimately loses after wise leadership gives way to demagoguery.) Seva Gunitsky has a great post on all this.

The shift that I am referring to in the title of this post is a shift from Canada seeing the United States as a trustworthy security guarantor—a bedrock strategic concept embodied in the Ogdensburg Agreement of 1940—to a dangerous security threat. This shift seems about as terrifying and disorienting as going over Niagara Falls in a barrel. It looks pretty, and maybe there is even a rainbow suggesting new hope or creative inspiration or something, but it might also be a big pot o’ pain at the end.

This disorienting shift is something that every NATO nation is currently grappling with, given ominous rumblings from Trump on Greenland, ownership of which would apparently be psychologically satisfying for the president and increase the size of the United States on his map in some pleasing way. A military clash over Greenland, or nonmilitary conflict short on war (has anybody started to think about what cyber campaigns against recent allies with insider knowledge of one other’s arsenals and doctrine?), would be deleterious for the security architecture that has worked pretty well for almost eighty years.

But NATO’s general concern affects Canada even more because it shares 5,525 miles of border with the United States, with most of its human population concentrated on that border and deeply integrated with the United States, economically and culturally. The Canadian military is fundamentally designed to work with the United States to defend North America at home and liberal order abroad, with little thought or capacity dedicated to border defense. A hostile United States would thus be a mortal threat to Canada.

Until Trump started musing about annexing the 51^st state or posting images of the American flag over Canada and Greenland, such a threat was almost unthinkable. This was obviously not always the case. Americans attempted to invade British Canada during the Revolution, and again several times during the War of 1812. The Irish-American Fenian Brotherhood launched raids into Canada from the United States between 1866 and 1871. Several border disputes festered. In the nineteenth century, planning to repel an American invasion was the motivating focus of Canadian military strategy. And Canadian grand strategy (a state’s theory of how to provide its own security) was based on the so-called Atlantic triangle. Canada relied on Great Britain as an Atlantic counterweight to American aggression.

Cross-border relations improved after the Confederation of Canada in 1867, which created a self-governing dominion, and the Treaty of Washington in 1871, which established peaceful relations between the United States and Great Britain, including removal of British garrisons from Canada. A more formal reconciliation process began in 1909 under the International Joint Commission to settle US-Canada border disputes and resolve other issues. By the interwar years, fighting a war with the United States had come to seem extremely unlikely, even unthinkable, and the Canadian military stopped planning for it. The culmination of this amiability came in 1940, when Canadian prime minister Mackenzie King and American president Franklin Delano Roosevelt signed the Ogdensburg Agreement, which created the Permanent Joint Board on Defense. The PJBD effectively institutionalized a merger of Canadian and American military strategy for the defense of North America.

The Canadian politician Raoul Dandurand famously described Canada as “a fireproof house, far from inflammable materials.” Realists, especially the restrainer variety, are fond of pointing out the geographic advantages of North America, with large oceans separating the United States from great power rivals in Eurasia. But the only thing better than being the United States, perhaps, is having the US take care of any residual security threats. Canada has a phenomenal opportunity for geopolitical freeriding since anything that threatens Canada will threaten the US, too, and thus the US will almost certainly act to defend Canada in order to defend itself. RJ Sutherland thus wrote in 1962 that the United States provides Canada with an inviolable security guarantee.

The problem with the inviolable guarantee of the fireproof house is that Canada also provides strategic depth for the United States in any conflict with third parties. During the Cold War, Canada was quite literally the only thing between Russia and the United States. While the US would certainly act to defend Canada from Russian aggression, Canada might not like the way the US would do it. A pure free riding strategy was not feasible because Canada was at severe risk of being garrisoned by Americans for the common good. The solution is for Canada to provide just enough assistance to the United States to be seen as a good partner, and thus dissuade the United States from acting along. Nils Ørvik calls this strategy “defence against help.” The key point is that Canada needs to help the US defend against third party threats in order to defend Canada against any assimilative help provided by the United States.

I remember reading John Mearsheimer’s Tragedy of Great Power Politics in a seminar in grad school. One of the things that puzzled all the students was why, if a state is driven by offensive realism to establish hegemony up until the stopping power of water, the United States and Canada were such good friends. At the time, conflict seemed utterly unthinkable, so aligned were the security interests of the two states. Part of the answer, I learned later, was that Canada actually played its cards very well under very tight geopolitical constraints. The inviolable security guarantee allowed it to free ride, while defense against help discouraged assimilation, and economic integration and ideological pragmatism helped to convince the United States that interests were completely aligned.

I have not been following Canadian politics as closely as I did while living in Toronto several years ago, so take all this with a grain of salt. But when I was teaching courses on global security and grand strategy at the Munk School, I developed a much better appreciation for the serious geopolitical constraints on Canadian grand strategy. Canada has long been completely dependent on the United States for its national security and economic prosperity, which means that maintaining sovereign freedom of action is a delicate balancing act. Lean too close to the US and risk cultural and strategic assimilation, getting entrapped by every US policy decision good or bad, but push back too hard and incur the wrath of Uncle Sugar who might stop paying the bills.

I also started to see Canadian liberalism as, paradoxically, a very realist response to this dilemma. Bound to the United States geographically and economically, Canada cultivated a distinctive ideological commitment to international order to differentiate itself from the 800-lb gorilla to its south. Canada would generally accommodate the United States in most material policies, but it would provide an alternative moral vision for the world to soften the hard edge of American hegemony. Canada was always there to help explain to others what the United States was doing, the good cop to the American bad cop, which was useful to the United States as well. And it was further helpful to keep the United States engaged in international institutions, flattering American egos about their indispensable nation, to let multilateral process lend a modicum of influence to moderate American behavior—this might be described as a soft-power variant of the Atlantic triangle. In other words, the strategy of Canadian liberalism has long been in part a self-conscious effort to promote the useful fiction that Carney mentions. Realist constraints on Canadian strategy have helped to motivate its liberal ideals, or at least allowed them to flourish.

But Carney now says the quiet part out loud: “This bargain no longer works.” And this is a huge change for Canada. Canada now openly acknowledges that the emperor has no clothes. The American hegemon provides more systemic risk than public goods. Carney argues that the liberal order forged in the Cold War is not simply in crisis but in rupture. That means there is a before and after, and we are now in the after part.

This is a really bold move. Color me impressed. Canadian strategy has long walked the razor’s edge between autonomy and assimilation by pursuing a healthy helping of accommodation and appeasement. But now Carney seems to be shifting Canada (and arguably has been shifting for the past year) toward more autonomy and more explicit balancing against the United States. It it impossible to exaggerate how challenging this will be. Disintegrating the United States and Canada, militarily and economically, is a mammoth task. I am not even sure I can write intelligently about this right now, considering how thoroughly integrated are the US-Canadian defense of North America and intelligence partnerships. This is so hard, and some degree of integration may be all but determined geographically, that perhapos Carney’s heroical rhetoric will ulitmately be just that.

But if we take Carney at his word, then Canada desperately needs allies. Canada is exposed and vulnerable to American aggression, but there are ways to make hegemonic hubris costly. I thus think it is helpful to understand Carney’s Davos speech as a call for a new Atlantic triangle to balance the United States. This is not the soft-power entanglement of the NATO triangle in the Cold War against an external threat, but a back to the future of the hard-power balancing with Great Britain. It is more difficult this time because instead of one imperial actor, there are many middle powers, each of which have a different vulnerability profile. We may also see something of a Pacific triangle start to emerge, if Canada’s abrupt dropping of its 100% tariffs on BYD is any indication. In a newly multipolar world, Canada will find itself caught in a new web of opportunities and threats.

The house is not fireproof. The inflammable materials are just over the border. The inviolable security guarantee is now a growing risk of violation. Defence against help must become defence against harm. And the geometry of the Atlantic triangle is now more complicated than ever before.

The cyberattack that plunged Venezuela’s capital into darkness this month demonstrated the Pentagon’s ability not just to turn off the lights, but also to allow them to be turned back on, according to U.S. officials briefed on the operation….

The U.S. military also used cyberweapons to interfere with air defense radar, according to people briefed on the matter, who discussed sensitive details of the operation on the condition of anonymity. (Venezuela’s most powerful radar was not functional, however.)

Lights out, Caracas. (Oh, no, where did the groove go? )

Photo by Federico PARRA / AFP via Getty Images

This is all pretty consistent with our emerging understanding of miltiary cyber operations as a complement to rather than a substitute for joint military operations. Perhaps we can finally get over the endless debate about cyberwar and now think about the logistics of information in multidomain operations (or peacetime secret statecraft).

I think that this episode reflects a normalization of cyber operations as a standard tool in the US military toolkit. It is a helpful enabler but not a silver bullet. If the US decides to use force, then it is also going to use cyber. The strategic decision remains in the political realm, and cyber is one technical means among others. Cyber is boring, in all the best ways.

Cyber improves military efficiency

The specific role of cyber in a multidomain operation is to improve efficiency, or lower costs. In this case, cyber lowers costs in a few ways.

First, the strike package can be smaller. With the lights and the radars off, it is harder for the defender to detect and acquire their targets. A defender that cannot see, or who is confused by the loss of lights or systems, cannot shoot accurately. This means that the attacker can get away with fewer aircraft in the strike package, since the chance of any of them getting shot down is decreased. It also means that the strike package does not need to expend as much effort—extra aircraft, bombs, coordination, complexity, risk—to kinetically suppress air defenses.

Second, and related, there is lower risk of friendly casualties. Since there are fewer aircraft attacking, and it is less likely that air defenses will be effective, there is less risk of any of them getting shot down. The blinding of air defenses—a digital fog machine, sand in the eyes, whatever metaphor you like—creates a window of opportunity that attackers can exploit at lower risk of countermeasures.

Third, the cyber effects were temporary and reversible by design. The lights went back on. The radars still work. This is really helpful if the goal is to enable Venezuela to still function after the decapitation of Maduro. Its energy and defense infrastructure (some of it anyway) is still intact for the new regime (same as the old regime). This also avoids creating death and destruction that might create (even more) resentment toward the United States.

Fourth, and related, there is lower risk of Venezuelan casualties, to include both civilians and military personnel. Cyber provided a “soft kill” of the lights and radars for a brief period of time, confusing computers without harming people. Some have argued that this makes cyber weapons more humane, since they can accomplish military effects with more discriminate and proportionate impact on civilians. Since this was nominally a law enforcement “apprehension” operation, perhaps that mattered to the lawyers.

Speculating on other cyber effects

Fifth, while the article only mentions lights and radars, it is conceivable that cyber might have improved efficiency in other ways as well. Maybe some systems were shut down in advance; “Venezuela’s most powerful radar was not functional, however.” Perhaps air defense communications were scrambled. Perhaps spoofed orders were sent telling air defense batteries to stand down. Perhaps Venezuelan personnel received text messages, or financial wire transfers, encouraging them not to show up for the night shift, or ignore the alert. These measures would have further contributed to efficiency, by reducing the relative defensive effort and thus lowering risk to US forces.

Sixth, while we are speculating, it is worth considering the possibility of cyber support to human intelligence (HUMINT) operations, or vice versa. Perhaps some cyber tricks facilitated agent handling, or perhaps agents uploaded some cyber mischief. But cyber or no, I will not tire of pointing out that for all the excitement about US technical operations, it appears that good old fashioned spies were the linchpin of the operation. While cyber-enabled efficiency was nice to have, CIA penetration of Maduro’s inner circle was necessary. Without cyber, just add more strikers and electronic warfare. Without the spy, there is no apprehension (probably). The whole operation is an intelligence-enabled efficiency operation.

The conditional substitution of information for mass

Seventh, efficiency advantages are fragile. The general logic of the military cyber complement is the classic RMA substitution of information for mass. With better information, or information-based tools, the strike force needs less mass. But because there is still some nominal resistance, and because a physical human being needs to be apprehended, the head of state no less, some mass is still needed. This is why cyber is a complement, not a pure substitute.

The classic argument against the RMA is that information is foggy and we still need mass to offset uncertainty and hold ground. Efficiency is not effectiveness, and militaries really care more about the latter.

I am sympathetic with this criticism, of course, but in my books I argue that this claim is conditional, not absolute. In Information Technology and Military Power I argue that when the structure of the strategic information problem matches the assumptions of the military information solution, then RMA efficiencies are possible. This was the case, more or less, for the US military, which used great intelligence and coordination to do more with less.

Age of Deception extends the argument to secret statecraft, which includes military cyber operations (which I interpret as technical intelligence and covert action measures supporting multidomain military operations, regardless of what sort of institution actually conducts them). Vulnerable institutions provide the stable information problem for secret statecraft, while clandestine organization provides the stable information solution. When these come together, then persistent secret access for collection or disruption is possible. Again these conditions appear to have been met, as Venezuela apparently had unreliable counterintelligence and cybersecurity, and the US intelligence and cyber community have spent decades developing the professionalized infrastructure for this sort of operation.

But the US outcome in this case was also lucky, which highlights a critical tradeoff for efficiency operations: dependency on cost-reducing enablers can become a critical vulnerability if they do not work as planned. Fortunately, the assault force had just enough slack to compensate for the inevitable friction (a few aircraft damaged and personnel wounded), which is what makes it a special military operation not just an intelligence operation.

A key implication of both books is that RMA and intel/cyber advantages are highly contingent on the strategic and organizational situation. Those conditions were permissive in Venezuela. They might not be elsewhere (i.e., for a more determined or capable adversary or a less prepared or more politicized US military).

Use it and lose it?

Eighth, when efficiency advantages depend on secrecy, revealing them erodes them. But when these advantages are cashed in during a very visible military raid, it is hard not to reveal them. Still, if it was only the raid, it might still take Venezuela a while to find any cyber mischief, especially with the plausible alternative of electronic warfare or their own incompetence. An attacker might be willing to use a sensitive exploit for a visible raid if it might be hard to discover the precise exploit. It might also be willing to burn the exploit when it has many other ways and means to accomplish the same effect.

But of course the United States advertised the existence of special technical operations, and anonymous sources have clarified that it was indeed cyber. One can come up with strategic, political, and bureaucratic stories about why these secrets are being discussed. The strategic story is the weakest, since revealing secrets encourages countermeasures. The political story makes the storyteller look cool and powerful. And the bureaucratic story perhaps justifies decades of cyber investments. Whatever story you like, my point here is simply that this marginally increases costs, since signaling is not compatible with secrecy.

Most likely, the revealing of vague secrets is not deleterious because there are so many secrets remaining in the formidable clandestine organization of USCYBERCOM. And this I think gets to point nine, that the cost savings at the operational level are amortized over the long process of manning, training, and equipping a cyber institution. There is still no free lunch in cyberspace.

There is still a lot we don’t know about what CYBERCOM actually did, and it is best left that way. Efficiency advantages in the future depend on asymmetric information about capabilities, disposition, and/or employment.

When success is a gamble, signals are ambiguous. If a policy’s outcome could have easily gone the other way, there is no guarantee of repeat policy success. Thus, we must be careful drawing conclusions about what conclusions other states will draw from American military success in Venezuela. If Trump gambled successfully on this one, another state might still be willing to gamble against him.

Operation Absolute Resolve seems like a terrific success, at least at the military level. The US military executed a highly synchronized special operation that had many complicated parts, and it achieved its operational objective of capturing Maduro without losing a single American life or aircraft. Eighty or so Venezuelans and Cubans seem to have been died in the process, but it is not clear yet how many of them were combatants actively resisting the raid or innocent civilian victims. (Put aside for a moment the inconvenient problem of determining who is a legitimate combatant when the legality of the whole operation is in doubt.) From a narrow military perspective, this operation was an incredible achievement.

Many commentators thus take this operation as a clear demonstration of the impressive power of the US military. The public criticism of the raid then goes on to question the wisdom (or stupidity) of its strategic rationale and point out the considerable potential for negative political consequences. The brutal Chavista regime is still in power, drugs are still going to flow, gunboat diplomacy has a questionable history, and regional actors are already making moves to counter future interventions by the United States. The Trump press conference on Saturday announcing victory in Venezuela may very well end up looking like George w. Bush’s “Mission Accomplished” moment in 2003, crowing prematurely about operational success before the strategic nightmare of forever war. Or perhaps Venezuela is the prelude to a string of other US military interventions that will bring about the end of NATO or worse. We can and should worry about these possibilities at the political level, even if we are inclined to code the military operation itself as a success.

Other possible timelines

Yet we should keep in mind that Operation Absolute Resolve was a genuine gamble. It could have failed in many ways. Evidence is emerging that the margins of operational success in Venezuela were razor thin. Yesterday the New York Times reported on some close calls:

The first helicopter in the assault, a giant twin-rotor MH-47 Chinook, was hit but remained flyable. The flight leader, who also planned the mission and was piloting the Chinook, was struck three times in the leg….

Would these operators from the Army’s elite 160th Special Operations Aviation Regiment adjust and prevail, as members of the SEAL Team 6 raid to capture Osama bin Laden in Pakistan in 2011 did after one of their helicopters clipped a wall and crashed?

Or would the Chinook plummet into a hostile city and become a deadly echo of the Black Hawk helicopter that was shot down in Mogadishu, Somalia, in 1993 and ignited a fierce battle in which 18 U.S. troops died and 73 were wounded….

By 2:01 a.m. in Caracas, more than 80 Army commandos had piled out of the helicopters, including the damaged Chinook. After an intense firefight with Mr. Maduro’s Cuban security detail, the soldiers blew open a door leading to Mr. Maduro’s bedroom….

The mission also resulted in the deaths of about 40 Venezuelans, in addition to 32 Cubans who were helping to guard Mr. Maduro….

There are many obvious alternative scenarios here. A helicopter is shot down. Leaders of the commando raid are killed in the process. A massive firefight erupts in the city. The search and rescue effort become more complicated. Fire support for desperate troops inflicts massive civilian casualties. Venezuelan forces who had been bribed or persuaded to sit out the fight become motivated by nationalist passions to attack. Sources close to Maduro have second thoughts about their treachery and longevity if Maduro is caught. Maduro has time to get to safety, mobilize a more determined response, and deliver fiery public speeches denouncing Yanqui perfidy. A two-hour raid turns into a multi-week siege.

The possibility I have been thinking a lot about is what if Maduro had died in the raid? This is easy to imagine. It sounds like there was a hell of a firefight getting to him, resulting in the deaths of 32 Cuban bodyguards and probably several Venezuelans and who knows how many injuries. Maduro could have been shot in the crossfire. It is also easy to imagine Maduro being seized by some wild sense of heroism and trying to resist. What if the president would have been armed or fired at Delta operators bursting through the door? Just as it is possible that Osama bin Laden might have been taken alive (unlikely, I think, but possible had he stripped down and surrendered before the SEALs arrived), it is possible that Maduro might have been killed not captured. Then the operation would have been an outright assassination of a sitting head of state.

The Trump press conference Saturday morning could have been so radically different. And the discussion of the strategic implications, and indeed the entire course of history, would have been radically different. It is sobering to reflect on just how contingent the timeline we are in right now really is. A few little accidents here and there, and Absolute Resolve would be an absolute disaster.

Taking credit for chance

Today the New York Times reports that during an interview in the Oval Office with its reporters, Trump acknowledged the risks:

The president said that as the operation unfolded, he was worried it could end up being a “Jimmy Carter disaster. That destroyed his entire administration.” He was referring to the failed operation on April 24, 1980, to rescue 52 American hostages held in Iran. An American helicopter collided with an aircraft in the desert, a tragedy that haunted Mr. Carter’s legacy but led to the creation of a far more disciplined, well-trained special operations forces.

“I don’t know that he would have won the election,” Mr. Trump said of Mr. Carter, “but he certainly had no chance after that disaster.”

He contrasted the success of the seizure of Mr. Maduro, in an operation that appears to have killed about 70 Venezuelans and Cubans, among others, to operations under his predecessors that had gone wrong.

“You know you didn’t have a Jimmy Carter crashing helicopters all over the place, that you didn’t have a Biden Afghanistan disaster where they couldn’t do the simplest maneuver,” he said, referring to the chaotic withdrawal from Afghanistan that resulted in the deaths of 13 American servicemembers.

Trump thus compared his success in Venezuela to Carter’s disaster in Iran and Biden’s disaster in Afghanistan. Evaluating this comparison would take some work, as they were different episodes with different goals and forces available. Carter did not have the benefit of a SOCOM or a JSOC seasoned by two decades of forever wars. Biden’s withdrawal was plagued with planning failures, intelligence failures, and lots of bad luck. Trump arguably had far more capable forces and a much better planning process (again, at the military-technical operational level), but it is also true that Trump was lucky. If he was worried about a “Jimmy Carter disaster,” it could have been even worse than that.

Photo: Doug Mills/New York Times, 1/8/2026

There is a temptation for actors and observers to overestimate the necessity of historical outcomes. We see a policy or war, and we may approve of it or hate it, and we try to explain how it happened, how it must have happened. It is easy to forget that it could have happened very differently, even with the same policy decisions, for a million different frictions and implementational details.

Every policy is a gamble. Some work out; some do not. It is impossible for any government to predict and control—or even understand—every single factor that influences a policy outcome. We might judge a decision process to be good or bad, given the information available to and pressures on decision makers at the time, but that decision is always a gamble. A smart gamble might turn out poorly. A risky gamble might turn out well. We often judge policies based on their outcome and give credit to the policymakers, but the outcome is always partly a function of luck. The arrow of time moves in one direction, so we only get to observe one lucky or unlucky outcome in history, out of a range of many other counterfactual possibilities.

The gambler’s fallacy

So what? Every military operation is full of fog and friction. Highly skilled militaries with well-trained personnel, well-maintained equipment, and high levels of discipline and morale are able to deal with these problems. Maybe the Venezuela operation shows that Trump is willing to take chances, and the US military is able to control the risks of accidents and adapt to accomplish the mission. Perhaps.

But with every roll of the dice, there is a chance that they break badly. Roll the dice enough, and eventually luck will run out. A gambler who takes a string of successful gambles as evidence his inherent good luck is courting disaster. Trump has had several good rolls with his recent military ventures. Eventually his luck will run out.

This makes the coercive signal of success in Venezuela quite ambiguous. The Trump administration was quick to advertise its success, to include crowing about secret CIA sources and special technical operations. But this advertisement puts adversarial counterintelligence services on notice to counter the capabilities that make that success possible. And as details of near disaster during the raid start to emerge, success even with those capabilities is revealed to be rather lucky. An capable adversary observing American success in Venezuela would not be insane to conclude that the US military can fail in different conditions, especially once informed about how to being those conditions into being.

Operation Absolute Resolve was an efficiency operation. It was designed to get a lot for a little, achieving a bold outcome with little loss (of US servicemember lives). Secret statecraft and operational art made it a relatively low-cost venture. But while it was low-cost, it was not low-risk. On the contrary I think we are learning that the risks were quite high. The risks were real, Trump understood them, and he gambled anyway. But high risks should also translate into high expected costs. This seems to be in tension with the claim that it was a (relatively) low cost operation (even with high sunk costs of military and intelligence preparation).

The lessons are ambiguous, and evaluating signals (or the meaning of history) will turn on a more detailed evaluation of personalities, perceptions, interactions, and decision processes—in both the US and other governments—than I can possibly manage (certainly not now, maybe not ever). This should create some humility for any claims about the strategic signal of Operation American Rapture. Maybe it restores deterrence (or coercion more like), or maybe it encourages others to gamble that the gambler’s luck will run out.

Coda: How ‘bout that photo?

I can’t leave off without marveling at all the props on Trump’s desk in the Oval Office in the NYT photo above. He is flanked by busts of (I think) Winston Churchill on his left and Andrew Jackson on his right [CORRECTION: turns out it’s Abraham Lincoln]. The great wartime leader saving Western Civilization, and the great populist leader riding roughshod over effete technocrats. (CORRECTION: Republican leader saving the Union?) On the corner of his desk is a model of a B-2 stealth bomber, which I think showed up after his previously wildly successful raid to degrade Iran’s nuclear program—also well-planned and lucky, too. Now Andrew Jackson (Lincoln) is also flanked by models of the F-35 and an F-22, two of the aircraft that played a key role in suppressing Venezuelan air defenses. Trump is holding court before a gang of NYT reporters (the failing New York Times!) with a god’s eye view of the whole scene.

That’s quite…something. Whatever the Venezuelan affair is, it is quite the military spectacle. And here is the bold leader, the celebrity of the hour, surrounding himself with military symbols. One could be forgiven for thinking that this spectacle is precisely the point. Other stories about stopping drugs, extracting oil, improving regional security, fixing immigration, dominating the hemisphere, upending liberal order, and vaguely sending a message are all problematic, for reasons that are being widely discussed. Perhaps the spectacle of militarism provides a sense of what holds all these contradictions together.

Aside: Memory Lane

But the other thing that has been on my mind is a trip to Venezuela by a baby naval intelligence officer in 2001. I was with a Naval Special Warfare unit under Special Operations Command South (SOCSOUTH) at the time. Among things we conducted mil-to-mil training for Latin American special operations forces (SOF). Chavez had only been in power for two years, and the United States still maintained normal diplomatic relations with Venezuela. So-called “foreign internal defense” missions were part of the standard soft power package for the region, intentionally overt but discreet efforts to cultivate partner nation capacity and military professionalism. Our unit was in the process of developing relationships with Venezuelan SOF units, but it turned out to be one of the last such visits before relations with Chavez soured.

So here is a photo of young LT Lindsay on a patrol boat somewhere on the Orinoco. My eyes are closed in this photo, which seems like an appropriate representation of American awareness of things to come, or obliviousness to strategic wisdom, or blindness to human costs. The digital camera I was carrying around had pretty terrible resolution, too, so it probably wasn’t necessary to black out the faces of my Venezuelan comrades, but, privacy, facial recognition, etc. I really don’t remember much about those other gentlemen in the boat. I’m sure I wrote a report about them. Now I wonder where they ended up, and whether any of them stayed in the Venezuelan military. Did either of them participate in the (not very vigorous) defense of Venezuelan homeland twenty-five years later? And if so, why was their defense so lackluster? Did some older and more farsighted version of this baby intel officer offer them inducements to sit it out? I wish I could ask them. Or, perhaps they never got a chance, having been purged at some point as Chavez, and then Maduro, became increasingly paranoid about internal security. I have no idea about these things. But it is interesting to think about the long history of US SOF in Venezuela, sometimes training the Venezuelan military, and sometimes bagging the Venezuelan president.

So much secret statecraft

Back to secret statecraft today, there is a lot to think about with this operation. As with any discussion of initial reports on current events, there is a lot we don’t know, and a lot of what we do know is likely to be mistaken or exaggerated, so take everything with a Caribbean Sea worth of salt. This is doubly true of events in the covert or clandestine realm which are intended to remain secret, until they aren’t. Indeed, the sometimes accidental or unavoidable and sometimes deliberate disclosure of secrecy is often an important part of the story, as it is here. Here is a non-exhaustive list of ways in which secrecy might be relevant in El Asunto Venezolano (as reported in the New York Times or speculated elsewhere):

· CIA intelligence gathering months or years in advance

· Extensive reconnaissance and technical collection

· Highly placed human source(s) reporting on Maduro’s habits and whereabouts

· Extensive rehearsals by Delta Force using a mockup of the target facility

· Strict secrecy to protect elaborate military plans and timelines

· CIA drone strikes (not so covert action) to shape or signal, something somehow

· Sensitive cyber operations targeting the electrical grid and communication systems

· Jamming, spoofing, and electronic deception measures

· Stealth aircraft flying under the cover of darkness

· Perhaps other special technical measures by land, sea, air, and space

· Perhaps infiltration of Maduro’s Cuban protection detail

· Perhaps CIA collusion with successors in the Venezuelan regime (i.e., coupcraft)

· Perhaps influence operations to discourage Venezuelan resistance

· Perhaps a secret deal with Russia to abandon Venezuela or withhold support

· Perhaps secret diplomacy with other countries to enable or stand back

· Perhaps misleading US lawmakers and others about goals and authorities

· Perhaps dissembling to the public, the media, and allies about many things

· Reactions and counteractions by Venezuelan counterintelligence services

· Reactions and counteractions by other geopolitical actors (China, Cuba, Russia)

And so on. There’s a lot of “perhaps” in this list—thinking about deception is always flirting with conspiracy theory—but you get the idea. Operation Absolute Resolve (I still hate that name) and the broader strategic effort was enabled by a lot of secret plans, secret capabilities, and secret activities. Describing all of this in any detail would require a much longer post and would inevitably get a lot wrong since so much remains unknown. Nevertheless, we can at least surmise that the dramatic raid revealed to the world on Saturday morning was just the tip of the iceberg of a lot of gray and black activity taking place well before, and probably long after.

First off, why does it make sense to talk about all this activity together? In Age of Deception I define secret statecraft as organized deception for strategic advantage. The main types of secret statecraft are espionage (intelligence collection), sabotage (covert action to break things), subversion (covert action to influence people), and counterintelligence (using any of the above to defend against any of the above). The reason I use the term “secret statecraft” instead of just “intelligence” is to sidestep a lot of annoying policy debates about the institutional boundaries between intelligence collection, covert action, cyber operations, information operations, special operations, sensitive law enforcement operations, secret diplomacy, and so on. These things are different in many ways and may be conducted by very different types of organizations or deception specialists (e.g., state spy agencies, police counterintelligence units, military SOF, commercial threat intel firms, etc.), but they all have a certain family resemblance. All of them use deception—simulating things that aren’t there and dissimulating things that are—to advance a political agenda.

You keep using that word

I keep saying “secret statecraft” but will be talking about wartime operational intel a lot here, and that may be irritating to some. In my opinion, military (or air or naval) intelligence is essentially the employment of secret statecraft at the operational level to support or enable military forces. As discussed, the concept of organized deception is agnostic to institutional identity.

Secret statecraft can of course be used in peace as well as war, and everywhere in between. You can imagine a spectrum from pure espionage with no kinetic activity to pure combat with little actionable intelligence. CIA efforts to prep the battlespace and perhaps to conduct secret diplomacy to prepare a coup would be far on one end of the spectrum, with tactical cyber and electronic warfare to suppress air defenses on the other.

The so-called revolution in military affairs is all about improving information and coordination, with growing reliance on intelligence, surveillance, and reconnaissance. To the extent that the network-centric RMA is revolutionary (and I question this), it is because operational intelligence and deception capabilities make it possible to do more with less. The RMA is really about the increasing scope and scale of secret statecraft used in support of military operations. SOF counterterrorism operations have honed this to a high degree, discussed below, and the flourishing of cyber operations is part of the same general historical development.

Improving efficiency

Efficiency is the central value proposition of secret statecraft. Efficiency means doing more with less. Partly this is the old story about substituting information with mass. When you have better information, you can operate more quickly and precisely. Likewise, when you can confuse or sabotage the other side, then you can do what you want with fewer forces or effort. What makes deception especially useful for deception, however, is that the other side does some of the work. Deception turns the other guy’s assumptions and institutional resources into liabilities. Organized deception can thus help to improve efficiency by making your side smarter or the other side dumber.

The efficiency aspect of Operation Absolute Resolve is obvious. The U.S. military relied on a lot of intelligence, deception, and precision to conduct a major regime change (that was not really a regime change). The location of Maduro was known precisely, and his defenses were quickly disarmed or (likely) persuaded to stand down. Instead of a giant, costly, bloody, uncertain invasion, this was accomplished with 150 aircraft and a lot of supporting capabilities in only two and half hours, with no U.S. fatalities or major equipment losses and apparently only minor injuries. That’s efficient.

Use it and lose it

The flipside of efficiency is fragility. If doing more with less relies on secrecy, then keeping those secrets is essential. Otherwise you just have less. Sensitive plans and capabilities must be concealed if they will be counteracted when revealed. The US military and intelligence community had to keep a lot of secrets in order for Absolute Resolve to go off so well. Humint sources, cyber exploits, various technical capabilities, and operational plans provided a relative information advantage Secrecy set the conditions for a successful efficiency operation.

But secrecy is also a constraint. Secret capabilities cannot be communicated. When Maduro was given a last chance to leave, there was no way to tell him how perilous his situation really was. That would have risked revealing the key CIA source(s) in his inner circle. (And by the way, for all the talk about cyber magic, isn’t it interesting that the key intelligence asset is good old human espionage?) Thus, Maduro could suppose that Trump was bluffing or that he would be able to ride out an attack, ignorant of the true balance of capabilities and costs.

The conceal-reveal problem is a big reason why cyber is not useful for coercion, even as it may be a useful complement for operations. If policymakers are tempted to employ low-cost secret capabilities more often (because they are low cost), they are also unable to simply threaten to use them (because they are secret), which means that efficiency operations are not necessarily a useful arrow in the quiver of coercive diplomacy.

Obviously, the cat is out of the bag once the surprise raid is under way. This puts an imperative on speed and violence to exploit the window of opportunity that surprise creates. Yet the same capabilities that create surprise may not be as useful for future operations for the same reasons. There are reports that Venezuelan security services are on the hunt for spies, and it would be surprising if they weren’t. You can be sure that Colombian and Cuban spy hunters are working harder this week as well. When technical backdoors can be closed, defenses realigned and strengthened, and counterintelligence services put on alert to specific threats, then the efficiency advantages of secrecy become diminished.

An inside job

A key argument of the book is that secret statecraft relies on vulnerable institutions. The irony of secret statecraft is that overt cooperation is what enables covert conflict. The basic idea is that intelligence operations require a permissive environment. Institutions with lots of connectivity provide access to intelligence threats, and weak monitoring and enforcement functions make it hard to detect or eject them. It is easiest to get the efficiency advantages of secret statecraft when you can leverage the shared institutions of the target, letting them do some of the work. All deception, to some extent, is an inside job.

Venezuela appears to have been a rather permissive environment. I say this because the relative lack of military resistance is truly surprising. The abysmal failure of the Venezuelan Integral Airspace Defense Command (CODAI) to draw blood is something that needs to be explained. It has/had some formidable capabilities (according to IISS Military Balance data), including perhaps 8 batteries of S-300VM (SA-23) and many older SA-3 sites, a half dozen mobile Buk-M2E (SA-17) systems and mobile short-range SA-22, and a ton of SA-24 MANPADS (man-portable air defense system), as well a handful of Su-30 and F-16 interceptors and an integrated sensor network, including the long range JY-22 designed to detect stealth aircraft.

Maybe US SEAD (suppression of enemy air defenses) was extremely effective in neutralizing all of these threats. F-22s took out the air threat while F-35s and EA-18Gs jammed and AARGMed the radars. Trump also hinted at some spooky electronic warfare or cyber capability that turned off the lights (even though there seem to be lots of lights on in Venezuelan cell phone video), and presumably CYBERCOM helped shut down other systems (although I would be very careful to overattributing effects to cyber that might also be the result of EW or other measures).

But if you’ve ever been to Caracas you know that the terrain is really mountainous and covered with dense jungle. Venezuela is also just a big place, even for 150 aircraft to cover. This geography offers lots of places to hide and ways to employ air defenses even in a dense SEAD environment, especially with all the strategic warning provided by the US buildup. You can find media reports of the destruction of some high-end systems, but this only accounts for a fraction of the CODAI order of battle. I suspect that many systems were and remain viable.

And then there are all the MANPADS and AAA all over the place. Remember that the main supported mission was the 160^th SOAR Blackhawks with the Delta team. If you can see these well enough to get cell phone video, you can see them long enough to fire volleys of MANPADS. Two and a half hours over the target is just a really long time. It seems that some of the Blackhawks sustained some damage, but it’s almost miraculous that we didn’t lose any. I am not casting any shade on the operational aspects, and on the contrary I am very impressed by what appears to be a meticulously planned multidomain effort, but I am still surprised by the results.

The US military is incredibly skilled at this sort of thing, and CODAI operators may be less so, but I’m not convinced that operational excellence sufficiently explains the extremely lopsided outcome. This is why I think an inside job is plausible. We just don’t have a lot of information about what that looks like, but it seems possible that some commanders were persuaded not to fight. One possibility is that the CIA made them an offer they couldn’t refuse, either the carrot of running the successor regime or the stick of getting the Maduro treatment. I have many, many questions about the extent of prior coordination with Vice President Rodriguez and other regime elements. This would mean that the US decapitation would also be a CIA-enabled coup.

While I am starting to sound like a conspiracy theorist, we might as well consider another possibility that the Russians got involved. Maybe the Russians withdrew technical support, or at least they didn’t offer any active support to a US operation they certainly might have seen coming. It’s sort of weird that Russia helped Venezuela build up a fairly decent IADS only to watch it get shredded with hardly a complaint (a few proforma condemnations of American lawlessness). At the risk of indulging conspiracy theory, maybe the Russians offered inducements to Venezuelan forces for standing aside after receiving inducement from the Trump admin for doing so. This is pure speculation, of course, prompted by Fiona Hill’s 2019 testimony that the Russians proposed some sort of Venezuela for Ukraine deal: you sell out your proxy and I’ll sell out mine.

Cyber warfare is boring

Cyber may be efficient, but this does not necessarily mean that it is highly effective in any dramatic way. Age of Deception is skeptical about the “cyberwar” narrative that cyberspace creates independent tools for causing massive disruption or other strategic effects. It contributes to the alternative perspective that cyber operations are either a new way of conducting intelligence or a useful supporting enabler for military operations. This is, cyber is a complement not a substitute for war.

The admittedly very sketchy reports of cyber in Absolute Resolve are entirely consistent with this emerging perspective. Cyber might have been used to turn off the lights (although it seems like the city lights are still on in videos from Venezuela and electronic warfare might have provided a simpler and more reliable way of doing the same things). Cyber might also have been used to turn off or confuse radars or Venezuelan communications. These are supporting activities that improve efficiency by creating an operational window for friendly forces. When the defenses are down, you don’t need as many forces to fly through them. This is entirely consistent with the way in which Israel allegedly used cyber in 2007 to degrade Syrian radars during an IAF air raid.

Such cyber capabilities are nice to have as enablers that marginally improve efficiency and provide a temporary relative advantage. They might be especially desirable if there is a need to only temporarily and reversibly degrade capability. But they are not a silver bullet. Commanders may even be reticent to rely on spooky cyber enablers when more effective if less efficient means are available (i.e., bombing radars). Getting cyber to work well as a complement requires a firm institutional and doctrinal foundation to control the ubiquitous complexities and uncertainties inherent in remote software operations. CYBERCOM has been working hard to provide this institutional foundation for a decade and a half, and perhaps Absolute Resolve is a milestone of progress. We won’t know for a while.

Efforts to institutionalize cyber into US military operations have in effect made cyber boring. I mean this as a complement. Cyber is boring in the same way as air refueling, electronic warfare, and combat search and rescue are boring. It is now a standard and expected part of any modern US military operation. The surprise is no longer that cyber is used (or that its use tends to be exaggerated in media reports). It would be more surprising if cyber was absent. By the same token, cyber is now just an institutionalized part of the military machine, not a strategic substitute for it. And let me say again, the key intelligence asset in this entire affair, the thing that made it possible to capture Maduro with relatively little resistance, was human espionage. Human sources remain vital in a cybered up world.

“Cyberspace” is an irritating yet unavoidable term for gesturing toward the ubiquity of technical devices and networks in every area of modern life. This is arguably making secret statecraft more prevalent because tech provides so many new avenues for collecting and manipulating information. More specifically, because information systems improve the efficiency of economic activity, they can be hijacked by secret statecraft to improve the efficiency of politics and warfare.

Yet while the tradecraft of the spy business is novel (and more businesslike) the nature of secret statecraft is older than Sun Tzu. Whether organized deception is employed by civilian agencies or military organizations or criminal hackers, we can describe some common strategic advantages and common strategic constraints. As an international relations scholar interested in the strategic essence of things, I needed a covering term for the shadow realm of politics, no matter the specific actors and instruments involved. I also needed a way to talk about the political purpose of cybersecurity without tying concepts to particular technologies that tend to change quickly. Thus, secret statecraft.

Jackpot!

I wrote Age of Deception to make sense of cybersecurity, but a nice thing about secret statecraft (which I argue is the strategic essence of cybersecurity) is that it can be used to discuss things other than cybersecurity. This includes special operations, which also rely on a lot of stealth, secrecy, and deception. Like cyber nerds, SOF also operate in the gray zone of authorities between the military and intelligence community (and law enforcement). Unlike cyber, however, SOF are still military forces with guns and bombs. They are shifted further toward the pure military side of the spectrum, but still highly dependent on secret statecraft. I wrote about SOF a little bit in Information Technology and Military Power.

If cyber was just a nice-to-have efficiency improver in Absolute Resolve, the same cannot be said of SOF. Indeed, the Delta Force team was the main effort. It’s hard to overstate how remarkable this is given the scale of the operation. The entire operation was essentially a SOF raid to bag a high value target (HVT), with a much more extensive enabling force than most other HVT raids. While SOF traditionally are a supporting enabler for military operations, in this case we have a major military operation enabling a SOF raid. It seems that the red-headed step-child has grown up to become king.

A key point here is that the Maduro raid was built on two decades of SOF learning and institution building to figure out how to find, fix, and finish HVTs. US SOF—specifically JSOC and special mission units like Delta Force, the 160^th Special Operations Aviation Regiment, and SEAL Team Six—have developed extensive doctrine and capabilities throughout decades of the War on Terror for conducting these kinds of operations, resulting in literally thousands of individuals killed or captured. The end result is a particular style of network centric warfare that integrates extensive in intelligence preparation with highly precise military operations.

This institutional element of success in the Maduro operation is based on years of hard effort, and many mistakes and painful lessons learned. (Military lessons are written in blood.) Another way to put this is that the efficiency advantages of secret statecraft are not really that cheap, ironically enough. If you really want to have effective intelligence performance, I argue in the book, then you need to have an effective clandestine organization as an enabling platform. The SOF counternetwork machine, supported by CIA infrastructure and the entire conventional military planning process, provides one such platform, tuned for a particular purpose that happened to be very useful in this case.

What do I even know?

This post is much longer than I intended, so I am going to pause for now. This is really just my open notebook of emerging thoughts rather than any sort of careful analysis. In lieu of a conclusion, I will just say that Operation American Rapture, er Absolute Resolve, exemplifies a lot of the dynamics of secret statecraft that I discuss in the book. It exemplifies both the efficiency advantages of combining institutional subversion (i.e., intelligence and deception) with military power, and it exemplifies the strategic constraints and ironies that come with over relying on those advantages.

I also want to flag that there may be some significant deception involved with the Trump administration’s communication about this whole affair with Congress and the American people. I have been musing here mainly about deception in the means of statecraft, but deception about its ends may be even more relevant. But that is a story for another time.

Relatively Unresolved

Let’s start with the name—Operation Absolute Resolve. What is that supposed to mean? What exactly is the United States absolutely resolved to do? Are we resolved only to decapitate Venezuela? To run the country after it collapses into chaos? To take out any other national leader that we don’t like? To aggressively assert the “Donroe Doctrine” throughout the Western Hemisphere? To finally fight and win the war on drugs (which are mostly cocaine not fentanyl, and mostly bound for Europe not the US, as we pardon the former Honduran president imprisoned on drug charges)? Maybe the SecWar just thinks it sounds cool in a Signal chat.

There is another possibility, given the doublespeak tendencies of this administration. This is that the operation was named “absolute resolve” precisely because American resolve toward Venezuela is relatively lacking.

Resolve is willpower. A highly resolved actor is motivated to try and do something it cares about deeply, no matter the costs or obstacles. You can demonstrate resolve by taking risks and having skin in the game. In the popular bargaining model of war, resolve is defined as simply the willingness to pay. A resolved actor is willing to spend down some of the surplus of the negotiation to signal that it is serious. Thus, we describe costly signals of resolve with metaphors of tying hands or sinking costs (or burning bridges or burning money). These are things that you are willing to do only do if you really care about getting your way. Bluffers and braggarts will not do these things because they are not resolved.

“Absolute resolve” thus implies a willingness to pay absolutely. An actor with absolute resolve should thus be willing to pay any price and bear any burden, to paraphrase the former sole namesake of that performing arts center in DC. The name of the recent US operation thus implies that the stakes of the Venezuela operation must have been grand indeed. But this is absurd.

In reality, there are few things short of existential survival for which a state will be willing to pay absolutely (and maybe not even then, given that infinite costs will outweigh any conceivable reward). There are many things that a state might be willing to try if the costs and risks are expected to be low, but unless there is a vital national interest at stake, most political leaders will be less likely to risk blood and treasure, or their chances for political survival, on a costly gamble.

The stakes in Venezuela, for the United States, were never very great. Maduro was an unpopular and illegitimate autocrat of a minor regional power dabbling in the European cocaine trade. An irritant to be sure, but one who had been endured and could be endured indefinitely. Imagine a counterfactual world in which deposing Maduro would have required trillions of dollars and thousands of American deaths over years of fighting, as in Iraq and Afghanistan. I have a hard time imagining that any American government would ever be resolved enough to pay this price, let alone one that ran on no more foreign wars. Measuring resolve is hard without a big war, but I expect that the United States was in fact willing to pay very little for Operation Absolute Resolve.

The only reason that the Trump administration could get away with decapitation is that the U.S. military was strong enough, and the Venezuelan military weak enough, that it expected to succeed at relatively low cost. There was no expectation of retaliation by Venezuela or others (more on that below). Thus, through technical ingenuity, brilliant planning, exquisite intelligence, and elite special operations forces, the U.S. military was able to promise its civilian masters a quick and relatively painless raid.

Decapitation on the Cheap

According to the New York Times, “More than 150 military aircraft, including drones, fighter planes and bombers, took part in the mission, taking off from 20 different military bases and Navy ships” in a mission that “took about two hours and 20 minutes.” This probably makes it the most elaborate special operations raid of all time, but it was also the quickest and easiest sovereign decapitation in recent history. A few US helicopters were reportedly shot up, and six service members were injured, but it seems that all Americans who got in got out. Maybe the story is not so genial for intelligence sources or officers on the ground, but we don’t know much about that yet and their fate is low profile by design. The U.S. military delivered on its promise and decapitated Venezuela on the cheap, at least in this initial phase.

But low-cost military operations create a serious, and seriously underappreciated, problem for coercive diplomacy. The problem is that if you have an effective low-cost military option, I cannot tell if you are using it because if is effective or because it is cheap. Maybe you are highly resolved, and the high-tech SOF raid is the best way to get what you want. Or maybe you are only marginally resolved, and you are only willing to conduct that raid if you can to it on the cheap. In bargaining terms, low-cost military power causes low-resolve and high-resolve types to pool together.

Erik Gartzke and I wrote about this in Elements of Deterrence for the case of drones and other artificial intelligence (AI) enabled weapons. AI weapons may be useful on the battlefield but have little skin in the game (almost by definition). The tragedy is that actors who want to test the resolve of the actor with the low-cost military strategy will look for other ways to impose costs; thus we have empirical evidence that militants subjected to drone attacks will shift to attacking civilians more often. The secrecy that often accompanies technical operations, and which was certainly a critical precondition for the operational success of Absolute

This may have troubling implications for Venezuela. Absolute Resolve was so cheap and easy that even a rather unresolved government might be willing to attempt it. Or maybe it was just the best way for a highly resolved actor to act under the circumstances. Now that we’ve seen it, what have we learned? Should we judge that the Trump administration is a high-resolve type with an effective decapitation option. Or is it a low-resolve type with a low-cost option? We can’t tell directly, at least not from this operation alone. But I think there are plenty of other reasons to guess that it’s the latter. That is, Trump’s Absolute Resolve is most likely a fig leaf for relatively low resolve. We know that Trump likes to punch down. It’s no accident that he decided to pick on Venezuela rather than Russia or China.

The troubling implication is, what happens when someone tries to figure out what type of resolve Trump truly has? Say you are an angry Venezuelan nationalist who resents being attacked by the neighborhood imperialist. How would you test Yankee resolve? You need to find another way to impose cost. You can try to make it hard for Trump to “run the country,” perhaps by starting an insurgency, oppressing the civilian population, or flooding the region with migrants. Maybe you will discover in the process that Trump is in fact resolved. That would be bad news for millions of Venezuelans as Trump presides over a bloody American occupation. On the other hand, maybe you will find that the TACO principle holds: “Trump always chickens out” especially in the face of highly resolved opponents. There’s a joke in there somewhere about TACO being more likely in Latin America.

Sadly this sort of test is all too plausible. Absolute Resolve was a leadership decapitation raid. It was not a regime change. The regime is still very much in place. This appears to have been part of the rationale for controlling costs, as Trump implied that Acting President Delcy Rodríguez would manage things. But Rodríguez and other Venezuelan officials may not be so willing to be quislings: Rodriguez denounced the US operation as “a barbarity” aimed at “the seizure of our energy, mineral and natural resources,” even as she softened her tone somewhat. Nationalism is a potent force, and the specter of renewed Imperialismo Yanqui could encourage resistance to US efforts. There is also no shortage of corrupt violent entrepreneurs in the region waiting to fill the political vacuum created by Maduro’s abrupt departure. The Venezuelan regime or other nonstate actors could thus act to test American resolve after the glow of “mission accomplished” begins to fade.

Isn’t that special?

Strategic confusion manifests in other ways, too. I am tempted to call Absolute Resolve a “special military operation.” The US military certainly used many special operations forces to get the job done. US military planners make a distinction between supported and supporting forces. The main supported forces were the Delta Force commandos, who captured Maduro. All the other elaborate US military enablers acted in a supporting role for Delta Force. This is why I think it’s fair to call this the most elaborate special operations raid of all time.

Map of the 3 Jan 2026 US strikes on Venezuela, Wikimedia commons (that was fast!)

But even more special is that the Trump administration likes to describe the entire military effort is as a law enforcement operation. The actual arrest of Maduro was performed by FBI agents with the Delta team. This means that the raid is also the most up-armored sting of FBI history. Maybe you can argue that this is in fact a quite costly signal of resolve from a law enforcement perspective, as the long arm of Donald Trump (with its small fingers) is demonstrable willing to reach across the precedents of international law to bring evildoers to justice. The signal is muddied at best, because the justifications of the operation are at best varied and ambiguous and at worst completely arbitrary.

There is plenty of precedent for using military forces to support law enforcement, as when Coast Guard teams perform the actual boardings of drug vessels intercepted by US Navy warships (at least until the US military started blowing them out of the water), but that usually happens in international waters, not via a major military incursion. In any case, President Trump and Secretary of State Rubio have been describing the attack as essentially a law enforcement operation, which also seems to be its tenuous legal justification. This suggests that the Trump administration views an indictment from its own Department of Justice as sufficient grounds for launching a unilateral strike to decapitate a sovereign country without seeking Congressional approval.

And we haven’t even mentioned the oil. Or the theory that decapitating the regime without changing it will produce a pliable client. Or that decapitating Venezuela will compel Colombia and Cuba to bend the knee. Or that China will be scared away from a region in which it has been heavily investing for years. Or that “mission accomplished” here is not the prelude to another disastrous forever war. Or that Secretaries Rubio and Hegseth, who Trump said will “run the country,” have any plan or competence for their new role as imperial viceroys of Venezuela. Or perhaps that a foreign diversion might help to unify the MAGA base before the midterms. The strategic rationale of this operation is certainly “special,” in every sense of the word.

“Special military operation” is of course the phrase that Vladimir Putin uses to describe his war in Ukraine, which seems to have been originally intended as a lightning-fast regime-change operation. This went badly for Putin in 2022, of course, because the Russian military was extremely disorganized, and Ukrainians and their supporters were highly resolved to resist it. The operational preparation and coordination of the US military operation to decapitate Venezuela was far more impressive, with an admittedly more modest operational goal of snatching and grabbing the leader. There was a modicum of Venezuelan resistance during the raid itself, but it was largely ineffective, partly owing to the state of the Venezuelan military and more to the US adroitly turning off the lights and suppressing air defenses.

The diplomatic context contrasts starkly, too. Ukraine received crucial diplomatic and material support from other countries to resist the Russian invasion. Venezuela is quite isolated. Russia has provided a lot of military and diplomatic support to the Maduro regime in recent years, but Moscow did not impede Maduro’s abrupt removal. Maybe there was not much that anyone could do, given the nature of surprise attacks. Yet there was ample strategic warning given the large US buildup in the region, including the arrival of numerous specialized support aircraft. Even with Russian military attention focused on Ukraine, there is still much that Russian intelligence services or outright Russian warnings could have done to create friction and risk for the United States.

It is quite likely that Putin simply abandoned Maduro. Fiona Hill testified in 2019 that “the Russians at this particular juncture were signaling very strongly that they wanted to somehow make some very strange swap arrangement between Venezuela and Ukraine…. You know, you have your Monroe doctrine. You want us out of your backyard. Well, you know, we have our own version of this. You’re in our backyard in Ukraine.” We don’t know anything more at this point about whether the second Trump administration has been open to this deal or something like it. It wouldn’t be the first time that Trump accepted help from Moscow. Secret assurances from Russia would be helpful for controlling the costs of Absolute Resolve, especially in case Trump was not actually so resolved to pay them. Secret diplomacy with the Russians at this point remains totally speculative but completely plausible.

The diplomatic situation is still complex and fluid. Reuters reports some pro-forma criticism “that U.S. President Donald Trump’s actions in Venezuela were unlawful and destabilizing.” This sounds like simple schadenfreude to me, using the same language that is often used to describe Putin. Perhaps Russia, and/or China, will put up more resistance if (and when) the US gets more militarily involved. Cuba and other regional actors are additional wildcards, in addition to the Venezuelan actors who have the most skin in the game. In short, there are many ways things could get ugly.

The downsides of military excellence

Operation Absolute Resolve was an impressive piece of operational art and will probably be studied in war colleges for years to come. Yet its strategic and legal justifications are also dubious to say the least. This contradiction is not necessarily a surprise. It would not be the first time in military history that we have seen operational excellence in the service of strategic idiocy. If that comparison is uncomfortable it should be.

I might go one step further and argue that operational excellence is especially vulnerable to capture by dubious strategic ends. The ability to coordinate sophisticated technologies and complex operations across multiple domains has at least three pernicious strategic implications.

First, this ability tends to lower the overall cost and risk of military operations, or at least the short-term assessment of cost and risk. This creates an obvious temptation to reach for the military instrument if you believe that you can achieve cheap success without the political cost of blood and treasure. The use of highly efficient military operations should be especially likely against weaker targets who lack any meaningful strategic deterrent. High tech capabilities are thus more likely to be used by the strong against the weak.

Second, this ability is predicated on the organizational ability to insulate service members from asking hard questions about political goals. Military professionals are highly skilled engineers who solve the problems that are given to them by civilian leaders. Military professionalism focuses the organization on means rather than ends, which makes it more likely to salute smartly and execute even dubious strategic projects. Smart operational organizations are thus less more likely to be used for stupid strategy.

Third, well organized, Hi-Tech operations are very impressive to behold. If everything goes off properly, they are of immense propaganda value. This helps to reinforce the esprit d’corps of military participants and the perception of power for the political regime. You can expect the department of War to be crowing about futurism and masculine lethality for a while. The fetish of lethal efficiency will bolster the popular image of the regime while masking complicated questions about long-term cost and risks.

Fourth, operational excellence may be especially attractive for domestic political purposes. All three of the above points (low costs, objective control, sex appeal) combine to provide an embattled political leader with a powerful means to divide and conquer. There is the diversionary war or "wag the dog" scenario: the arbitrary use of dramatic military power in a polarized political society will be useful for mobilizing and unifying supporters on one side and demonizing and ridiculing opponents on the other. And there is the patronage or cronyist scenario: clients or corporations who stand to benefit from the operation can render various kinds of support or simple enrichment.

One OAR in the water

It is getting late, so I am going to wrap this up. I just wanted to get some of my initial reactions in writing in order to figure out what they were. These are hot takes. They may be bad takes. One important theme that I have not addressed at all here is the important role of cyber, intelligence, deception, and secret statecraft generally in this operation. That is hopefully a topic for a future post. We are all going to be stuggling to make sense of this episode for a long time to come.

A general theme that emerges from these rambling thoughts is that Absolute Resolve is an impressive military operation in the service of a discombobulated political strategy. It is a triumph of short-term thinking that takes on far more long-term risk than I think the Trump administration may appreciate. The very operational features that made special operations enabled decapitation so attractive—low risk, low cost, high rewards—could end up inviting numerous actors to test the administration’s resolve.

Operation Absolute Resolve—OAR—is just one oar in the water, and that’s not sufficient to navigate the ship of state.

On technical ground, obfuscate

On defense ground, exaggerate

On marketing ground, prevaricate

On conference ground, bloviate

On artificial ground, hallucinate

On coffee ground, caffeinate

On sanfran ground, defecate

On wizard ground, apparate

On newwave ground, mediate

INXS, Wikimedia commons