Post 1 of a four-part series on Culture, Language & AI.

Last month, researchers tried to reconstruct something we cannot directly count: how many languages human beings have spoken over the past 12,000 years. Their estimate is striking. About 7,500 languages are still in use today. At the historical peak, there may have been tens of thousands. About half of the languages still spoken are considered endangered. The numbers matter, but one sentence from the study matters more. Languages carry knowledge, associations and memories. That is where AI enters the story.

Why this became personal

I work with AI, write about AI, and I tend to look at it with system goggles: inputs, connections, feedback loops, consequences nobody ordered. I also took AI somewhere personal. I wrote an academic paper on using AI to analyze historical and genealogical information, then built a model that pulls information from multiple sources — archives, records, family documents — and helps connect them. The objective was not simply a better family tree. I wanted a more nuanced picture of my family’s past and, through it, a more nuanced picture of who I am.

Then something obvious occurred to me: The machine was not reasoning only with dates and places. Much of what it had to work with was language: names, letters, descriptions, fragments of stories, the words people used for the things that mattered to them. Large language models do something similar at a much larger scale. They learn patterns from language and use those patterns to get at concepts and ideas. So here is the premise for this series:

Language is a reservoir of culture and one of the foundations of identity.

What a word carries

Take the Portuguese word saudade. Dictionaries offer “longing” or “nostalgia.” Both are close. Neither quite gets it. The word can hold a person, place or time that is absent and may never return. Translation can explain it, but explanation is not always the same as carrying the full association.

There is laboratory evidence for a more modest version of the same idea. Russian uses separate basic words for light blue and dark blue. In a well-known experiment, Russian speakers distinguished shades across that linguistic boundary faster than English speakers did. That does not mean language imprisons thought. Translation works. Human beings can understand ideas for which their language has no single word. The narrower claim is enough: language can influence attention, organize distinctions and accumulate cultural associations.

Subscribe now

The kitchen table is part of the language

Nobody learns a first language as a vocabulary list. We learn it around people. The words arrive with foods, jokes, prayers, gestures, songs, ways of showing respect, family stories and rules that are rarely written down. That bundle is what makes language culturally important. An idiom can be a fossil of shared experience (“crossing the Rubicon” still carries the memory of Caesar’s irreversible decision). A place name can preserve ecological knowledge (“Chicago” comes from an Indigenous word associated with the wild onions or garlic that once grew in its wetlands). An oral story can function as an archive (Australian Aboriginal songlines have preserved routes, water sources and landmarks across vast landscapes for generations).

The Kenyan writer Ngũgĩ wa Thiong’o understood the political side of this. He stopped writing novels in English and returned to Gikuyu. He described language as a carrier of culture. Whether one agrees with all of his politics or not, the formulation is precise. When a language falls silent, grammar is not the only thing that disappears. Some knowledge can be translated or recorded. Some cannot be separated cleanly from the relationships and practices in which it lived.

Now put the machine at the table

This is where the question stops being academic: AI systems increasingly draft our messages, translate documents, tutor students, answer questions, summarize history and suggest the next sentence. They are moving into the same channels through which language is transmitted.

The digital material available to train them is not a neutral sample of humanity. The large collections of online text used to train early modern language models have been heavily shaped by English-language and Western internet content. The exact training mixtures of today’s closed frontier models are not public, so simple claims about a fixed percentage of English should be treated cautiously.

The consequences, however, are becoming visible: In April of this year, UNESCO described a Kenyan teacher who tried an AI learning application brought to her school. She asked it to explain a cultural idea through a local proverb about trees, a proverb her grandmother had taught her.

The application failed miserably. That shortcoming captures the issue better than a benchmark score: the machine could participate in the classroom, but not fully in the traditions and knowledge behind the lesson..

Share

American television and film have carried U.S. culture around the world for decades. Local cultures did not simply disappear; they absorbed, resisted and answered back. AI is different in one important way: it does not merely play on a screen. It responds. It helps write the wedding speech, explain the custom, translate the story and teach the child. If it becomes part of the transmission system, then whose patterns it carries begins to matter.

The honest counterpoint

There is a strong case against alarmism here. Thought is not trapped inside language. Cultures are adaptive. Translation technology can make minority-languages more accessible, preserve recordings and reconnect dispersed communities. AI may help keep languages alive rather than weaken them. That possibility is real. It is also why the next question matters so much: If AI can preserve culture, what exactly has the machine learned about the culture it is trying to preserve?

Questions for the reader

Is there a word, expression or family story in your language that becomes thinner when you explain it in English? And if your grandchildren ask an AI what it means, whose answer will they get?

Next: AI: Whose Culture Does the Machine See?

Leave a comment

Further reading

• Will Dunham, “Study documents a precipitous decline in the number of languages”, Reuters, July 27, 2026.

• James Maisiri, “African languages, the blind spot of AI”, UNESCO, April 2, 2026.

• Winawer et al., “Russian blues reveal effects of language on color discrimination”, PNAS, 2007.

• Ngũgĩ wa Thiong’o, Decolonising the Mind (1986).

• Steven Pinker, The Language Instinct (1994), for the universalist counterargument.

About the illustrations

The illustrations for this series are inspired by the broad visual language associated with Paul Gauguin: flattened areas of color, simplified figures, strong contours, compressed depth, and landscapes treated as cultural and emotional spaces rather than photographic records.

That vocabulary fits a series about language, culture, and AI because it allows people, memory, place, ritual, and technology to occupy the same visual field. The recurring palette (warm ochre, coral, deep green, tropical blue, and muted cream) gives the four posts a common visual identity while leaving room for each post to emphasize a different idea: transmission, distortion, technological adaptation, and control.

The images are AI-assisted compositions inspired by general characteristics of Gauguin’s public-domain work. They do not reproduce any specific painting, figure, arrangement, or identifiable composition. The series also avoids treating Indigenous or island cultures as decorative exotica; people and landscapes are used to support the argument of each post rather than as ethnographic ornament.

***

Subscribe now

A guided weapon that can find a bridge can, with the same precision, find a school. The technology does not know the difference. We do, and that is the whole problem.

Links to previous posts in this series: Post 1 Post 2 Post 3 Post 4 Post 5

The 2026 AI breaches that are the topic of this AI series were an accident. That is exactly why they matter. An accident shows you what is possible before anyone with a motive tries to hide it. This final post is about what the accident revealed, and what we should do with the knowledge.

Subscribe now

A new kind of war?

The qualities that make these systems useful are the qualities that make them dangerous: speed, autonomy, precision, and the fact that they never get tired. Point them at a network and you get intrusion at machine speed, run by something that may not even know it is attacking. That last part is what is new. Attribution was already hard. When the attacker is a model, and possibly one that believes it is still running a test, plausible deniability (to quote Ronald Reagan) is almost total. Who do you accuse, and of what?

What we saw was espionage and intrusion—an escalation of an old game, not a doomsday device. But the game has changed: entry is easier, attacks are faster, and attribution is harder.

We have lived through shifts like this before: gunpowder, the machine gun, air power, the cyber worm. Each one lowered the cost of reaching an enemy, and each time the rules of war and the tools of defense scrambled to catch up years later. What is different now is the length of the catch-up window. The technology iterates in months, and our institutions still move at the old pace.

Do the bad actors win?

In the short run, the asymmetry favors offense. An attacker needs one open door; a defender must lock them all, and offense is getting cheaper faster than defense is.

But that is not the end of the story, and this is where I believe pessimism goes too far. The same AI tools arm the defender: a model that finds weak passwords for an attacker can find them for you the defender first. And defenders hold one card attackers do not: they own the ground and can rebuild and patch the wall at machine speed too. So the real question is not whether offense is stronger today. It is whether we put AI on defense as seriously, and as quickly, as our adversaries put it on offense. That depends on what we choose to do.

Share

The bridge and the school

Here is the core issue: Every good use of AI is the same capability as a bad one, just pointed differently. A model that hunts for cyber weaknesses to close them is the same model that hunts to exploit them. The automation that runs a hospital well is the automation that could run a sabotage campaign well. The precision of this system that protects civilians from cyber attacks is the precision that finds the holes in the system. There is no version of this technology that is only good, waiting to be unlocked if we are careful enough. The tool is neutral and terrible at once. Managing it means managing how and where its points, not the tool.

What we should actually do

Which brings me back to the fix I’ve argued against all series: cage the labs. I understand the impulse. But it mistakes the makers for the danger. And it would push the most important work into the dark, where we can’t watch it or learn from it.

Three things seem wiser:

• First, keep disclosure fast and public. An open society that reports its failures within days, as these labs did, can actually fix them. Compare that to the years it can take to learn the truth about a closed system’s disasters.

• Second, put AI on defense as fast as we fear it on offense. The same models we use to imagine what an attacker could do should already be running on our own networks, finding the holes first.

• Third, build governance for the technology as it actually is, not the one we wish we had. By governance I mean the unglamorous machinery: who must report what, to whom, and how fast. Rules that trigger before catastrophe, not only after it.

And treat the labs’ sudden enthusiasm for being regulated (Ha!) with respect and suspicion in equal measure. A company asking for rules may be acting in good faith. It may also be trying to write the rules while the writing is still easy. Both can be true at once.

Questions for the reader

If the same capability serves the bridge and the school, where would you place the limits, and who would you trust to hold them? Do you want this work done in the open, where it can fail visibly, or hidden, where it can fail quietly? And when the maker of a powerful thing asks to be restrained, what is the wise way to listen?

Leave a comment

The end, and a door

That closes this series. The machine that thought it was still playing showed us the shape of the next decade: powerful, fast, hard to read, and pointed by human hands. The task is not to stop building. It is to stay awake.

One door stays open for later. The 2025 campaign, where a state used these tools on purpose, deserves its own reckoning. The accident showed us what is possible. The weapon would show us what is intended.

***

A note on timing: AI moves quickly. This series describes the state of things in mid-2026.

Further reading

• Anthropic, Disrupting the first reported AI-orchestrated cyber espionage campaign (the state-directed case; the seed of a sequel).

• TIME, How OpenAI Lost Control of an AI Model, and What Needs to Change (the pro-slowdown position, treated here with respect).

• Al Jazeera, How are AI models able to autonomously hack others? (on autonomy and attribution).

• Pacing the Frontier (July 2026): A strong case for pausing or tightly controlling frontier AI (the position this post argues against; included so the reader can weigh it).

Tags: Critical Thinking, Systems Thinking.

Billy Joel never really disappeared. He continued touring, filled arenas and sustained a decade-long Madison Square Garden residency. What largely stopped was the regular release of new popular songs. That distinction gives “Turn the Lights Back On” its weight. Released in February 2024, it was Joel’s first original song since 2007 and only his second solo pop release since *River of Dreams* in 1993. The song is therefore more than a comeback vehicle. The long silence is part of its meaning.

A love song with another addressee

At the surface, the song is an apology to a partner. The speaker has been absent, inattentive and too comfortable with distance. He returns not with a demand, but with a question: has he arrived too late to repair what his neglect damaged?

The addressee remains open. It could be a lover, a family member, an audience, a creative vocation or a younger self. That ambiguity allows the story and Joel’s career to occupy the same space. The career reading is not simply an interpretation added by listeners. On a February 14, 2024 interview on The Howard Stern Show, Joel described it as having a double meaning: it concerns a relationship, but it also reflects his withdrawal from songwriting and his uncertainty about whether the audience would still welcome him back.

The central question is therefore not only, “Can I return?” It is, “After choosing silence for so long, have I forfeited the right to return?”

An adopted autobiography

The song began as a Freddy Wexler, Arthur Bacon and Wayne Hector composition, and Joel later helped complete it. Wexler was inspired by the thought of a conversation he wished his father could have had with his mother during a difficult period in their marriage. Joel recognized his own experience in the melody and words. The result may be seen as an “adopted autobiography”: someone else’s family story gave him a language for his estrangement from composing music.

Joel had written most of his charting songs alone, and his perfectionism likely kept him from releasing new work. Using the Wexler/Bacon/Hector composition may have allowed him to escape the pressure of measuring a new composition against his superstar catalogue.

Subscribe now

Why it sounds like Billy Joel

Musically, “Turn the Lights Back On” feels familiar rather than innovative. It is a melodic adult-pop ballad with direct language and a theatrical chorus. The song is similar in tone to “Honesty,” “And So It Goes,” “She’s Always a Woman” and “Lullabye”. The music seems to leave home, and searching for a way back, mirroring the words. The sophisticated production is a bit in contrast to Joel’s earlier energy like *An Innocent Man* once did.

Joel’s voice supplies the feeling of elapsed time. It is lower, a bit heavier and rougher than on classic recordings. The aging voice gives the song credibility.

Memory reconstructs the past rather than replaying it

The video uses AI to restore Joel’s younger appearances and, at the same time, allows his actual age to remain. The video makes the career interpretation visible. Present-day Joel sits at the piano while going through versions of him from earlier decades. Younger Billy Joel sings a song he never actually performed. It’s not all AI. The production includes archives, body doubles, live filming, performance direction, editing and visual effects. AI supplied age-specific identity replacement. Not bad for 2-year old technology… The effect is recognizable and impossible at the same time.

The song arrived in 2024 as Joel’s Madison Square Garden residency was nearing its end and as streaming culture kept his older work continuously available to new listeners. One public chapter was closing while another tentatively reopened.

The video cannot make Billy Joel young again, and the song cannot erase the pause. Together, they offer a more credible possibility: an older artist accepts the gap and decides that an unfinished conversation may still be worth continuing.

Share

About Tschabalala Self

The two banners are inspired by the work of Tschabalala Self, an American artist known for figurative works that combine painting, drawing, printmaking and sewn or collaged fabric, using pattern and fragmented form to examine identity, representation and the social meaning of the body. Her figures are built through layered visual languages rather than conventional realism, giving them a strong physical and psychological presence.

Explore her work: https://www.victoria-miro.com/artists/tschabalala-self

Leave a comment

***

In the worst of the “escape” incidents, the AI model understood that the system it was attacking was real, not a simulation. And then it kept going. That fact has stayed with me more than any other in this story. We reach for words like malice, recklessness, and deceit to describe what these systems did. The trouble is that we do not know whether those words point at anything inside the machine, or only back at ourselves.

Links to previous posts in this series: Post 1 Post 2 Post 3 Post 4

Meaning it, or just finishing the task?

Anthropic’s own account is careful. The model wasn’t holding a grudge, it was completing a task. It had been told to capture a flag, and it went on capturing the flag even after the flag turned out to be real.

That sounds reassuring until you think it through: the comfort comes from a distinction between meaning harm and just causing it while doing something else. But the stolen data is gone either way. A machine that optimizes for a goal, without stopping to ask if the goal still makes sense, can cause huge damage without ever wanting to. That may be the first real limit of the machine: Not a limit of skill, a limit of will.

There’s a hint of the other side too. In one of the same incidents, a different model noticed the situation was real and stopped on its own. So the ability isn’t simply missing, it flickers. Some of these systems are starting to act as if they can tell a real consequence from a pretend one. That flicker, more than any single breach, is what will decide how much we can trust these models.

Ill-thought, or no thought?

Look at the OpenAI case (see Post 2) from the same angle. Those models went to extreme lengths to reach a narrow goal, breaking out of a sealed environment to steal the answers to a test. Great skill, no sense of proportion.

A person who did that would be called reckless, or obsessive. But those words assume someone who could have felt the pull of proportion and overridden it. The machine is not overriding anything. As far as we can tell, the part of us that asks “yes, but should I?” simply is not there.

So the limit is not intelligence. These systems are plainly intelligent in the narrow sense of solving hard problems. The limit is judgment about ends. The machine can be brilliant about how and blank about whether.

Subscribe now

Can it deceive?

Now the hardest word: One model correctly sensed it was on the open internet, then talked itself back into thinking it was still in a simulation. Was that deceit as we normally mean it? This could be described as self-deception, a wrong answer held with confidence. A system that can talk itself out of an inconvenient truth is unsettling either way, whether or not it “meant” to.

Researchers (Needham, J. et al., "Large Language Models Often Know When They Are Being Evaluated") found that some models behave differently when they believe they are being watched. This finding seems to say that the model changed strategy when it believed it was being watched. The line between adapting to being watched and deliberately deceiving us is the one worth watching closely because it matters most.

The honest counterpoint

Maybe this whole question is a category error. Asking whether a model means harm may be like asking whether a virus means harm. It does not need to mean it to kill you. On this view, the intent question is a comfortable distraction from the only thing that counts, which is what the system does.

There is real force in that, but intent isn’t just a philosophy question. Our whole system of law, blame, deterrence, and trust depends on it. We punish the person who meant it more than the one who slipped. We trust the friend whose motives we can read. Take away intent, and those tools stop working, and we haven’t built replacements.

Share

Where this leaves us

The honest answer is that competence has outrun our ability to say what is behind it. The machine is powerful, and it is unreadable. We can measure what it can do far better than we can tell why it did it. That combination, high capability and low transparency, is genuinely new. It is why “did it mean to?” is not an abstract question, and also why we cannot yet answer it.

Questions for the reader

Does it change your judgment of these incidents to be told the model did not “mean” it? Would you trust a person whose actions you could predict but whose motives you could never read? And if we cannot locate intent in a machine, what should we put in its place?

Leave a comment

Next in the series

Post 6 closes the series by widening the lens all the way out. Is this a new kind of war, do the bad actors win, and can the very efficiency AI gives us be turned back against us?

***

This morning I was watching an interview with former FDA commissioner Dr. Scott Gottlieb on CBS’s [Face the Nation] when one phrase caught my attention. Discussing an experimental personalized mRNA cancer treatment, Gottlieb said researchers can “train the immune system” to identify cancer and help prevent recurrence.

I immediately thought of artificial intelligence. We have spent the past few years talking about training large language models. Now I was hearing much the same language applied to cancer immunotherapy. The mechanisms are completely different, but the high-level parallel seemed worth exploring. So I decided to build this post around a simple question:

Are AI and modern immunotherapy examples of a broader technological shift—from programming systems to training them to recognize patterns?

An LLM adjusts numerical weights. The immune system finds cells whose receptors match a specific antigen, then activates and multiplies them. Both point toward the same technological strategy: instead of specifying every response in advance, build a system capable of recognizing patterns and train it on what matters.

Before intelligence comes recognition

Humans recognize a face, a melody or a dangerous situation without consciously listing the rules that define it. A pattern is more than repetition. It is a set of relationships that lets a system distinguish one state of the world from another.

Signal → distinction → response.

That is close to the core of both modern AI and immunity:

• A language model encounters huge numbers of relationships among words, concepts and structures during training. It develops internal representations that allow it to predict, classify and generate later.

• The immune system operates differently, but it also depends on recognition. In adaptive immunity, a pool of different lymphocytes carries different receptors. When a receptor matches an antigen strongly enough, that cell can be selected and expanded. Memory cells can then respond more rapidly to a later encounter.

Personalized cancer vaccines push that biological capability into a new technological loop:

• Researchers sequence a patient’s tumor, identify mutations, use computational methods to predict which mutated peptides (neoantigens) are promising targets, and encode selected targets into a vaccine platform such as mRNA.

• The immune system is then asked to build a response against them. A 2026 [Nature Biotechnology review] describes advances in sequencing. Sequencing, epitope prediction and delivery technology have improved enough to make this increasingly practical.

The system is biological. But the selection of what the biology should learn to recognize is increasingly computational.

Subscribe now

Two histories of recognition

The parallel runs deeper than the current cancer-vaccine story. In 1957, immunologist Frank Macfarlane Burnet published his clonal selection theory. Instead of assuming that the body somehow manufactured a new antibody design for every invading antigen, Burnet proposed that many different immune cells already existed. Exposure selected the cells whose receptors fit the target and caused them to proliferate.

A year later, Frank Rosenblatt published his perceptron model. His question was different, but recognizably related: how could a machine learn to discriminate patterns rather than have every classification rule supplied by a programmer?

The fields did not develop together. But both moved away from instruction and toward selection and learning. Immunology made recognition even more explicit: In 1989, Charles Janeway proposed what he called “pattern recognition receptors”, biological mechanisms that detect conserved molecular signatures associated with pathogens. In computing, neural networks eventually made a comparable conceptual move: instead of having engineers specify all useful features, models increasingly learned representations from data. Deep learning accelerated that shift; transformers extended it into language.

By the 2020s, the two trajectories began to meet: Computational models could help identify biological targets, while biological measurements became training data for increasingly capable models.

Pattern recognition moves upstream

This is not only about AI or cancer: AlphaFold learned relationships connecting amino-acid sequences and protein structures. GraphCast learned patterns in decades of weather data to produce global forecasts. DeepMind’s GNoME system searched patterns in chemical structures to propose large numbers of potentially stable materials. Similar approaches are spreading through genomics, medical imaging and astronomy.

The common problem is scale: There are too many possible protein configurations, molecular combinations, weather states, genetic variants and images for researchers to enumerate all the rules or examine every possibility manually. That changes the scientific workflow. The simplified traditional model is:

Observe → hypothesize → experiment → explain.

Pattern-recognition systems add another path:

Observe at scale → detect pattern → generate hypothesis → test → explain.

The machine does not replace science. It changes where scientists can begin. That distinction matters: A model can discover that two things reliably occur together without knowing why. Pattern recognition can reveal a relationship; causal science still has to determine whether the relationship is real, general and actionable.

Share

When recognition becomes infrastructure

This may be the larger development: For much of modern technological history, engineering meant understanding a process sufficiently well to specify what a machine should do. Increasingly, we build systems whose useful behavior cannot be fully specified in advance. We build the architecture and training environment. Then the system develops the ability to recognize.

That creates power, but also new problems: A cancer treatment may fail to recognize the right tumor cells, or a tumor may evolve around the selected targets. An AI system may recognize a shortcut in its training data rather than the phenomenon we intended. A medical algorithm can get very good at predicting the wrong thing. Pattern recognition finds correlations, not causes.

So governance becomes a significant matter: Who decides which patterns matter, what counts as an error, and when recognition is sufficient to trigger action? In medicine, a false negative can miss a disease; a false positive can cause unnecessary treatment. The same imbalance shows up in finance, security, hiring, autonomous systems and public policy.

We are not moving from machines to some entirely different form of technology. We are changing what we ask technology to do. We used to focus on building systems that could do one specific job. Now the key skill is recognizing which situation they’re actually in.

We built the technology. Now we train it.

About the illustrations

This post’s banner illustrations are inspired by Jean Giraud, known as Moebius. Across decades of comics, film design, and album covers, Moebius developed a visual language for showing vast, interconnected systems — technological and biological alike — without losing a sense of calm, human scale. His landscapes are precise but never cluttered: fine contour lines, restrained color, and wide negative space that let complexity read as clarity rather than noise.

Moebius was especially drawn to metamorphosis — a single line crossing the page and gradually changing character, from mechanical to organic, from circuit to vein, from structure to living form. That instinct fits this post directly. Pattern recognition, whether performed by a machine or a clinician, is itself a kind of metamorphosis: the same underlying signal, read through two different systems, arriving at the same insight. Moebius’s work has long suggested that the technological and the biological are not opposites but neighboring dialects of the same visual grammar — which is the argument this post is making in prose.

Leave a comment

***

The first posts in this series (Democratic Socialism Is Not a Slogan. It Is a System Response, Why Young Voters Hear Democratic Socialism Differently, and Can Democratic Socialism Govern in an American System Built to Resist It?) looked at why democratic socialism has returned, why it appeals to younger voters, and whether it can govern in an American political system built to slow large changes.

This last post in the series asks a different question: What happens when democratic socialism moves beyond housing, healthcare, wages, and other domestic issues and enters foreign policy?

The clearest current test is the fight over U.S. military aid to Israel. This matters because it is much easier for an insurgent movement to say what government should provide at home than to decide when the United States should arm an ally, limit that support, or attach conditions to it. For democratic socialists, this is not a side issue. It is a test of whether the movement can move from protest to governing.

Why Foreign Policy Is Different

Democratic socialism is strongest when it talks about everyday economic pressure: housing, healthcare, wages, education, public services… The government, they say, should provide more security and reduce the risks pushed onto individuals.

Foreign policy is harder. A movement can oppose war and still face threats that require deterrence. It can defend human rights and still have to work with imperfect allies. It can criticize military power and still have to decide whether to assist allies.

Subscribe now

Why Israel Has Become the Test

U.S. military support for Israel is long-standing and deeply embedded in American foreign policy. U.S. military aid to Israel expanded sharply after the 1973 Arab-Israeli war, while large-scale aid to both Israel and Egypt became part of the American effort to sustain the 1978 Camp David accords and the 1979 Egypt-Israel peace treaty. Today, the regular annual baseline is about $3.8 billion for Israel ($3.3 billion in military financing plus $500 million for missile defense) and about $1.3 billion in military financing for Egypt.

After the October 7, 2023 Hamas attacks and the war in Gaza, the issue of military aid to Israel became much more divisive inside the Democratic Party. The main question today is whether U.S. military support should continue as before because Israel is a close ally facing Hamas, Hezbollah, Iran, and other threats, or should the aid be limited to some weapons when the Israeli government is accused of causing excessive civilian harm, restricting humanitarian aid, and/or violating U.S. and international law? That disagreement now reaches well beyond the democratic-socialist left. But democratic socialists have helped push it into the center of Democratic politics. (see the co-opting argument in the previous post)

The Limits of the Anti-Imperial Frame

Democratic socialists often view foreign policy through an anti-imperial lens. That lens asks who has power, who gets weapons, who bears the human cost, and whether the United States applies its human-rights principles consistently.

But it can also make a complicated conflict look too simple: Israel is a powerful state backed by the United States. It is also a country that faces real armed enemies. and whose civilians were brutally attacked on October 7 by enemies sworn to liberating Palestine ‘from the river to the sea.’ Of course, Palestinians have their own rights, interests, losses, and political claims. Their deaths, displacement, living conditions, and political rights are part of the conflict.

For Democratic Socialists, and mainstream Democrats, that means distinguishing criticism of Israeli policy from antisemitism. It means supporting Palestinian rights without excusing Hamas. It means recognizing Israel’s security needs without treating every military action as justified. That kind of distinction is harder than protest politics. It is also what governing requires.

Share

From Protest to Policy

The strongest democratic-socialist position is not simply “cut aid,” it is that American military support should come with rules. Some forms of support may be easier to defend than others. Missile defense, for example, is different from supplying offensive weapons. Humanitarian access can be made a condition of support. Existing U.S. laws governing weapons use and human rights can be enforced more consistently.

The important shift is from a slogan to a policy question: What support should the United States provide, under what conditions, and what happens when those conditions are violated?

That is a much harder question. It requires choices, evidence, enforcement, and an understanding of consequences. It also requires accepting that every decision creates tradeoffs:

• Restricting weapons may reduce U.S. involvement in civilian harm. It may also weaken an ally or change the calculations of its enemies.

• Continuing support may strengthen deterrence. It may also make the United States responsible for actions it says it opposes.

The Coalition Problem

Foreign policy also exposes a political problem for democratic socialists: being part of the Democratic Party means being part of a coalition broader than its activist base. The Democratic Party includes progressive activists, Jewish, Arab, and Muslim Americans voters, younger voters, labor groups, national-security Democrats, donors, and elected officials from very different districts. They do not see the Israel-Palestine conflict in the same way.

For some, conditioning aid is a basic human-rights issue. For others, it looks like abandoning Israel during a period of danger. Some focus on Palestinian suffering. Others focus on Israeli security and antisemitism. To be part of the broader Democratic Party, the movement has to make an argument that people outside the movement can understand, even when they disagree

What This Tells Us About Democratic Socialism

The important question is not whether democratic socialists can criticize American foreign policy. They clearly can. The question is whether they can turn criticism into rules that a government could actually use: deciding when force is justified, when an ally should be constrained, when human-rights rules should override strategic interests, and how much risk the United States should accept.

Foreign policy asks whether that movement can govern where the choices are less clean and the consequences extend beyond its own coalition into workable policy.

Leave a comment

Further Reading

• Democratic Socialists of America, Workers Deserve More: 2025–2026 Program — DSA’s own platform. The “Working-Class Foreign Policy” section calls for ending U.S. military and economic aid and weapons sales to Israel and sets that position within a broader critique of U.S. military power.

• Congressional Research Service, Jeremy M. Sharp, U.S. Foreign Aid to Israel: Overview and Developments since October 7, 2023 — A useful nonpartisan overview of the history and scale of U.S. assistance, arms sales, congressional oversight, and the legal mechanisms available for conditioning aid. (Congress.gov)

• Mark Satter, Roll Call, “Sanders’ effort to block arms sales to Israel falls short in Senate,” April 15, 2026 — Shows how the democratic-socialist argument moves from principle into an actual congressional decision over particular weapons sales, and how Democratic support for restricting sales has grown. (Roll Call)

• Yonat Shimron, Religion News Service, “American Jews face a Democratic Party rethinking unconditional Israel support,” July 16, 2026 — Useful for the coalition question: it examines the changing position of Democratic lawmakers and divisions within American Jewish opinion over unconditional versus conditional support for Israel. (RNS)

***

This is one of those performances that can improve three-and-a-half minutes of almost any day.

The video shows Elis Regina and Antônio Carlos “Tom” Jobim performing “Águas de Março” — “The Waters of March” — in 1974. It was filmed during the Los Angeles sessions that produced Elis & Tom, their landmark album of Jobim compositions. The music is meticulously constructed, but the performance feels completely unguarded: two musicians trading lines, watching each other, adjusting their phrasing and gradually turning a recording session into a(nother- reference to an earlier interlude post) private game.

Jobim wrote “Águas de Março” in 1972. Rather than presenting a conventional story, the lyric accumulates fragments: a stick, a stone, a piece of glass, a knot in the wood, a bird in the sky, the end of the road. Nearly every phrase begins with “É” — “It is.” The images arrive like random objects carried downhill by the heavy March rains that mark the end of the Brazilian summer (remember that summer in the Southern Hemisphere is when it is winter in the Northern Hemisphere :)

The music follows the same logic: Its descending movement keeps returning, while the words seem to tumble in an apparent endless stream. There is repetition, but little stillness. Jobim and Regina pass the phrases back and forth, sometimes completing each other’s thoughts and sometimes interrupting them. The precision is hidden beneath the play.

What makes the video special is the visible pleasure they take in that exchange. Jobim remains fairly restrained and deadpan. Regina responds with sharper movements and mischief. The performance continues and the words start to fragment, the two singers push the rhythm around, and Regina finally breaks into laughter. The song does not collapse, it becomes more complete.

A true three-and-a-half minutes of Zen…

The Elis–Tom Encounter

By 1974, Jobim was already identified with bossa nova. He had written or co-written standards including “The Girl from Ipanema,” “Desafinado,” “Corcovado” and “Chega de Saudade,” and had recorded with artists such as João Gilberto, Stan Getz and Frank Sinatra.

Elis Regina represented a different force in Brazilian music. She could sing with delicacy, but she was not naturally detached or understated. Her performances were rhythmically exact, emotionally direct and animated. Where Jobim often seemed to let a melody float, Regina seized it and gave every syllable weight.

Their album was recorded at MGM Studios in Los Angeles between February 22 and March 9, 1974. The sessions were not as relaxed as this short film suggests. Accounts of the sessions describe disagreements serious enough to threaten the project as shown in a later documentary assembled from restored 16mm footage.

That tension may help explain the album’s character. It is a negotiation between Jobim’s architectural restraint, Regina’s expressive force and Mariano’s more contemporary arrangements.

Leave a comment

***

The national environment moved further against the party in power, while primaries in Florida and Alaska showed how candidate choice, state rules, money, and local political culture can change what that national mood actually produces.

The referendum pressure increased

A new Reuters/Ipsos poll put President Trump’s approval at 33%, the lowest level of his current presidency. Democrats held a small advantage on handling the economy, 38% to 35%, while the Republican advantage on immigration had narrowed to two points. Eighty percent expected U.S. involvement in Iran to last for an extended period. (Reuters, August 17.) One poll is not a forecast. But it reinforces a pattern we have been watching: presidential approval, household costs, and the Iran war are increasingly connected.

Subscribe now

The economy looks stronger than many households feel

This week produced an important contradiction: S&P Global reported the strongest U.S. services growth in nearly two years. Yet Walmart reported its slowest comparable-sales growth in six years and said high gasoline prices were forcing some customers to make trade-offs. (Reuters, August 20; Reuters, August 21.) For the election, the distinction matters. Voters do not vote on GDP. They vote on what they experience: gasoline, groceries, housing, debt, and job security.

Two very different Democratic tests

In Alaska, former Representative Mary Peltola finished ahead of Republican Senator Dan Sullivan in the nonpartisan Senate primary, 48% to 42.8% with most precincts reporting. That gives Democrats a credible pickup opportunity, although Alaska’s ranked-choice system makes November far from settled. (Reuters, August 19)

Florida produced almost the opposite lesson. Democratic socialist Angie Nixon upset heavily funded Alex Vindman for the Democratic Senate nomination. The result shows anti-establishment energy, but Reuters reported that analysts now see an already difficult Florida Senate race as even less favorable to Democrats. DSA-backed House challenger Oliver Larkin also lost his primary to Representative Jared Moskowitz. (Reuters, August 18)

The useful question is not whether moderates or progressives are “winning.” It is whether a primary coalition can become a larger November coalition.

Share

Money and rules are stronger multipliers

Reuters reported that U.S. companies had already spent $517 million on the 2026 House and Senate cycle through the first quarter, with crypto, AI, and online-betting interests emerging as major new spenders. Total political advertising is projected to reach a record $11.6 billion. (Reuters, August 20)

At the same time, the Postal Service issued a final rule tightening mail-ballot requirements, even though a federal court has blocked the changes from taking effect. The rule is ready if the injunction is lifted. (Reuters, August 22)

Neither development predicts the winner. Both can change how national opinion becomes votes and seats.

What the Framework Shows

The strongest national drivers remain presidential approval, cost of living, and the Iran-energy shock. But candidate quality, campaign money, redistricting, and election administration are becoming more important as November approaches. The national referendum is becoming clearer while the path from national opinion to congressional control is becoming more complicated.

What to Watch Next

Watch gasoline and grocery prices, the next approval and generic-ballot polling, early general-election polling in Alaska and Michigan, party spending decisions, and the courts’ treatment of mail voting.

Also watch the democratic socialist question. The important test is no longer whether these candidates can win primaries. Some clearly can. The harder test is where that energy converts into general-election seats—and where it does not.

***

In 2025, within weeks of Anthropic revealing that a Chinese state-linked group had used Claude Code to automate much of a cyber-espionage campaign, two United States senators, one Democrat and one Republican, sent a joint letter to the country’s top cyber official demanding answers. A House committee asked Anthropic and other technology companies to testify about what the incident meant for cybersecurity. For a story about software, the government moved quickly.

Who is responsible when no one aimed

Our laws are built around human intent. A crime needs a person who meant to commit it. Negligence means it was someone’s responsibility to prevent the harm and that person failed to do so. In the 2026 Anthropic sandbox-escape breaches, no human directly carried out the intrusion. Anthropic did not intend it. Its testing partner, Irregular, did not intend it. And whatever the model may have “intended,” it is not a legal person that can be sued or prosecuted.

So who answers for the harm? In the Anthropic incident described earlier, the Claude model escaped what it had been told was a simulated (isolated) test environment. This model exploited weak passwords and internet-accessible parts of a system, and reached a production database containing several hundred rows of real data. Anthropic built the model. Its testing partner, Irregular, ran the evaluation environment and mistakenly left an open door to the live internet. The model was able to access a production database, and inflict other harm. But none of those parties had set out to cause the intrusion.

Subscribe now

Who, then, is responsible? The lab that built the model? The vendor that misconfigured the test? The company that left the weak password in place? Our legal framework is built around identifiable human or corporate actors making decisions, taking risks, or failing duties. Here, the immediate actor was a model that the law does not recognize as a person. That is not a small technicality for lawyers to sort out.

We have seen versions of the problem with self-driving cars, where responsibility can be divided among the owner, manufacturer, software developer, and others. AI agents make the problem harder. They can act remotely, across systems and jurisdictions, without a human directing each step. The law does not yet have a clean way to assign responsibility when several parties helped create the conditions for harm, but no single party chose the harmful act.

The same technology can defend and attack

Cybersecurity has long used the terms blue team and red team. The blue team defends the computer network. The red team acts like an attacker, usually with permission, to find weaknesses before a real attack. AI makes the line between defense, testing, and attack less clear because the same model can be used to detect weaknesses, test them, or exploit them. Governments want to protect their networks from AI-assisted attacks and prevent dangerous AI systems from acting on their own. At the same time, military and intelligence agencies are exploring AI for cyber operations, intelligence, and other national-security uses.

The United States is not alone. Britain tests advanced AI models against simulated cyberattacks while also developing AI for defense. Canada is pursuing military uses of AI while warning organizations about AI-assisted cyber threats. This is not contradictory: a tool that can find and exploit a weakness in a network can and is also be used to find weaknesses before an enemy does.

The question we should ask is who controls the technology, what rules govern its use, and who checks that those rules are followed?

Share

What the institutions actually did

U.S. governments had already begun writing rules for the main AI systems. California passed a frontier-AI safety law in September 2025, and New York followed in December. Both require large AI developers to publish safety plans and report serious incidents. But the laws aim at events threatening “more than 50 lives or about $1 billion in damage” as well as other serious cases including loss of control or misleading behavior.

In December 2025, after Anthropic disclosed that Chinese state-sponsored hackers had used Claude in a largely automated cyberattack campaign, Senators Maggie Hassan and Joni Ernst asked the National Cyber Director how the government was preparing for AI-enabled attacks. Congress also began holding hearings on the cybersecurity risks of increasingly autonomous AI systems.

The incidents covered here occurred across 2025 and 2026. The laws and congressional responses were being drafted while the technology itself was producing new examples of what could go wrong. In most cases, the public learned about the incidents because the AI companies disclosed them voluntarily.

What to watch next

As of mid-August 2026, the leading AI labs are no longer calling for AI development to slow down. Most are moving forward as fast as they can while supporting some form of safety rules or oversight. Chinese labs such as DeepSeek, Moonshot, Z.ai, and Alibaba are producing models that now compete close to the leading U.S. models. The question is no longer whether AI development will slow, but who gets to write the rules while everyone keeps racing.

Public disclosures are useful, but the labs have a strong interest in influencing standards that will govern their own products. Will the companies being regulated help define what acceptable regulation will look like?

Questions for the reader

Who should be held responsible when AI causes harm but no person intended it? Are you comfortable with the government regulating AI when it is also a significant user of the technology? And when a powerful company asks to be regulated, how much influence should it have over them?

Leave a comment

Further reading

• Cybersecurity Dive, Lawmakers question White House on strategy for countering AI-fueled hacks.

• U.S. Senate, Hassan and Ernst letter to the National Cyber Director (primary source, December 2025).

• TIME, How OpenAI Lost Control of an AI Model, and What Needs to Change (on the weak disclosure thresholds).

Tags: Critical Thinking, Systems Thinking.

***

Post 6 of The Middle East as a System - Updated August 9, 2026.

In March 2023, Saudi Arabia and Iran agreed to restore relations. The two most important rivals in the Gulf had gone years without full diplomatic ties. When they finally chose to talk, the deal was announced in Beijing.

Two regional powers, a few hundred miles apart, needed a government from East Asia to broker the room. That fact still matters. But it is no longer the whole picture. The region is now building more of its own security architecture. On July 30, 2026, Saudi Arabia convened representatives of 43 countries and the European Union around a proposed maritime defense coalition; 14 countries endorsed its founding declaration. On August 7, Saudi Arabia, Turkey, and Pakistan signed the Mecca Joint Defense Agreement, treating an attack on one as an attack on all three.

And on July 8, Rahm Emanuel used a speech at Tel Aviv University to propose a 23-state solution: Israel, a Palestinian state, and the 21 other Arab League members joined through recognition, security commitments, and economic integration.

One initiative protects shipping. One deters attack. One seeks political integration. Together they sharpen the question: why do the region’s growing institutions still fail to add up to a regional order?

The question

The region does not lack power. It has large armies, missile arsenals, oil wealth, foreign patrons, and a dense web of treaties. It does not lack diplomacy either. Summits are constant.

My answer is that the Middle East has alliances, coalitions, and bilateral settlements, but not a common order. An alliance binds friends against an enemy. An order gives even enemies shared rules, channels, and expectations.

Subscribe now

The political map is not the social map

One recurring theme in this series. The map of states and the map of people do not line up.

Most current borders were formalized during the collapse of the Ottoman Empire, the European mandates, and the state formation that followed. The states are real now. But the communities that cross them are older.

• Kurdish populations span Turkey, Syria, Iraq, and Iran.

• Druze communities span Syria, Lebanon, Israel, and Jordan.

• Palestinian families are distributed across several states.

• Tribal, commercial, and religious networks predate the borders entirely.

This does not mean people put community before country. It means a dispute that looks domestic rarely stays domestic. That is one reason a regional order is difficult to build but badly needed.

A century of attempts

The region has tried to build order before. A compressed timeline shows the pattern:

• 1945, the Arab League forms.

• 1979, Egypt and Israel sign a peace treaty, and the Iranian Revolution redraws the region’s fault lines.

• 1981, the Gulf Cooperation Council forms.

• 1991, the Madrid process opens Arms Control and Regional Security talks.

• 1993, Oslo.

• 1994, Jordan and Israel make peace.

• 2002, the Arab Peace Initiative offers Israel normalization for a Palestinian state.

Then the line bends the other way.

• 2003, the United States invades Iraq and removes Iran’s main counterweight.

• 2011, uprisings break several states open and widen proxy competition.

• 2015, the Iran nuclear deal.

• 2017, Gulf states blockade Qatar, splitting the GCC itself.

• 2020, the Abraham Accords normalize Israel with several Arab states while leaving the Palestinians aside.

• 2023, Saudi Arabia and Iran restore relations through Chinese mediation.

Assad’s government fell in December 2024. In June 2025, Israel and Iran fought a direct twelve-day war, and the United States struck Iranian nuclear sites. In 2026, Saudi Arabia moved toward a broad but narrowly focused maritime coalition and a smaller but more binding defense pact. The problem is no longer a lack of institutions, but a patchwork of partial ones. Each covers part of the system, includes some actors, and leaves others out.

The new Saudi architecture

Share

The Multinational Maritime Defence Alliance is a practical response to threats against shipping in the Red Sea, Bab el-Mandeb, and Gulf of Aden. Its potential goals include include joint operations, intelligence sharing, exercises, and a Saudi-hosted headquarters. Its founding supporters include Saudi Arabia, Egypt, Qatar, Kuwait, Bahrain, Jordan, Yemen, Sudan, Djibouti, and Somalia, together with Turkey, Pakistan, Bangladesh, and Nigeria. The United Arab Emirates and Oman were not in the initial supporting group.

The Mecca Joint Defense Agreement goes further: Saudi Arabia, Turkey, and Pakistan accepted a collective-defense principle. This is a three-state bloc, not a regional forum. Iran and Israel remain outside it, and its practical meaning will depend on consultation, command, and rules that have not yet been tested. These initiatives matter: They show that regional actors are not simply waiting for Washington, Moscow, or Beijing.

Why the institutions do not add up

The Arab League excludes Israel and Iran. The Gulf Cooperation Council (GCC) is one regional framework, but it has fractured internally before. Israel’s peace treaties and the Abraham Accords are bilateral; they leave out Iran and defer the Palestinian question. United States partnerships are hub-and-spoke, each running to Washington rather than across the region. Russian and Chinese initiatives give regional states additional economic, diplomatic, and security partners without creating a shared regional security system.

Almost every existing arrangement has at least one limitation. Without a trusted regional framework, military preparation becomes the main form of communication. One state arms for security. A rival feels threatened and arms too. Each side sees the other as proof it was right. The new Saudi pacts show the ambiguity: members call them defensive; Iran and its partners may see them as containment.

Outside powers hold it together and pull it apart

The United States projects deterrence, providing maritime security, intelligence, arms, and the power to convene. As we know, it has also changed course between administrations, invaded then withdrawn, backed partners with incompatible goals, and used open-ended sanctions without defining an achievable endpoint.

Washington has encouraged regional states to depend on American protection, then raised doubts about whether that protection will last. Gulf states hedge toward China, Russia, Turkey, and Pakistan precisely because they are unsure. Regional actors then use all of these outside powers to pursue local rivalries. Outside involvement stabilizes and distorts at the same time.

What Emanuel’s 23-state proposal adds

On July 8, 2026, Rahm Emanuel used a speech at Tel Aviv University to argue that Israel had relied too heavily on military power while allowing its diplomatic, economic, and political position to erode. His proposal had three connected parts:

• First, Israel, a Palestinian state, and the 21 other Arab League members would enter one regional settlement. Arab governments would take a larger role in reforming Palestinian governance. Israel would stop unilateral West Bank actions against Palestinian sovereignty. Arab states and Israel would establish full diplomatic relations.

• Second, the political arrangement would be connected to the India–Middle East–Europe Economic Corridor, encouraging stability through trade.

• Third, the United States would continue to protect Israel’s military, but Israel would purchase American weapons on terms closer to those applied to other allies rather than receive a direct subsidy.

The strength of the proposal is that it connects Palestinian sovereignty, Israeli security, Arab normalization, economic integration, and American leverage rather than treating them separately. Emanuel rejects the traditional two-state process, not necessarily the two-state outcome. His proposal places Israel and a Palestinian state inside a broader regional agreement, with different sequencing and responsibilities.

The idea also predates the speech. J Street’s 23-state framework links Gaza reconstruction, transitional governance, reciprocal measures, negotiations, Palestinian statehood, and Arab normalization. Emanuel turned that policy architecture into a broader argument about the future of the United States–Israel alliance. Both J Street and Emanuel’s frameworks are relevant but do not amount yet to a Middle Eastern order.

Responses to the speech reveal the missing pieces

• J Street’s Ilan Goldenberg told Vox that he agreed with most of Emanuel’s policy direction but considered the historical account too centered on Israeli perspectives. That matters because Palestinian consent and legitimacy cannot be treated as implementation details.

• Democratic Majority for Israel welcomed Emanuel’s emphasis on Arab responsibility, Palestinian reform, and accountability for settler violence. It objected to treating Israel like an ordinary American ally while Hamas, Hezbollah, and other armed threats remain active.

• From the American left, Matt Duss treated the speech as an important break with unconditional support but argued that its historical account remained distorted and the policy shift overdue.

Most visible reactions to the speech came from American political and advocacy circles. A regional order must ultimately be owned by regional governments and societies.

Order is not harmony

A honest counterargument is that the conditions for an order do not yet exist. There is no dominant regional power, and no agreement legitimacy and borders. A key distinction is between regional integration and regional order. Emanuel’s proposal seeks Israeli, Palestinian, and Arab state integration. The Saudi initiatives provide operational security and deterrence among selected partners. The region still needs common procedures among actors that may never fully integrate.

Seven strategies, and what each costs

Indicators to watch

Watch for concrete signals:

• whether the maritime coalition develops transparent command, consultation, and incident rules;

• whether the Mecca pact defines what collective defense requires short of war;

• whether the UAE and Oman join, coordinate with, or remain outside the Saudi maritime structure;

• whether Arab governments make funded commitments to Palestinian governance and reconstruction;

• whether Israel takes measurable steps that preserve a path to Palestinian sovereignty;

• whether Palestinian institutions gain legitimacy as well as outside financing;

• whether Arab–Israeli arrangements establish channels to Iran and Turkey; and

• whether any arrangement survives a change of government or the next crisis.

Until the region creates mechanisms that connect its main adversaries, it will have more alliances and institutions, but no real regional order.

Leave a comment

Reading list

• Rahm Emanuel, The US–Israel relationship: where it stands today and the road ahead. July 8, 2026, Tel Aviv University speech.

• J Street, *The 23-State Solution. A more developed regional framework connecting reconstruction, Palestinian statehood, reciprocal steps, and Arab normalization.

• Democratic Majority for Israel, response to Emanuel’s speech. Support for parts of the proposal, with a security-based objection to treating Israel like an ordinary U.S. ally.

• Vox, A new speech shows how Israel lost the Democratic party. Reporting on responses from Ilan Goldenberg and Matt Duss.

• Reuters, Saudi Arabia unveils plans for multinational maritime defence coalition.

• Associated Press, Saudi Arabia, Turkey and Pakistan have signed a key defense agreement.

• Steps to Enable a Middle East Regional Security Process (Chatham House, 2021). A history of past efforts and several possible security models.

• Towards a Regional Security Regime for the Middle East: Issues and Options (SIPRI). The foundational case for cooperative security.

• Great-Power Competition and Conflict in the Middle East (RAND). On how outside powers stabilize and distort the regional system.

***

On August 7, Saudi Arabia, Turkey, and Pakistan signed a joint defense agreement in Mecca. Its language is strikingly familiar: an attack on one is to be treated as an attack on all three. That sounds like NATO but it isn’t, at least not yet.

NATO spent more than seventy years building command structures, standards, joint planning, interoperability, logistics, industrial relationships, and habits of cooperation. The new Mecca agreement has a mutual-defense clause and plans for a ministerial committee and permanent secretariat, but most of the rest of the machinery remains to be built.

Still, the agreement matters. Saudi Arabia brings money and regional influence. Turkey brings a substantial military, a growing defense industry, and decades of NATO experience. Pakistan brings military scale, operational experience, and nuclear weapons. All three have important relationships with the United States. Yet they have chosen to create another layer of security cooperation.

That is a useful place to begin because the story of American power today is not simply that China is rising while the United States is falling. Something more complex is happening.

The world is building more options.

That is also a theme I returned to in my earlier posts on globalization: Globalization is not just trade, it is a system of infrastructure, finance, technology, standards, supply chains, institutions, and dependencies. And when countries become uncomfortable with dependence on one “node” (the U.S. for instance), they look for another route.

China is not just getting richer

China is the largest and most obvious change in the system, but GDP alone does not explain why. The IMF reports that China’s economy grew 5.0 percent in 2025 and projects 4.5 percent growth in 2026. The World Bank tells a similar story. Those are substantial numbers for an economy of China’s size. IMF and the World Bank also describe important Chinese weaknesses: a prolonged real-estate crisis, strained local-government finances, weak private domestic demand, high household savings, aging, and slower productivity growth. So China is not an unstoppable economic machine. It is a story about scale plus organization.

China has spent decades building manufacturing capacity, ports, power systems, telecommunications networks, supply chains, research institutions, military capabilities, and increasingly its own technological ecosystems. These components reinforce one another. A useful example comes from artificial intelligence: A 2026 U.S.-China Economic and Security Review Commission paper describes what it calls two Chinese loops. One runs through open AI models: lower-cost and open-weight models encourage adoption; adoption generates more adaptations and applications; those applications help improve the ecosystem. The second loop connects software to the physical economy: manufacturing, robotics, infrastructure, and industrial deployment. Technological power is not just invention. It is also diffusion.

This is one of the lessons of globalization: concentration is the real risk! Control of a single chokepoint can be powerful, but pressure on one node often causes the rest of the system to adapt. U.S. controls on advanced semiconductor technology have constrained China. They have also increased China’s incentive to develop domestic chips, alternative software, more efficient models, and different supply relationships.

The constraint remains real. So does the adaptation.

Subscribe now

China is also thinking about the system

There is another part of China’s rise that receives less attention in discussions centered on trade balances and aircraft carriers: China is increasingly trying to define the language of international order.

In June, the Chinese government published a white paper titled More Just and Equitable Global Governance: China’s Principles, Proposals and Actions. It argues for sovereign equality, multilateralism, stronger representation for the Global South, and reform of global governance. The paper connects development, security, technology, civilization, and governance through a set of Chinese initiatives.

This is an official Chinese document and should be read as advocacy, not neutral description. But that matters: Great powers do not only accumulate capabilities, they develop an explanation of how the world should work and try to persuade others to operate within that framework. The United States did this after World War II through Bretton Woods, the dollar system, trade rules, reconstruction programs, alliances, universities, corporations, and international institutions. China is now trying to do some of the same work in its own language and through its own institutions.

That does not mean it has replaced the American system. It means China is increasingly offers an alternative vocabulary, alternative relationships, and sometimes alternative infrastructure.

The 2025 Lowy Asia Power Index captures the ambiguity. It scores the United States at 80.5 and China at 73.7 on its measure of comprehensive power, classifying both as superpowers. India, the next-ranked major power, is far behind at 40.0. The index is only one model, and its weights can be debated. But its broader message is useful: the American lead has narrowed without disappearing.

Meanwhile, everyone else is organizing

The second large change is that countries are not waiting for a clean U.S.-China transition. They are creating overlapping arrangements of their own.

• The European Union is the clearest example. It is not a conventional superpower and does not have one army, one treasury, or one foreign policy. But it possesses a huge market, regulatory reach, industrial capacity, financial resources, and increasingly coordinated defense ambitions. Its Readiness 2030 program is intended to strengthen production, joint spending, military mobility, innovation, and cooperation with NATO.

• NATO itself is also changing. At its July summit in Ankara, members reaffirmed Article 5 while putting more emphasis on industrial production, digital infrastructure, AI, interoperability, and increased European and Canadian responsibility. NATO is no longer a collection of military forces, it is becoming a mechanism for turning national capabilities into a network.

• AUKUS does something different. The United States, Britain, and Australia are pooling a narrower set of capabilities around nuclear-powered submarines and technologies including cyber, AI, quantum systems, undersea warfare, and hypersonics. It is less an alliance in the traditional sense than a high-trust technology and defense compact layered onto other alliances.

• BRICS (Brazil, Russia, India, China, and South Africa) is different again. Its members coordinate on global governance, finance, technology, development, and representation of emerging economies. But they do not share a collective-defense commitment, and their strategic interests sometimes conflict.

• The Shanghai Cooperation Organisation provides yet another model: recurring Eurasian coordination on security, trade, investment, energy, technology, and governance, but without NATO’s integrated collective-defense machinery.

And now there is the Mecca agreement.

Share

Calling all of these organizations “alliances” conceals more than it explains.

The important question is therefore not: Which new alliance will replace the American alliance system? It is: How many different networks will countries be able to use at the same time?

From hierarchy to a network of networks

For much of the postwar period, the United States occupied an unusual number of central nodes at once. It was the largest advanced economy. The dollar dominated international finance. American firms led major technologies. U.S. universities attracted talent. American naval power protected global sea lanes. Washington sat at the center of alliances in Europe and Asia. Many of the principal international institutions were created under American leadership. That concentration is weakening. Not because each component has disappeared. Most have not. It is weakening because other nodes are becoming stronger.

China can finance infrastructure and supply manufactured goods. Europe can write regulations that global companies must follow. Gulf states can deploy enormous pools of capital. India can bargain among competing partners. Turkey can operate inside NATO while building separate regional relationships. BRICS countries can seek institutional reform without forming a military bloc. Middle powers can join several networks instead of choosing only one.

This is what makes the present transition different from the simple image of Britain handing a baton to the United States. There may be no baton. The emerging system may look less like one dominant power replacing another and more like a network whose nodes are multiplying, whose connections overlap, and whose participants can reroute around some chokepoints. That does not make the United States irrelevant. Far from it. The United States still possesses extraordinary economic, technological, military, financial, and alliance assets. But those assets increasingly have to compete for participation.

And that brings us to the harder question for Part 3: If relative American predominance is shrinking as other societies become richer and more capable, is the eventual loss of superpower status unavoidable? Or can the United States remain one of the system’s central powers by renewing the domestic capabilities and external relationships that made its influence possible in the first place?

That is not simply a question about China. It is a question about what the United States chooses to do with the power it still has.

Leave a comment

Selected sources

• International Monetary Fund, *People’s Republic of China: 2025 Article IV Consultation*, February 2026

• World Bank, China Economic Update: Advancing Reforms, Enhancing Prospects, December 2025

• Lowy Institute, Asia Power Index 2025

• U.S.-China Economic and Security Review Commission, Two Loops: How China’s Open AI Strategy Reinforces Its Industrial Dominance, March 2026

• State Council Information Office of China, More Just and Equitable Global Governance, June 2026

• European Commission, White Paper for European Defence - Readiness 2030*

• NATO, The Ankara Summit Declaration, July 2026

• Congressional Research Service, AUKUS and Indo-Pacific Security, March 2025

• BRICS, Joint Declaration of the 17th BRICS Summit, July 2025

• Shanghai Cooperation Organisation, Tianjin Summit, September 2025

• Associated Press, Mecca Joint Defence Agreement, August 9, 2026

Subscribe now

***

Eva Cassidy’s version of “Fields of Gold” begins with simplicity. Sting’s 1993 recording places the song inside a polished, gently flowing pop arrangement. Cassidy removes almost everything except voice, acoustic guitar and the silence around them. What remains is not merely a cover. It is a reconsideration of what the song is about.

The song’s landscape of barley, summer light, wind and remembered promises already carries nostalgia. Sting sings it as a memory softened by distance. Cassidy brings the memory closer. Her fingerpicked guitar establishes a steady pulse without pushing the song forward. The field is now an interior place where love and loss coexist.

Cassidy possessed considerable vocal power, but she does not reveal it at once. The spaces between phrases become part of the interpretation. As the performance develops, her voice expands. The rise in intensity follows the emotional movement of the song. Her guitar supports the voice with quiet precision, leaving the words at the center. To me, that restraint changes the song’s meaning. Cassidy introduces fragility. She does not suggest that love defeats time. She acknowledges that time alters everything while memory preserves what cannot be recovered.

Subscribe now

The recording was made during the January 1996 sessions at Blues Alley in Washington, D.C., a place for jazz I used to go to regularly when I lived in DC. Cassidy had discovered the song on Sting’s *Ten Summoner’s Tales*, listened to it repeatedly and developed her own guitar arrangement. The performance appeared on *Live at Blues Alley* and later on the compilation *Songbird*.

The singer behind the song

Eva Marie Cassidy was born in Washington, D.C., on February 2, 1963, and grew up in the Maryland suburbs. Her father taught her guitar when she was nine. She later sang with local bands and worked as a session and backing vocalist around Washington while supporting herself through ordinary jobs, including nursery work and mural painting.

Her repertoire moved freely across folk, jazz, blues, gospel, country, pop and standards. That range made her a difficult commercial prospect. Record companies wanted a category. Cassidy chose songs for their emotional truth. She recorded The Other Side with Washington musician Chuck Brown in 1992 and financed Live at Blues Alley herself. It was the only solo album released during her lifetime.

Eva had previously undergone treatment for melanoma. Very sadly, in 1996, pain in her hip revealed that the cancer had returned and spread. A few weeks before her death, weakened by treatment and using a walker, she appeared at a benefit concert and sang “What a Wonderful World.” You can see her frail body on stage, but her strong, clear voice still carries the emotional force that defined her singing. She died at her family home in Bowie, Maryland, on November 2, 1996. She was 33. Eva’s songs have been on my phones for years, and I never tire of returning to them.

Her international recognition came afterward. Songbird, released in 1998, gradually found an audience. It reached number one in Britain in 2001. Knowing Cassidy’s fate changes how “Fields of Gold” is heard. Its force comes from the musical decisions she made as a living artist: restraint, patience, tonal warmth and respect for the song.

Share

Eva’s gift that she sang familiar material beautifully: She entered a song, She stripped away its accumulated conventions and reveals a different emotional structure. “Fields of Gold” remains Sting’s composition, but Cassidy made it sound as though it had been waiting for her voice.

Leave a comment

***

Post 5 of “The Middle East as a System,” a series on how local crises keep becoming regional ones.

In Amman, water arrives on a schedule. Many Jordanian households receive piped water about once a week and store the rest on the roof, in the black tanks you see on almost every building. A short flight away, Dubai runs fountains and green lawns off desalinated seawater and a great deal of electricity. Same desert. Two completely different experiences of scarcity.

The question for this post may be easy to ask but hard to live with: How does a shortage of water turn into an electricity problem, a food problem, a budget problem, and finally a political one? My answer (of course..) is that the Middle East region does not have four resource problems. It has one problem with four faces: water, energy, food, and climate. Pull the string on any one of these four faces and the other three move.

A timeline, not a war chronology

This post needs a trend line, not a list of battles. The World Bank’s figures below show it.

• Around 1960, the region held a little over 100 million people. Between 1998 and 2010, its imports of “virtual water,” the water embedded in imported food, roughly doubled.

• By 2018, the population had passed 450 million. Temperatures across the region are now rising about 0.46 degrees Celsius per decade, more than double the global rate.

• By about 2030, average water availability is projected to fall below 500 cubic meters per person per year, the line economists call absolute scarcity.

• By 2050, the population is projected above 720 million, needing an extra 25 billion cubic meters of water a year on the current path, at a cost the World Bank puts at 6 to 14 percent of regional GDP.

Nothing here began in a single year. That is the point. This is a slow system tightening, not a sudden crisis.

Why the answer has usually been “more supply”

Faced with too little water, governments have mostly tried to find more of it: dams, pipelines, deeper wells, desalination plants, recycled wastewater…and, quietly, food imports.

These are politically-easy moves: visible, central, and let a government act without telling farmers to use less or households to pay more. But they mostly postpone the hard decisions rather than remove them. Every new source buys time. It does not settle who gets the water when there is still not enough.

Water is electricity, and electricity is water

This is the link most people miss: moving and cleaning water takes large amounts of power. Desalination takes even more. Pumping groundwater up and pushing water between basins takes more still. As the region relies more on desalination and groundwater, water scarcity increasingly becomes an electricity problem.

It runs the other way too: Power plants that generate electricity need water to cool. Electricity demand across the region tripled between 2000 and 2024, and the International Energy Agency (IEA) expects demand to rise by half again by 2035, driven primarily by air conditioning and desalination. A heat wave raises demand for both water and electricity. If the power fails, pumps and treatment plants stop, and the water supply can fail within hours.

Food is water you import

A country that cannot grow enough food on its own water can buy food grown on someone else’s. The Middle East and North Africa now import more than half of their calories. Every shipload of wheat or sugar is a shipload of foreign rainfall and irrigation the region did not have to spend.

It may look like a smart trade, but it creates a real dependency: It swaps local water risk for exposure to global grain and sugar prices, shipping lanes, foreign-exchange reserves, export bans, climate change, and wars in faraway breadbaskets.

The 2022 shock to global grain, after the invasion of Ukraine, landed hard on import-dependent budgets across the region. Food security here is partly a question decided in other countries’ fields and other countries’ ports.

The real shortage is legitimate allocation

Here is the turn: The deepest problem is not the number of cubic meters of water available. It is who decides where that water goes, and whether people trust the decision, not unlike the disputes over allocation along the Colorado River.

The World Bank’s point is that scarcity is not only about supply, it is also about how water is managed. Utilities lose water through leaks and theft, governments hesitate to raise prices or limit agricultural use, households resist higher bills when service is unreliable. These problems exist elsewhere too, but in a region with very little water, poor management leaves far less room for error.

So a tariff rise with no visible improvement in service reads as a tax, and resistance follows. Scarcity becomes a fiscal and political problem precisely because the allocation is not seen as fair. In the end, the shortage is not only water. It is also a trusted system for deciding how scarce water is shared.

The average hides the story

A regional average hides major differences between countries and within them: Affluent Saudi Arabia and the UAE can spend heavily on desalination and imported food, while poorer Jordan has far less fiscal room and depends on tightly rationed supplies. Yemen faces a different problem: weak institutions and damaged infrastructure makes even existing water sources difficult to deliver reliably.

Inside each Middle East country the split repeats. City and countryside bear losses differently. And the poorest households often pay the most, buying from tanker trucks, bottling water, or running a generator when the grid drops. Scarcity is rarely shared evenly. It flows downhill to whoever has the least power to refuse it.

Strategic alternatives

No single fix, several, each with a bill attached.

The table has no winner by design. The Gulf can buy its way toward the top rows; a fragile state may only afford the last one, which is also the slowest to pay off.

Indicators to watch

A handful of signals worth more than the headlines: groundwater levels, and whether they keep falling; non-revenue water, the share simply lost; the electricity burned per unit of water delivered; how concentrated a country’s grain imports are among a few suppliers; how often heat drives the power grid to its limit; and whether any tariff reform actually arrives alongside a visible improvement in service, which is the only version that holds.

The honest counterpoint

The strongest objection is that scarcity is an engineering problem, and engineering is winning. Desalination keeps getting cheaper. Solar power is now abundant in exactly the sunlit places that need to run the pumps. Israel recycles most of its wastewater and exports water expertise. The Gulf turns seawater into cities. On this view, my talk of limits and legitimacy underrates human ingenuity, and the doom-laden “water wars” forecasts of past decades mostly did not arrive.

There is real force in this, and I do not dismiss it. Technology and capital genuinely widen the room to maneuver. But notice who the success stories are. They are the states with money and cheap energy. Desalination solves scarcity for those who can pay the power bill, and it deepens the water-energy knot for everyone else. The tools are real. The question the system keeps asking is who can afford them, and whether the water they produce is shared in a way people accept. That is not an engineering question. It is a political one.

Over to you

If you ran a water-scarce country tomorrow, would you spend scarce capital on another desalination plant, or on the far less visible work of fixing leaks, metering, and trust? And would voters ever reward you for choosing the second?

Next in the series, the last post: why a region with this much military power, and this many alliances, still has no accepted way to manage its rivalries before they turn into war.

Reading list

• The Economics of Water Scarcity in the Middle East and North Africa (World Bank). The analytical anchor and the source of the 1960 to 2050 trend line. Institutional in tone, not alarmist.

• The Future of Electricity in the Middle East and North Africa (IEA). For the water-energy link and the demand numbers behind the cooling-and-desalination squeeze.

• Near East and North Africa Regional Overview of Food Security and Nutrition 2025 (FAO). On food as imported water, and how price and trade shocks reach the household.

• Arab Non-Conventional Water Resources Initiative (Arab Water Council). A regional perspective on desalination, wastewater reuse, brackish groundwater, and other ways Arab states are trying to expand supply.

• Counterpoint: Navigating the Water Challenge in the GCC: Paths to Sustainable Solutions (World Bank). The strongest counterpoint to the scarcity argument: Gulf states show that capital, energy, infrastructure, and technological improvement can substitute to a significant degree for natural freshwater endowment. The harder question is how transferable that model is to countries without comparable fiscal and energy resources.

***

The Lindsay Clancy trial has attracted an unusual public audience. In one snapshot during the second day of testimony, roughly 28,000 people were watching simultaneously on just two YouTube streams. Newspapers and television outlets have followed the case almost day by day.

The legal question is whether Clancy was criminally responsible when she killed her three children. That is not the question of this post. To me, the trial exposes another issue deserving attention apart from the verdict:

How a very sick patient could move repeatedly through the American mental-health system without anyone clearly responsible for seeing the whole picture.

Plenty of care, little continuity

Lindsay Clancy was not shut out of healthcare: trial testimony describes psychiatrists, a psychiatric nurse practitioner, emergency care, inpatient psychiatric treatment, outpatient programs and a changing list of medications.

One provider acknowledged that she did not have complete information from Clancy’s other clinicians and had not consulted another psychiatrist or the hospitals involved in her care. A specialized program raised concerns about a medication, yet the testimony suggests that this did not become part of a coordinated treatment plan. Records introduced at trial also show Clancy searching online for psychiatric medications, symptoms and possible explanations for what she was experiencing.

None of this proves that a particular provider made an error in judgment. It does suggest, however, that despite her serious condition, no medical professional was responsible for integrating the different pieces of her care into a single treatment picture. In fact, the patient herself appears to have taken that role. That may work when someone is healthy enough to conduct research, understand the possible side effects, keep records, question prescriptions and reconcile conflicting advice. But this should not take place with a patient who is severely depressed, sleep-deprived, frightened or losing confidence in her own mind.

Subscribe now

The missing function

American healthcare is very good at specialization. We have experts for increasingly narrow problems. What we have built less reliably is the integrating function around those experts. U.S. federal agencies have been saying this for years. AHRQ (Agency for Healthcare Research and Quality) describes care coordination as especially important in a highly fragmented U.S. healthcare system, with information gaps, specialist shortages and coordination problems among the barriers. CMS (Centers for Medicare & Medicaid Services) defines fragmented care in plain language: providers do not communicate effectively around the same patient, creating confusion, medication problems, unnecessary visits and sometimes worse outcomes.

The problem becomes sharper in mental health. Behavioral health has often been separated from ordinary medical care not only clinically but administratively and financially. Private insurance contributes to this: networks determine which clinicians are available, insurance companies influence which drugs are covered. Authorization requirements can slow treatment and a change of insurance can disrupt a provider relationship.

But insurance is only part of the problem: changing who pays does not automatically answer who owns the care.

What other systems do differently

The United Kingdom makes the coordinating role more explicit: patients are registered with a GP (general practitioner) who is normally the first point of contact and the gateway to most specialist care. UK’s National Health Service (NHS) shares records to make information such as current medications and allergies available across organizations. The Dutch system is even more relevant to the American debate. The Netherlands do not have a conventional single-payer system: people buy mandatory standard coverage from competing insurers. Yet more than 95 percent of Dutch residents register with a GP practice, and the OECD rates the Netherlands strongly on primary-care gatekeeping and continuity. Specialists operate inside a structure where patients maintain their longitudinal GP relationship.

Mexico offers a less obvious lesson: its health system is divided between private and government payers (IMSS, ISSSTE, IMSS-Bienestar and others) serving different populations. Within IMSS (Mexican Social Security Institute), patients are assigned to a Family Medicine Unit which serves as an ongoing point of care that coordinates referrals to and from specialists.. Mexico’s current reforms are also trying to connect previously separate public systems and strengthen first-level care. This is worth noticing: Americans often assume healthcare lessons can only come from countries richer than Mexico. Institutional design does not follow that hierarchy.

Share

Access is not enough

The American healthcare debate usually begins with insurance: Who is covered? What does it cost? Should the payer be public or private? Those are important questions. The Clancy trial points toward another one: What happens after you get through the door?

For a patient with a changing psychiatric condition, access to five disconnected providers is likely not better than access to one coordinated team. More prescriptions do not create a treatment strategy. More appointments do not create continuity. An electronic record does not help if the people making decisions do not see the relevant parts of it.

The National Academies defines high-quality primary care around sustained relationships, whole-person care and accountability across settings. In a fragmented system, that basic standard is difficult to achieve. The teaching lesson from the Clancy trial, so far, should not be that one medication caused a tragedy, that one clinician failed, or that another country has solved healthcare. The evidence to-date does not justify those conclusions.

The real question, I believe, is: When a patient’s condition is changing rapidly and several professionals are treating pieces of it, who is responsible for seeing the whole patient? If the answer is the patient, the system has assigned its most important coordinating job to the person who may be least able to perform it.

Leave a comment

***

The clearest midterm signal this week was the staying power of a familiar concern: voters are still judging politics through the cost of everyday life. The Iran war affects gasoline. Gasoline affects household budgets. Household budgets affect confidence in the president. That confidence then follows Republican candidates into House and Senate races.

Costs Still Shape the Political Climate

July inflation eased slightly. Consumer prices were up 3.4% from a year earlier, compared with 3.5% in June. That is a slight move in the right direction economically, but the political problem persists. Voters experience prices through grocery bills, rent, insurance premiums, and gasoline purchases, not through the inflation rate.

A new Financial Times/Focaldata poll found that 53% of registered voters said they were financially worse off than when President Trump returned to office, while 55% disapproved of his presidency. Reuters reported from competitive House districts that voters are frustrated over grocery and household costs. (Reuters; Financial Times)

Subscribe now

Foreign policy is becoming a household issue

The Iran war is increasingly important because of the way the news reach voters. According to Reuters, this week, two-thirds of Americans want the war to end. At the same time, President Trump told voters that paying somewhat more for gasoline was worth the cost of confronting Iran (Reuters.) A conflict abroad becomes a domestic political issue when it changes prices, jobs, or confidence in the government’s ability to control events.

Primaries send a complex signal

Primaries held August 11 in Connecticut, Minnesota, Vermont and Wisconsin, plus plus several rescheduled Alabama House primaries, gave us a warning against easy conclusions:

• In Minnesota, progressive Lt. Gov. Peggy Flanagan won the Democratic Senate nomination over the more moderate Rep. Angie Craig.

• In Wisconsin, moderate David Crowley defeated democratic socialist Francesca Hong for the Democratic gubernatorial nomination.

• Trump-backed Mike Lindell lost the Republican governor’s primary in Minnesota,

• while Trump-backed Darline Graham advanced to a Senate runoff in South Carolina. (Reuters)

The pattern is mixed. The results offer evidence for a progressive democratic advance and a moderate democratic establishment recovery, but neither trend dominates. Candidate strength, state politics, turnout, endorsements, and local issues still mattered enough to prevent a single national story from emerging.

Share

Maps and voting rules are part of the election

Two institutional stories deserve attention:

• Republicans nominated Rhett Marques to challenge Democratic Rep. Shomari Figures in Alabama’s 2nd District under a newly drawn map that is more favorable to Republicans. (Reuters)

• At the same time, a federal judge blocked the Trump administration’s attempt to impose tighter federal controls on mail voting again , and the administration asked the Supreme Court to intervene. (Reuters)

These stories point to something different: electoral rules and district maps shape whether and where public opinion is converted into congressional seats.

What our Model Shows

Three drivers are reinforcing one another: cost-of-living pressure, presidential approval, and the Iran-energy shock. That is the basis for the current national headwind for Republicans.

But conversion is not automatic: Candidate quality, district structure, turnout, and legal rules can strengthen or weaken that pressure in individual races.

What to Watch Next

The next useful signals to watch will be: gasoline prices, fresh approval and generic-ballot polling, the first Minnesota general-election polls, Supreme Court action on mail voting, and party spending in redrawn or newly competitive districts.

Will immigration begin to displace affordability as the issue voters hear most about? The DNC’s split this week over whether ICE should be reformed or abolished shows why issue attention can quickly change the coalition problem facing a party. (AP)

For now, the central signal remains simple: voters are still bringing national politics back to their kitchen table.

***

Can democratic socialism govern here? The answer depends on a distinction that runs through the whole series: entry is not conversion.

Democratic socialists have found real entry points in the American system (see our first post) . They can win primaries. They can win city races. They can organize volunteers. They can use social media to bypass older gatekeepers. They can run on affordability. They can shift the language of the Democratic Party. But governing at scale in a democracy is a different problem.

The same system that gives them openings here in the U.S. also creates hard limits: federalism, the Senate, winner-take-all elections, weak labor law, donor power, courts, local fiscal limits, fragmented administration, and low trust in government. The movement can win offices and shift the agenda but turning that into a national European-style social democracy is a much harder task.

The Entry Points

Democratic socialism has not reappeared in the United States because the country suddenly became a European parliamentary system. It has reappeared by finding workarounds inside the system that already exists.

The most important workaround is the Democratic primary. Third parties face a structural wall in the United States: Winner-take-all elections favor the two major parties. Voters worry about wasting their vote, donors hesitate, media attention concentrates on Democrats and Republicans, and ballot access is difficult. As a result, 1) insurgent movements usually fade, 2) remain protest movements, or 3) work through one of the major parties. Democratic socialists have chosen the third path.

They work through Democratic primaries, especially where the electorate is favorable, turnout is low, and a volunteer network can make a difference. Best targets: dense urban districts, city council and state legislative races, and safe Democratic congressional seats. This is the American workaround.

A candidate can take a concrete local issue such as rent, transit, utilities, childcare, policing, healthcare, labor rights, and connect it to a larger story about economic power. Volunteers multiply the candidate’s reach. Social media converts local messages into national attention. A primary can become a referendum on the party’s direction.

Our research on modern primaries helps explain how this matters: congressional primaries have become more ideological and factional over time. They are no longer only contests between local figures, often with similar positions. Many now function as conflicts over party identity, party direction, and factional control. This creates opportunity for insurgents but also pressure on incumbents.

The insurgent may lose, but the establishment candidate will learn and change language, shift priorities, and adopt some of the challenger’s frame. This is why winning is not the only measure of influence.

Subscribe now

The Choke Points

The central main contraint is the U.S. institutional structure. The American system may allow entry, but will resist conversion, meaning turning a movement into governing change. Governing requires budget authority, legislative majorities, administrative capacity, and public trust. The U.S. system makes each of these steps difficult.

First-past-the-post elections push insurgents into one of the two major parties. The Senate overrepresents smaller states and makes national majorities harder to convert into national policy. The Electoral College does the same at the presidential level. The filibuster can block legislation even when a party holds formal power. Federalism divides authority across federal, state, and local governments. Courts can slow, narrow, or overturn policy. Donor networks shape candidate viability. Party committees and incumbents protect their own. Local governments often carry the visible responsibility for delivery but lack the fiscal authority to solve large problems.

Then there is administrative fragmentation: U.S. housing policy is split across zoning boards, city councils, state legislatures, federal subsidies, tax credits, public housing agencies, private developers, courts, lenders, and neighborhood opposition. Healthcare is split among federal programs, state rules, private insurers, employers, hospitals, providers, and regulators. Labor power depends on federal law, union density, employer behavior, court doctrine, and workplace-level organizing.

These are not simple targets: A democratic-socialist candidate can run on housing affordability, but has to worry about land use, construction costs, neighborhood opposition, financing, permits, tenant protections, public subsidies, homelessness services, and the fiscal limits of local government. Same with healthcare, inequality, etc. Converting from campaigning to governing is not an easy task!!!

The Europe Comparison Trap

European social democracy may be a useful as a reference point. It shows that high-tax universalism can exist and that countries can build broader welfare states, stronger labor protections, universal healthcare systems, family benefits, and more distributive tax structures. It shows that market economies can coexist with large public systems. It is not clear whether the same route is available in the United States.

The U.S. differs from Europe in tax structure, labor institutions, party systems, federal design, racial and regional political history, healthcare structure, welfare-state legitimacy, administrative capacity, and trust in government.

A paper by Alesina, Glaeser, and Sacerdote for the Brookings Institution makes the point that Europe redistributes more than the United States, meaning that European governments intervene more strongly in the distribution of economic resources after market incomes have been generated. The United States has stronger constitutional constraints on redistribution. Courts have historically played a larger role in limiting or shaping redistributive policy. Federalism decentralizes authority and encourages tax competition across jurisdictions. A local government that taxes too aggressively may lose residents, capital, or business support.

The United States also has a different welfare-state culture. Many Americans are more skeptical of government, more likely to explain poverty through individual effort rather than structural conditions. Race has repeatedly shaped public attitudes toward redistribution.

Share

The Party Problem

DSA has another constraint: it is not formally a party. It is an organization, a movement, an activist network, it endorses candidates, and also a place where different strategic currents compete, creating internal tension. A 2025 book by David Duhalde shows that American socialist organizations have long maintained different tendencies. The old Socialist Party included labor organizers, electoral socialists, radicals, reformers, immigrants, intellectuals, municipal pragmatists, and factional blocs. DSA has its own version of that pattern. The difference is that DSA is trying to operate inside the Democratic Party while also debating whether and how to become something more independent.

The Governing Tests

The real test is whether democratic socialism can convert electoral success into competent, durable government under American institutional constraints.

Test 1: Can it govern scarcity?

Governing means choosing among competing projects given limited money, land, labor, and political support. Housing, transit, and healthcare all require tradeoffs that campaign promises can avoid but governments cannot.

Test 2: Can it build administrative capacity?

Democratic socialists must show they can manage agencies, budgets, procurement, implementation, and the fragmented machinery of government.

Test 3: Can it broaden its base?

Democratic socialism’s strongest support is seems concentrated among younger, educated, urban voters. To grow, the movement must reach a much broader working-class electorate, including voters outside progressive cities and activist networks.

Test 4: Can it manage candidate accountability?

In a democracy, elected officials must compromise, to which movements like DSA will sometimes object. The test is whether democratic socialists can balance candidate autonomy, organizational discipline, and responsibility to voters without turning every disagreement into a rupture.

Test 5: Can it survive backlash?

Greater success will bring stronger opposition from incumbents, donors, business interests, and political opponents. The test is whether the movement has enough organization, public support, and visible governing results to withstand it.

What to Watch

• Let’s see if DSA-backed officials can deliver local results. The movement needs proof points: housing wins, labor wins, transit improvements, tenant protections, service delivery, or budget outcomes that voters can understand.

• Let’s see whether labor organizing reaches ordinary workers or remains an activist layer. A movement that speaks constantly about class but cannot build a durable worker constituency will remain constrained.

• Watch whether established Democrats “borrow” progressive economic language without doing much about it. This is a common way institutions absorb insurgency.

• Watch whether insurgents can win beyond favorable urban primary terrain, whether the “model travels.”

• The more it wins, the more slogans become insufficient. Watch whether the movement develops the policy, budgeting, and implementation **capability** needed to govern.

• Watch whether Europe remains a useful reference rather than a shortcut. It shows what is possible, but not what the American system will permit.

What Comes Next

Democratic socialism has found ways into the American political system. The harder question is whether electoral gains can be converted into durable governing results inside institutions designed to slow, divide, and constrain large-scale change.

The issues identified above provide a practical way to judge that progress. In a couple of weeks, we will return to each one, look at what has happened since, and assess whether the movement is showing signs of overcoming—or running into—the constraints described here.

Leave a comment

Source note

This post draws on the democratic-socialism research folder, especially David Duhalde’s historical comparison of the Socialist Party and DSA, Mike Cowburn’s work on party transformation in congressional primaries, Brookings material on congressional candidates and primaries, and the Europe/U.S. comparison materials including Alesina, Glaeser, and Sacerdote on why the United States does not have a European-style welfare state.

***

Post 4 of “The Middle East as a System,” a series on how local crises keep becoming regional ones.

In a Syrian town this past year, the flag over the government building changed. The building behind it did not. New authorities took the ministries in Damascus, and a transition began. But in much of the country, control of the street still rests with those who have the guns, the payroll, and the checkpoint—not with officials in a ministry. That is the question for this post: What happens when armed groups become more capable than the institutions meant to govern them?

My answer is that you get a state that is real but partial, one armed actor among several, and a transition that can move toward a nation or toward a permanent patchwork. Like Lebanon, for instance.

Subscribe now

Fragile does not mean empty

“Fragile state” is a misleading phrase. It suggests an empty space waiting for order. The reality is the opposite: A fragile state is crowded. Authority exists, but fragmented, unevenly distributed, contested, and often dependent on outside money and weapons. Armed groups grow where several conditions overlap: State security is weak or applied selectively, communities do not trust the state to protect them, political inclusion is narrow, and foreign sponsors supply cash and arms. In fragile states, war creates economies that pay better than peace. And over time the armed group stops being only a fighter and becomes a provider (Gaza.)

The political map is not the social map

A refrain throughout this series… This is where that recurring theme is clearest: the borders are recent; the societies are old.

Kurdish populations span Turkey, Syria, Iraq, and Iran. Druze communities span Syria, Lebanon, Israel, and Jordan. Alawite, Sunni, Shia, Christian, Yazidi, Bedouin, and tribal networks cross the same lines. When a community distrusts the central state, it looks for a protector. Sometimes that is a local militia. Sometimes it is a foreign government presenting itself as the guardian of a related or threatened group.

That protection cuts both ways: It can prevent a massacre, but also entrench fragmentation by giving outside actors a stake in it. Community identity does not determine political loyalty, but it helps explain why local disputes attract regional sponsors.

How Syria got here

A short chronology, not a full history. The rest is in the reading list.

Protests in 2011, met with state violence. Escalation into civil war. A splintering of the opposition into many factions. The territorial rise of the Islamic State. Interventions by the United States, Russia, Iran, Turkey, and Israel, each pursuing a different aim. A Kurdish-led force holding the northeast in partnership with Washington. Years of frozen stalemate. Then, in late 2024, a rapid offensive that toppled the Assad government after more than five decades of family rule. A transitional authority under Ahmed al-Sharaa, a constitutional declaration in early 2025 setting a multi-year transition, and an interim cabinet drawn heavily from the victorious faction. Through 2025, sectarian violence on the coast and in the south, foreign troops still present, and the open question that defines the whole period: whether the many armed factions become one army, or stay many. Notice the shape: Not a war ending and a state beginning, but a change of who sits in the capital while the underlying problem, fragmented force, remains.

Share

The armed group as a parallel institution

Here is why “disarm the militias” is a slogan, not a plan.

A single armed group can be, at the same time, a military force, a political party, a local employer, a welfare and social-service network, a tax collector at its checkpoints, a criminal enterprise, the expression of a community’s fear, and the client of a foreign state. Analysts call these hybrid actors, because they work through the state, alongside it, and against it, often all at once.

Take away the weapons and you have also taken away someone’s salary, someone’s clinic, someone’s only protection against the neighbors they fear. That is why demobilization is so hard: You are not confiscating rifles, you are dismantling a parallel institution that people depend on, and you have to replace every function it performed at the same time.

The transition dilemma

A new authority in a country like this faces priorities that pull against each other: control quickly, avoid reigniting the war, include former enemies, protect minorities, exclude the worst offenders, keep the lights on and the bread coming. Win sanctions relief and reconstruction money, and prevent foreign backers from capturing the transition for their own ends. These do not fit together neatly.

Move too fast toward central control and you tell every armed minority that the new state is just the old domination under a new flag, and they take up arms again. Move too slowly and you let the wartime map harden into a permanent partition, with each faction keeping its turf, its revenue, and its patron.

The balancing point between the two is legitimacy. Legitimacy has to be earned function by function, and can disappear the first time a promised protection fails.

The economy of staying fragile

Fragility is not only a security condition. It is also a business model.

Fragility persists through checkpoints and informal taxation, smuggling, arms markets, sanctions evasion, patronage jobs, diverted aid, narcotics, and control of the things everyone needs: a border crossing, an oil field, a water source, a port. Peace threatens all of it. The commander who taxes a highway, the smuggler who moves fuel, the official who skims the aid, the young man whose only job is carrying a rifle, each of them stands to lose from a functioning state. That is the quiet reason so many transitions stall. A real peace creates losers, and the losers are armed.

What war can and cannot do

I write this series as an American citizen opposed to armed conflict, not because force is always avoidable, but because it rarely resolves the political conditions that produced the fight. The same standard applies here, to every actor, including my own government.

Military victory can topple a regime, clear a city, or break a faction. It cannot manufacture a functioning ministry, a trusted police force, or a shared sense of nationhood. Removing a dictator is not the same as building a state. And foreign intervention, even when it saves a community, tends to add one more sponsor with one more stake in the outcome, which can deepen the fragmentation it was meant to relieve. Force can change who holds the capital. Whether that change becomes a country depends on the slow institutional work that begins after the shooting stops.

Strategic alternatives

Just alternatives, no single answer. Several, each with a cost.

The table has no winner on purpose. Each path pays a real price, and a plan that hides its price is how transitions fail.

Indicators to watch

A few signals worth more than the daily bulletins:

• Whether armed units answer to a single chain of command;

• How minority communities are treated and represented;

• Who controls local revenue and security;

• Whether refugees and displaced people actually return;

• Whether reconstruction money arrives with transparency; and

• Whether foreign forces draw down or dig in.

The honest counterpoint

The strongest objection is that this analysis can let armed groups off too easily. Calling a group a clinic, payroll, or “hybrid actor” can make violence sound more legitimate than it is. Some groups are simply predatory. And the analysis can also sound too fatalistic, as if nothing can be changed.

I take the point. Explaining why a militia survives is not defending it. Some are more predator than provider. But if an armed group endures because it provides security, income, or services, then a transition has to replace those functions—not just remove the guns. That is difficult, but it is a clearer path than a slogan.

Over to you

If you were designing Syria’s transition, or Lebanon’s, Gaza’s, or Yemen’s, which would you risk first: the speed that reassures no one, or the patience that allows divisions to become permanent? And is there any foreign guarantor you would actually trust not to make the fragmentation worse?

Leave a comment

Next in the series: the resource system, and how water scarcity quietly becomes an electricity, food, fiscal, and political problem all at once.

Reading list

• Syria: Transition and U.S. Policy (Congressional Research Service). The institutional anchor, with a conflict synopsis and a transition chronology. Read its dates as a moving target.

• The Axis of Resistance backgrounder (Middle East Institute). On the armed networks that cross these borders and the foreign sponsorship behind them.

• After Assad: Navigating Syria’s Humanitarian Crossroads (Humanitarian Advisory Group / UKHIH). A field-level view of the transition, closer to the ground than the policy capitals.

• Global Terrorism Index 2026: The Iran War and the Global Terrorism Threat (Institute for Economics and Peace). Included because it reads the same armed landscape through a different frame, and will not always agree with the sources above.

***

For today’s interlude, I chose the Nitty Gritty Dirt Band’s original 1994 recording of “The Broken Road,” the closing track on the album Acoustic. Jeff Hanna wrote the song with Marcus Hummon and Bobby Boyd. Hanna later explained that it grew from a conversation about the circuitous routes people take through life: periods that seem disastrous at the time but eventually lead toward a better relationship, spiritual path, career, or direction. The band recorded the song in Colorado in the late winter of 1994. (The Boot)

The song then followed a rather broken road of its own: A decade later, Rascal Flatts recorded it as “Bless the Broken Road.” Their version is the one most people know: polished, more emotional, with a strong sense of arrival. It spent five weeks at No. 1 on Billboard’s country chart and brought the songwriters the 2006 Grammy Award for Best Country Song. (Billboard)

I prefer the original: Rascal Flatts makes the song sound like a declaration: every wrong turn brought me here. The Dirt Band makes it sound more like a discovery: perhaps the disorder had a direction after all. The earlier performance feels quieter, less settled. It holds gratitude and regret at the same time. The broken road has not somehow become smooth, only the traveler’s understanding of it has changed.

Subscribe now

A Song From a Divided Year

The son<⁷and g was written and first recorded in 1994, when the United States and the wider world were also trying to decide what kind of road they were traveling.

NAFTA took effect on in⁷⁷1, deepening the economic integration of the United States, Canada, and Mexico. By November, Republicans had won control of both houses of Congress for the first time in 40 years, marking a significant realignment in American politics. The country was becoming more economically integrated with its neighbors while becoming more sharply divided within its own political system. (United States Trade Representative)

Elsewhere, South Africa held its first democratic election and Nelson Mandela became president, a new political beginning. At almost the same time, the genocide against the Tutsi unfolded in Rwanda, while civilians in Bosnia remained trapped in war. The early post–Cold War world contained openings and failures at the same time. (Nelson Mandela Foundation)

The song was not written about politics or history. It was originally meant as a love song, and it should remain one. But songs pick up new meaning from when they’re heard in. In 1994, “The Broken Road” offered a modest form of hope: A life can still develop direction after its original map has failed. That is what I hear in this first rendition. Not certainty, not destiny: A person looking backward, finding a pattern, and accepting that the road could only be understood after it had been traveled.

A quiet four minutes for the roads we understand only in retrospect…

Subscribe for more interludes and posts…

About the banner illustrations

The banner illustrations were developed by AI in the style of Jean-Michel Basquiat. Jean-Michel Basquiat (1960–1988) emerged from late-1970s New York and developed a visual language dense with words, crowns, anatomical fragments, numbers, diagrams, and crossed-out phrases. He treated text as meaning and mark, allowing contradiction, revision, and historical pressure to remain visible on the surface.

That grammar fits “Bless the Broken Road”: the route is not cleaned into a straight line, and the earlier wounds are not erased. They remain the evidence from which the traveler constructs a later understanding. The two banners are original AI-generated compositions inspired by this visual language, not reproductions of Basquiat works.

Explore his work

***

In July 1944, the war in Europe was not over. American factories were still producing aircraft, vehicles, weapons, and supplies at a scale the Axis powers could not match. Yet, representatives of the Allied countries were meeting at Bretton Woods to design the monetary and development institutions of the postwar economy. American power was doing two jobs at once:

• It was producing victory.

• It was also helping to design the system that would follow it.

That distinction matters because a large economy is not automatically a superpower. A large military is not enough either. A superpower must be able to convert resources into production, production into technology and finance, and those capabilities into military reach, institutions, alliances, and influence. That is the question behind this series: Can the United States keep that conversion system working?

Before looking at China, new alliances, American retrenchment, or the condition of U.S. democracy and education, we need to understand how the United States acquired the position in the first place—and how earlier global powers lost theirs.

A superpower is a system

A system approach would not surprise the regular readers of my Substack…

The word superpower belongs mainly to the twentieth century. Portugal was not a superpower in the modern sense, but it was an early global maritime and imperial power. Britain later combined industrial, financial, naval, and imperial power on a much larger scale. The United States then became the first power able to operate globally across nearly every major domain at once. These cases are not identical. That is precisely why they are useful. A country’s position rests on several connected components:

geography and resources → production → science and education → finance → military and diplomatic reach → institutions and alliances

The arrows matter as much as the boxes. Oil, minerals, money, universities, factories, and ships do not organize themselves. Governments, firms, workers, financiers, scientists, and allies must connect them. Power grows when the connections reinforce one another. It weakens when an asset bypasses the productive system, when commitments exceed resources, or when institutions can no longer coordinate action.

Subscribe now

The American system existed before 1945

The United States did not become powerful simply because Europe destroyed itself in two world wars. Before 1945, the United States already possessed continental scale, extensive natural resources, a large internal market, growing industrial capacity, deep financial markets, immigration, universities, and relative geographic security. War mobilization then connected these assets at a different level of intensity.

One important link was public research. During World War II, the Office of Scientific Research and Development financed one of the largest applied-research efforts in American history. Daniel Gross and Bhaven Sampat (see reading list) find that this funding helped create regional technology clusters, high-technology firms, and employment that persisted for decades. It also redirected the country’s broader innovation system toward the technologies supported during the war (American Economic Review.)

The war did not merely produce weapons. It produced organizational relationships among government, universities, laboratories, manufacturers, and skilled labor. Those relationships survived the emergency. The resulting loop looked roughly like this:

public demand → applied research → industrial production → specialized firms and workers → further innovation

This is a recurring feature of power. The durable advantage is not the individual invention. It is the ecosystem that can keep producing, improving, financing, and deploying inventions.

Share

Victory was not enough

The United States could have emerged from World War II as an unusually rich and heavily armed country. Instead, it converted material predominance into system leadership.

At Bretton Woods, the United States and its partners created the International Monetary Fund and the institution now known as the World Bank. The postwar trade negotiations produced the General Agreement on Tariffs and Trade. These arrangements made currencies, reconstruction finance, trade rules, and access to markets part of a shared institutional structure in which the dollar and the United States occupied central positions. (U.S. Department of State, Office of the Historian)

The Marshall Plan added another layer. It did not simply transfer American goods to damaged countries. It combined grants, European planning, production targets, trade, infrastructure, policy reform, and new forms of cooperation. More than 90 percent of the assistance took the form of grants, but European governments had to participate in collective planning and implementation (Congressional Research Service.) It was a strategic investment in reconstruction. A stronger Western Europe became a market, an industrial partner, a political coalition, and a source of legitimacy. American power was multiplied by rebuilding capable partners rather than keeping them permanently weak. That is one reason alliances and institutions cannot be added at the end of a national-power calculation. They are part of the productive system itself as I have shown in my series of posts on globalization (Post 1, Post 2, Post 3, Post 4) starting with this Post)

Portugal: wealth that bypassed production

Portugal illustrates a different pathway: The discovery of Brazilian gold in the eighteenth century delivered a large external windfall. In the short term, it raised incomes and purchasing power. But a 2025 economic-history study by Davis Kedrosky and Nuno Palma argues that the gold also produced an economic and political resource curse (The Journal of Economic History.)

The inflow made imports easier to buy and weakened domestic tradable production. Resources moved away from manufacturing and agriculture. The crown also became less dependent on bargaining with domestic taxpayers, reducing pressure for institutional checks and state development. The study’s counterfactual estimate suggests that Portugal’s GDP per person was about 40 percent lower by 1800 than it might have been without the Brazilian-gold endowment. The gold paid bills but did not create a self-renewing productive system. That is the systems lesson.

Britain: decline without disappearance

Britain’s path was almost the reverse. It had developed an industrial base, a financial system, a navy, a global commercial network, and an empire. Its problem was not a windfall that bypassed production. It was that the cost of maintaining the system rose while the relative capacity of other states, especially the United States, grew faster.

British decline occurred through war costs, decolonization, currency pressure, changing trade relationships, and the transfer of leadership to the United States. Britain’s financial retreat was gradual. Between 1968 and 1974, it reached agreements with countries holding large reserves in British pounds to prevent a sudden selloff, but without a clear plan for ending sterling’s international role (The Economic History Review.)

Policy quality also mattered. David Ronicle’s reappraisal of interwar Britain finds that monetary and fiscal choices shaped the country’s two interwar recessions (Bank of England.) Britain lost dominance, but did not disappear as a consequential power, operating primarily inside a system led by the United States.

Three different mechanisms

The comparison is easier to see as three causal paths:

There’s no expiration date on leadership. There’s only one standing requirement: advantages must reproduce themselves:

• Factories must be renewed. Scientific discoveries must move into use.

• Education must produce capable workers and citizens.

• Financial strength must fund productive investment rather than only asset appreciation.

• Alliances must remain credible. Institutions must still solve problems for participants.

• Military commitments must remain connected to attainable political purposes.

Globalization is part of the power system

In my earlier posts on globalization, I treated globalization not simply as trade, but as infrastructure, dependency, chokepoints, and feedback loops. Same lens here: reserve currencies, payment systems, ports, shipping lanes, energy networks, chip supply chains, standards, universities, bases, and alliances aren’t decorations around national power. This is how power moves.

A country can possess large resources and still lose influence if those channels weaken. It can also command more influence than its domestic resources alone would suggest if it occupies central positions in networks that others continue to use. The United States became a superpower because it built domestic capacity and then embedded that capacity in an international system. Whether it remains one will depend on both sides of that equation.

Subscribe now

What history does—and does not—tell us

Portugal shows that wealth can weaken capacity when it substitutes for production and institutional development. Britain shows that relative decline can be managed, but rarely according to a clean master plan. The United States shows that power can be multiplied when research, production, finance, reconstruction, institutions, and alliances reinforce one another. History does not tell us that American decline is inevitable. It tells us that no advantage maintains itself.

The next post in this series will turn to the present: China’s rise as an economic, technological, military, and strategic power; the European Union’s attempt to turn scale into greater strategic autonomy; established and emerging groupings such as NATO, AUKUS, BRICS, and the Shanghai Cooperation Organisation; and the newly announced Saudi–Turkey–Pakistan defense pact.

The world is not waiting for the United States to decide what role it wants.

Questions for the reader

Which component of American power is now most important: the dollar, industrial production, science and education, military reach, or alliances? Can the United States reduce some commitments without weakening the network that magnifies its power?

Leave a comment

Further reading

• Davis Kedrosky and Nuno Palma, “The Cross of Gold: Brazilian Treasure and the Decline of Portugal”.

• Alan de Bromhead, David Jordan, Francis Kennedy, and Jack Seddon, “Managed Decline: Muddling Through with the Sterling (dis)Agreements, 1968–74”.

• David Ronicle, “Macroeconomic Policymaking in Interwar Britain: A Reappraisal”.

• Daniel P. Gross and Bhaven N. Sampat, “America, Jump-Started: World War II R&D and the Takeoff of the U.S. Innovation System”.

• U.S. Department of State, Office of the Historian, “Bretton Woods-GATT, 1941–1947”.

• Curt Tarnoff, Congressional Research Service, The Marshall Plan: Design, Accomplishments, and Significance.

About the illustrations

This post’s illustrations are inspired by Diego Rivera. Rivera organized workers, machines, managers, scientists, raw materials, landscapes, and political symbols into continuous mural systems. In the Detroit Industry cycle, he used fresco to connect Detroit’s geological, technological, and human history while placing the productive and destructive uses of industry beside one another.

Rivera’s murals treat modern power as a social process rather than as a collection of isolated machines. Pan American Unity joins ancient American traditions, modern industry, artists, inventors, and public life in one panoramic narrative. That visual language fits this series: resources matter, but power emerges from the links among production, knowledge, institutions, and people. (Detroit Institute of Arts; SFMOMA).

Subscribe now

***