Illustration by Althea Xiu

When ChatGPT first appeared in November 2022, many people thought writers would rebel. How could they compete with a tool that could generate seemingly credible articles on nearly any topic, complete with structure, transitions, and references? My own editorial work doing reports for management consulting firms dried up within a few months. They didn’t need writers when they had bots.

But within a couple of years, many writers have made their peace with GenAI. They have come to appreciate the help and convenience of the tools. The more significant rebellion came from readers. They’re the ones who feel threatened. It took only 18 months after the release of ChatGPT for the term “AI slop” to enter the mainstream. That label is a reader’s complaint.

Studies have shown that writers who rely on AI systems have diminished neural engagement. The same may well be true for readers. We know when we’re reading something written by a bot. We can sense our synapses fritzing, and it feels uncomfortable. We gravitate instead toward prose that reflects real connection with another person’s mind. The advent of GenAI seems to have whetted appetites for human authorship: for stories and essays that engage us, created by minds that care.

The method I describe here is for people who have good ideas, but don’t find it easy to write—especially in the competitive environment of an AI-enabled mediasphere. They want to get the writing done quickly and easily, but they also want a connection to readers that you can’t get from an LLM. I developed this method gradually, over 30 years as an editor of business publications, helping engineers and managers become thought leaders. Then I revised it after extensive experience with GenAI, using the tools to streamline the tedious parts of the process without losing the writer’s cognitive commitment.

I know there are hundreds of writing methods, and some of them probably disagree. This method works consistently for me and the people I coach and edit. It works because it removes some of the hurdles in your mind, but it keeps your mind in the lead.

It’s a seven-step recipe. You gather material, write a “curtain-raiser” (opening), create a core passage that expresses your idea directly, figure out a closing, choose a plot that gets you to your closing, sort everything into the plot, and then write and refine. The sequence feels counterintuitive at first, because the order in which you create the article is not the same order in which readers experience it. But if you were a highway engineer, you wouldn’t expect the design and building of a bridge to feel much like the experience of driving across it.

Like any recipe, this method is best learned by rote and then improved upon in practice. As you get more experienced, you transcend the need for recipes. But you won’t forget these steps; they’re simple enough to keep in mind. Your LLMs will also remember them. They can play back the jargon—curtain-raiser, core passage, closing, plot, scaffolding—in consistent ways, and because they are primed to associate combinations of words that will work well together, they can help you with the tedious aspects of this task.

Step 1: Gather and Annotate

Start by accumulating raw material: notes, observations, interview transcripts, clippings, the intellectual detritus that forms your article’s foundation. Assemble all this without attempting to organize it.

As you gather, your mind begins processing connections. Encourage this. Add commentary about what things might mean, questions that arise, and thoughts about framing controversial elements. Include fragments of transcripts, previous writings, and anything potentially relevant. Pay particular attention to the elements of any relevant human story: experiences, reactions, and expectations. What happened? How did it start? How did the relevant people feel about it? What surprised them? How did they respond? What happened next?

You may decide you need to interview people to learn more about your topic. The interviews are part of the gathering phase. Record and transcribe all of them; don’t rely on your notes, because even the best note-taker (including AI note-takers) will filter out important details. Later, you may need to check the details with the people you interviewed.

Step 1 trains your memory and improves your grasp of the whole story. Do not assume you will use everything. Much of this is scaffolding—as valuable as the scaffolding on a building, which eventually gets discarded but which is necessary for construction.

AI can accelerate Step 1. Use it to conduct comprehensive searches, saving time and ensuring thoroughness. Ask it to identify gaps in your research or suggest related topics you might have overlooked. But remember: AI is searching its training data and the web, not your brain or the reasons why you care enough to write this article. The connections you make in your mind are as important—more important—than anything AI can tell you.

Step 2: Write the Curtain-Raiser

Think of your article as a play in a theater where the audience can leave at any moment without blocking anyone’s view. Each sentence must hold them in their seats. Each paragraph must earn the next. The curtain rises, and you have perhaps 30 seconds before they decide whether to stay.

What do people see onstage when the curtain goes up? This could be an anecdote, a striking fact, a hypothetical scenario, an image that serves as a microcosm for your entire piece. It is not comprehensive—it cannot tell everything. It is a hook, and more than a hook. It lets people glimpse what the rest of the material will do for them.

Consider the opening of a book most people know: Harry Potter and the Philosopher’s Stone. A man named Dursley is the director of a drill-making company in England. He is obsessed with appearing normal, and shuns his wife’s sister. On his way to work, he thinks he sees a cat reading a map. He yells at his staff. Going home, he sees people in cloaks, gossiping about a family called Potter. That’s the curtain-raiser. There is no Hogwarts, no Voldemort, no wizards, not even Harry Potter yet. But there is a sense of intrigue, and an implicit indictment of the mundane complacency of everyday life. That’s enough to draw you in.

Daniel Kahneman’s book Thinking, Fast and Slow has a curtain-raiser about his colleague Amos Tversky, with whom he developed many of the ideas. Jim Collins opened Good to Great with a story about an executive who asked him a tough question. Amy Edmondson’s curtain-raiser for The Fearless Organization was a story about a hospital nurse who didn’t speak up when a doctor put newborn twins at risk. William Goldman started his classic Adventures in the Screen Trade by talking about the dawn of the motion picture industry, when stars of the stage were too embarrassed to use their names. Though my examples here are drawn from books, you’ll find that most chapters and compelling articles have curtain-raisers as well.

The curtain-raiser is a story, a passage, an idea, or even a slogan that you can’t stop thinking about. It doesn’t tell the whole story of your article. It draws the reader in, just as it drew you in. You can ask AI to suggest potential openings, based on your notes from Step 1, but only you know which image or catchphrase has been haunting you. Get it out of your head and onto the page.

(My curtain-raiser in this article is the first three paragraphs about readers’ skill erosion, which I think is an under-explored impact of AI on people.)

Step 3: Craft the Core Passage

If you had only two paragraphs to convey your entire message, what would they say? That is the core passage. It is not an academic précis or a promise of what you will tell readers. It is the kernel itself: what is important and why it matters.

Journalists call this the “nut graf”—the passage that explains what happened, who was involved, where and when it took place, and why anyone should care. Business writers talk about the “elevator speech,” though your purpose here is not to sell but to communicate. Readers could walk away from a good core passage thoroughly educated about your subject.

It might take a full day to write these two paragraphs. Time well spent: they anchor everything else. Without a solid core passage, you risk writing yourself into thematic cul-de-sacs with no logical escape.

You may hesitate to put the core passage up front because you worry it will give away your best stuff. But in non-fiction today, readers don’t have the patience to wait for your message to emerge. If you articulate it clearly and cogently at the front, you’ll be able to build on it all through the rest of the narrative.

AI can help here. Feed your notes and ask it to compose a nut graf about your core theme. Use the output as a starting point, a mirror to see whether your thinking is clear. But do not accept it as written. AI’s judgment is not as good as yours. Refine it until it captures exactly what you need to say.

(My core passage in this article begins with the phrase, “The method I describe here.” You could read just the three paragraphs that follow and get the gist of the article. I was tempted to leave out the steps, because I am always afraid of giving away my secrets right at the beginning. But like most articles, this one is stronger when it puts the core ideas up front.)

Step 4: Write the Closing

What song do you want the audience humming when they leave the theater? The closing represents your destination for the article. This step forces you to consciously think about how your readers might feel about your ideas. Until you write it, you do not fully know where your ideas are headed.

The closing might be a coda, a call to action, a bridge to further exploration, encouragement that readers can master what seems daunting, or a reminder to consider broader impacts. It can also be coming attractions for the next article in the series or the next chapter in the book. Choose the closing based on the emotional and intellectual resonance you seek to create. AI can suggest options, but only you know the effect you want.

(In this article, my closing is a reminder of the connection at the heart of writing.)

Step 5: Determine the Plot

Having written your curtain-raiser, core passage, and closing, you must now chart the path between them. Think of this as flying over the terrain of your material. What pattern will best carry readers from start to finish?

There are many plot structures, and I find it best to pick one at a fairly high level of abstraction. You’ll probably have a sense by now of what you need to say to connect your core passage to the closing. Here are a few plot structures that I use from time to time:

Choose a structure that best serves your material and your readers’ needs. AI can suggest organizational structures, but you are the one who understands the shape of your argument. If you’re writing a book or a series, every chapter gets its own plot, and they don’t have to be the same.

Sometimes the plot takes the form of a framework: an assortment of elements with explanatory power. For instance, the “limits to growth” system story is a framework in which accelerating changes hit a moment of equilibrium. The elements in a framework generally fit together. Either they add up to a change (the seven habits of highly effective people), or they influence one another (the interplay between human capabilities and AI capabilities), or they represent an emergent phenomenon in which one leads to another (the “crossing the chasm” framework of emerging technologies).

(The plot of this article is a prescriptive path, with seven steps.)

Step 6: Sort Everything into Your Plot

Your plot becomes an outline. Go back to all your gathered material from Step 1. Now sort it into the section where it belongs. You can use AI for this, but recognize that AI will not always put everything in the best possible order.

Be ruthless about discarding everything that does not fit. It’s like the scaffolding on a building. You needed it to construct the final article, but you do not keep it.

You may find some sections worth keeping, even if they do not fit this article. They can be applied to another piece of work.

Step 7: Write and Refine

In this final step, you are ready to write the entire article. You operate at a closer, more intensive level of detail. Since everything is sorted into sections, you can work on one section at a time rather than trying to hold the entire piece in your head. For each section, verify facts, acknowledge reader expectations, and stay true to your own integrity.

AI can help organize the material for a section, suggest transitions, or create first drafts. But you will need to work through the material yourself, one sentence at a time. As a rule of thumb: each paragraph should make one point. The sentence that sums up this point should generally come at the top of the paragraph. Each sentence should have a single thought. This rule of thumb is not universal; you’ll have many opportunities to break free of it. But be conscious when you do.

After you have finished the article, set the piece aside for at least half a day to gain distance. Then read it aloud as a whole, listening for gaps, redundancies, or places where the flow stumbles. Your ear will catch what your eye missed.

You can also give your complete draft to AI and ask: What have I left out? What gaps do you see? Use it to check facts or identify areas that need clarification. This does not substitute for reading aloud. If something does not sound like you, change it until it does.

Writing with Your Mind in Mind

Throughout this process, remember that you are working on your mind. Every piece of scaffolding you create and discard, every draft you revise, every unnecessary paragraph you cut—all of this builds your facility with words and concepts. The material you discard is not waste. It is the exercise that made your final structure possible.

Jim Collins cut 30 percent from his manuscript of Good to Great purely for quality’s sake. When he read another author’s manuscript, he recommended cutting another third. Anne Lamott reminds us that we need “shitty first drafts.” The willingness to produce imperfect work and then improve it is generally more effective than trying to get it right the first time.

This method is formulaic in structure but does not lead to formulaic results. Like any good recipe, it bootstraps you past the need for recipes, building your skills until the process becomes natural.

We are entering an era when machines will generate most of the written text we see. Much of it will be adequate. Very little of it will connect. Those of us who develop our own voices will have an advantage that compounds over time. Readers know when something connects. We know it because we can feel it in our minds. Our challenge, as writers today, is to invoke that feeling, in a way that makes a difference, without getting overwhelmed by the task.

My inboxes are flooded with articles about improving the experience of AI. It’s like being accosted by dozens of sorcerers’ apprentices, all raising questions that require trial-and-error to answer. Whose ChatGPT cheat codes are the best? Do we use plain-language prompts or slashes? How do we get large language models to stop flattering us or making us stupid? How rigorously must we check for hallucinations? Do we trust AI systems with deleting our email?

This cacophany of sudden expertise reminds me of the 1990s. The internet was just going mainstream, and people were wondering which search engines to use and which links to trust. We once again stand at a threshold of a new, life-changing technological medium. AI is already replacing search engines; it will soon be the primary medium for information and interactive entertainment.

The technology for all this — the models, servers, and infrastructure — is ready. But technology is not enough to create a truly valuable, trustworthy communications medium. It never has been. You need formats for that, and GenAI’s formats are still immature.

When the internet was going through similar growing pains, the great multimedia pioneer Ted (“Xanadu”) Nelson compared it to the “pre-director” stage of motion pictures. The movie camera was so new that only a few people knew how to use it. They became the first film directors. Through their efforts, over the next few decades, the format elements that we associate with movies emerged: Titles, long shots, close-ups, cuts, pans, dollies, and ultimately the conventions of the soundtrack.

That’s what AI is like now. The format innovation has only just begun.

You could compare today’s AI to broadcast television before the development of the half-hour time slot, the news desk, and the situation comedy. To the newspaper before the headline/subhead combination, the single lead photo, and the Who-What-When-Where-Why “nut graf.” To the internet browser before the highlighted link and the conventions for pop-up windows. To social media before memes, dance trends, and cat videos.

A Brief History of Format Innovation

Formats don’t develop the way technologies do. Technologists have hierarchies, R&D budgets, and specialized expertise. They manage innovation from the top down. Format innovators work bottom-up, through trial and error, borrowing and adapting the work of other creative people. They experiment and their successful experiments are copied until they go viral, and then they become norms.

I first saw the power of formats when I was a journalism student, researching the history of slick magazines. The invention of the rotary press in the 1880s allowed print runs of more than a few thousand copies at a time, enabling the first mass audience publications. Art directors from Germany’s Bauhaus school fled to the US and joined emergent glossy-paper magazines (the “slicks”) in the 1930s, experimenting with new ways of combining text and images, often stealing ideas and designs from each other. Photos “bleeding” to the page edge without margins. Pull-quote boxes. Three-column text pages at the back of the book (to set off the ads). It took about 60 years for the modern magazine format to evolve, and even now it still influences the design of online journals — for instance, in the short headline and longer subtitle format at the top of this article.

In the 1980s, I became an early devotee of computer conferencing systems like the CompuServe and the WELL, forerunners to today’s Reddit, X, Bluesky and Substack. Here, too, the most significant features were formats: conventions like the “you own your own words” policy on The WELL, the simulation of emotions through emoticons and acronyms, the tone of voice of user-created text and the recognition that the quality of online conversation, like the quality of a dive bar, depends on having regulars show up. The conventions of trolling are formats; the word “trolling” itself appeared on Usenet in 1992, adapted from a fishing term for running a baited hook through the water. Other formats, like the threading patterns for comments and responses, were encoded into the platform software, but always in ways that reflected the preferences and mental habits of the most loyal, dedicated participants.

A few people, like Stewart Brand on The WELL, have played the role of format godfather for the internet, in the same way that Adolph Zukor and Irving Thalberg played that role in the early-twentieth-century movie industry. They established successful media, making highly skilled use of formats, in a way that audiences responded to and others couldn’t help but copy.

Good platform designers pay attention to formats. When something gets popular, they find ways to install it as a feature. It’s like streetwear influencing couture — or like Zoom calls influencing the camera-facing, split-screen aesthetic of podcasts.

Joe Ziska, an AI executive who writes about long-wave technological development, has a theory that explains why format development lags behind hardware and software engineering. Technologies go through five stages of development, starting with scarcity: in this phase, the devices are so expensive and difficult that only specialists and devotees have the necessary access and patience. Then comes a tools phase, in which the devotees create formats; a scaling phase, in which the tech matures and rolls out to non-technical users; an abundance phase, in which it becomes a way of life; and a commodity phase, in which you have to constrain it like you would a weed. (Email is now in that phase. So are toaster ovens.)

For all its ubiquity, AI is still in the scaling phase. The creative community of users is still catching up with the technology. But it is possible to look ahead and guess at what kinds of formats the new system can and should generate.

In fact, where AI is concerned, we should be trying to predict the formats of the future, with the hope that we can influence them. For we’re going to need good formats: formats that naturally channel AI in humane, transparent, and benevolent directions. The technical parameters will determine how AI is developed, but the formats will determine the behavior of the people who use it.

The Previous Round of the Next Wave

I admit that my track record for this kind of prediction is mixed. In 2001, the Global Business Network published a report I wrote called The Next Wave of Format:

Next Wave Of Format

2.92MB ∙ PDF file

Download

Download

It was about the future of the internet. Search engines and e-commerce were still new; smartphones did not yet exist. The dot-com crash had recently happened, and many internet companies were going under.

But I was optimistic. “There is something terribly right going on,” I wrote. I made the same case I’m making here, arguing that formats would emerge as people coalesced around them, and the internet — both as a marketplace of commerce and a marketplace of ideas — would develop its prevalent behaviors based on the formats people preferred.

That part of the prediction has held up. The conventions that organize the web today are formats in exactly the sense I meant. The “like,” the retweet, the thread, the story, the scrolling feed, the quote-post: each one took its cue from users. Each one reduces the number of decisions a user has to make, and each is now so embedded in our expectations that any platform without them feels broken.

But my background misled me. I have a masters in journalism from UC Berkeley and a career steeped in the practices of organizational learning (which is highly influenced by conflict resolution). I predicted formats that would emphasize clarity, objectivity, and “just-the-facts” reporting, like the mainstream news media of the 1970s. The internet would distinguish paid content from editorial. It would bring together opposing points of view in balanced discussions where people revealed the assumptions underlying their perspectives and found common ground.

I thought popular media would evolve into something like a Studs Terkel oral history combined with a truth-and-reconciliation commission report. The closest we have is Wikipedia, where volunteer editors practice “productive friction,” comparing perspectives. The audience for this type of colloquy, however, is relatively small.

I also assumed new formats would be developed by the kind of professionals who had built formats before: content shops, driven by craft and curiosity. I thought there would be many more curators; I forgot that curation, like book reviews, is often more boring than the original work. I dismissed advertising as an outmoded business model, to be replaced by micropayments. But advertising did not die. It became the engine of platform support. The inclusive reporter I imagined, patiently weaving perspectives together, was not elevated by the new medium. The new medium routed around him and chose the partisan pundit instead.

I understood the concept of “enshittification,” though Cory Doctorow hadn’t yet coined the word. I knew that commercial interests tended to undermine the quality of their offerings over time; I’d seen it happen in manufacturing. But I assumed, just as mainstream journalism had overcome the yellow sensationalism of the 1890s, that quality would rise to the top. I didn’t realize that the “objectivity” of mainstream media was just another format, and that yellow journalism would make a comeback because people liked it.

But what I did get right was the decentralization of format opportunity. Everybody may not be famous for 15 minutes, but everybody can be the designer of their own media space, adapting social media to their own purposes. Ideas and artwork have always been plentiful, but it’s never been as easy to publish them. As Steve Piersanti, founding editor of the Berrett-Koehler publishing firm, points out, there are 15 times as many book titles published each year today as there were in 2005 — but the total number of books sold is the same.

Those who want to reach an audience (and make a livelihood) are compensating by using AI. We are coming to realize that AI systems are not our competition. AI systems are the brokers. The language models, which assemble words and pixels based on the probability that someone will find them valuable, are already arbitrating whose voices are heard.

We’ll need better AI formats to make sense of all this. Of course the tech companies still control the technology, but there’s a limit to what technologists can dictate when it comes to human behavior. Unlike social media, GenAI systems are diverse and interconnected; it seems unlikely that any single one will dominate. Instead, the language models will compete for popularity, and those that incorporate the formats people prefer will tend to capture larger audiences.

Because GenAI plays other roles than media, its format innovation will be complicated. As B2B and military uses evolve, formats will trickle out to civilian use five or ten years later, like the word “fubar” after World War II. Large language models and their complementary tools (including sensors, personal robots, and data analytics) will also create vast new opportunities for format innovation in other media. I can already see this when I create a Substack post or a website. Claude Cowork is sitting browserside, helping me figure out how to tweak the settings.

The Formats That Will Shape AI

So let’s talk about the formats that will emerge to use GenAI effectively. We can’t predict with certainty which will prevail, but we do have some clues. There will be at least three categories of formats for language models. Some will make control of the systems easier, especially for people needing sophisticated responses. Some will try to earn our trust. And some, inevitably, will be used to extract revenue.

MAKING CONTROL EASIER

The basic LLM format of prompted conversation box has already become second nature. The chat interplay — your prompt, the model’s reply, the thread that accumulates, including the one-line ruminations that simulate thinking to show the model is active — is a format. It spread the way formats spread: meme-like, self-evident, copied across every product within months, with no one trained to use it.

But this format is still very young, and it has some evolving to do. Prompt engineering is good for one-time interactions, but it gets tedious over time. And public LLMs do not automatically learn from experience, except through add-ons like skill files and handoff scripts. That’s a design feature; LLMs that learn from experience are too easy to co-opt. Therefore, an LLM has personality, but not much of one, and not consistently.

Companies, and individuals, are rapidly developing workaround formats to help with efficiency. There are endless lists of shortcut prompts like /STEELMAN (to ask for an opposing argument) or /CTA (three calls for action). But these are futile efforts, in my opinion, to overlay the systematic formats of code on top of the chaotic, probabilistic, ambiguity of natural language and broad training data. Designing coder-style formats for language models is like trying to set up commercial breaks for an oracle.

There are already emergent drop-down menus that adjust on the fly based on guessing your intent; or personalities that show up depending on clues about the user’s mood. These will continue to become more sophisticated. Formats will go beyond providing different chatbot personalities. They’ll filter information and perspective, so that bringing up a bot’s avator will be like adopting a new way of looking at the world. The LLM will be like a schoolhouse in which we call forth different digital tutors to guide us. Sometimes these will be provided by the LLM itself, but it’s likely that third-party plug-in avatars will proliferate, coming to life when we request them, and continually learning from their interactions with their favorite people.

We’ll need formats, in the form of visual or auditory cues, to tell us when we’re talking to the base model and when we’re talking to the complement. We’ll also need formats for helping us design and refine our own complementary guides, and versions of ourselves to interact with others. We’ll need a wide variety of formats for dealing with errors, because people without coding experience will be generating their own software, and many errors will follow.

EARNING OUR TRUST

The infrastructure for gaining trust is far more difficult to create, and needed more than ever. Suppose I create an avatar that represents me and answers routine email. How do people know if they’re corresponding with the “real” me? Do I elect to keep that mysterious? If so, I risk losing others’ confidence. I’ll need an equivalent to a political campaign ad’s closing format: “My name is Art Kleiner, and I approve this bot’s message.”

Since provenance is more difficult to guarantee than ever, there will be formats for identifying origins — ways to see the sponsorship, ownership, and authorship of what you were reading. An AI answer arrives as fluent, sourceless prose, and the work of discovering where any piece of it came from falls entirely on the reader. There is already evidence that this laissez-faire approach to verification is taking a psychological toll on people. In a world of deepfakes, misinformation, and reporting out of context, people yearn to know what they can believe, and how much of that believability they can rely on. Even sites that clarify their process for resolving contentious disputes, like Wikipedia, are continually accused of bias.

As economist Herbert Stein said, when something cannot continue forever, it will stop. For that reason, I believe there will be a solution to the provenance problem. This will be more format solutions: ways of structuring text and images so people can click to the identifiable source. Wikipedia pioneered this with its voluntary editorial agreements, and other systems will probably follow.

Provenance is a mechanism for building trust. But trust doesn’t just happen; it must be earned, through consistency, transparency, and reliability. Those who aren’t in the coalition must be able to recognize the format signals and count on them, just as we currently count on an .edu email address to designate an academic sender, or red and green colors on a traffic light to designate when to move.

One forerunner is the cryptographic provenance standard C2PA / Content Credentials. This is a technological solution, under development by a coalition of tech companies (Adobe, Microsoft, Google, OpenAI, BBC, New York Times, Sony, Samsung), working under the Linux Foundation, being driven by regulation (EU AI Act Article 50; California SB 942). It will only take off when there are formats that make it inviting and easy to use, and those will emerge gradually, on top of the technical base.

There may be formats to indicate range and depth: whether an answer rests on a settled body of knowledge, a contested partisan view, or an uncertainty. There may be formats for helping judge the reliability of an answer, added on top of the language model itself, verified by a third party. In the world of scientific journals, the format of peer review used to carry weight; now, we rely on the quantity of citations. Both have been called into question. We may need formats like Yelp review stars, summaries of Amazon reviews to indicate what people think of ideas, or credibility ratings for curators like the statistics format for ranking commenters on Quora.

We may also need formats to show the gap between what the model retrieved and what it interpreted from that retrieval. The two arrive woven into one sentence, in one confident register, and nothing in the current format marks what was interpreted or why. A reader who could see that gap at a glance would have a more incisive relationship with the information. The “at a glance” element is critically important. We want to know the truth, but we gravitate to systems that don’t make us work hard to find it.

Finally, we will need formats to help us control the vast amount of personal data collected about us. Juliette Powell, my coauthor on The AI Dilemma, who died in 2025, saw this coming. Years before most people gave it a thought, she co-founded with Intel Labs a research project called WeTheData that mapped the hidden economy of personal information; the work drew notice at the World Economic Forum, the New York Times, and MIT Technology Review. We are all vulnerable to having that information—our likenesses, our finances, our purchasing habits—collected and used to manipulate us.

Laws like the European Union’s General Data Protection Regulation (GDPR) set a floor for privacy, but a law is inert until a format makes it usable. GDPR normalized the consent button that greets us on every website, asking whether we want all, some, or none of our browsing data collected. That button is a perfect example of what a format can do and how it can fail. It makes the law usable, but it also shows how a format can get captured and engineered into problematic patterns. It’s all too easy to click “Accept All,” rather than go through the fatiguing process of voicing our data risk appetite. An honest AI-based format, which told us the risks in plain language, guessed our preferences, and asked us to confirm that guess, would make higher levels of trust possible.

Juliette understood that controlling your data begins with the literacy to interrogate it. “Could you query a data set?” she asked in our book. That is a request for a format. If we are ever to truly own our data, we will need formats that make plain exactly what we are granting—and what we are entitled to ask in return.

GENERATING REVENUE

Like just about every “free” online service before them, AI systems are finding compromises that monetize their offerings. In February 2026, ChatGPT began showing ads. They appear below the answer, clearly labeled and visually separated. OpenAI has explicitly said that the ads will have no influence on what the model says.

Unfortunately, advertising always influences the content, at least indirectly. That’s why CEO Sam Altman had previously said advertising would be a “last resort.” Cigarette advertising is the reason why very few magazines in the 1950s and 1960s published articles about the dangers of smoking. (The exceptions were Reader’s Digest, which was profitable enough to ignore advertisers’ pressure, Mother Jones, which did not rely on advertising for survival, and Mad Magazine, which carried no advertising.)

For AI systems, advertising will affect the ways that platforms attract audiences, and that in turn will shape the personalities of the system. Data about customer use will be inseparable from the format design. Using AI systems will mean being constantly accosted by sales pitches in ever-more-insidious formats.

Formats that generate revenues are not necessarily bad. Some of them will fund independent creators. Some will be transparent about the way they do this. Inevitably, as always with media, LLM audiences will become sophisticated. They’ll come to learn which language models have which hidden biases, linked to which collections of advertisers. Pop-up formats may emerge to help people recognize when ads are trustworthy and authentic, and when they’re scams.

Ads are always catalysts for format innovation. In a digital environment, subscriptions may also be. We may see formats that aggregate subscriptions, reveal hidden costs, or hide real costs. Already, there are formats that make it easy to link software to our bank accounts. GenAI will develop formats to help us handle the confusion associated with those links.

Joe Ziska’s work—and many other efforts to track technological change—suggest that technological systems tend to learn from the efforts that came before. History doesn’t circle; it spirals. When the internet was developing its formats, bottom-up groups like the Electronic Frontier Foundation (EFF) and top-down registries like the Internet Corporation for Assigned Names and Numbers (ICANN) emerged to help steer it in positive directions. Many of the most prescient commentators on current-day technology had roots in these efforts. There has also been a link between internet commentary and format generation. We need new collaborative efforts to help AI realize its positive potential.

We can’t know for sure yet, but LLMs and GenAI seem to be more open to format innovation than social media were. Facebook, Tiktok, Instagram, YouTube and LinkedIn are all tightly controlled. When a new format is introduced, it comes from the top. AI systems invite more experimentation and participatory design.

Format design tends to happen in an ad hoc fashion, independent of the regulations. Maybe this time we can be more deliberate and self-aware of the formats we use. We can do more to design and test formats to advance transparency, reliability, credibility, and other human values. As we did in the early days of the Internet, we may need to meet and think together about the formats we will need, and the institutional support necessary to engender them.

It’s only been three and a half years since ChatGPT was introduced publicly. A generation that has already watched social media be taken over might be able to develop the formats we need while the AI systems are listening. This is not a question about the technology. It is a question about us.

I recently published a Substack article on predetermined elements: what we already knew for sure about the 2030s. First on the list was “climate change gets worse.” Almost immediately commenters accused me of succumbing to “thick succulent climate #fearporn” and disaster roulette. We’d all heard this 30 years ago; wasn’t I repeating the same old cliché?

It made me curious. I’ve been reading material in this field so long – everyone from Amory Lovins to Bjørn Lomborg – that I had assumed that there were three things that everyone agreed on about climate change:

1) the temperature of the Earth is changing over time;

2) the reason is the greenhouse gas effect; and

3) this is the cause of extreme weather.

These comments made it clear that not everyone agrees on these three elements. But were these objections reasonable? Or was I being trolled?

I am mindful of Upton Sinclair’s famous aphorism: “It is difficult to get a man to understand something, when his salary depends upon his not understanding it.” That logic has been used ad nauseam to accuse everyone on all sides of the climate change debate. I don’t want to be accused of bias, and I want to find out what’s real. If there’s ground truth about climate change, I’d like a reasonable layperson’s understanding of it.

So as an experiment, I decided to bring out two tools, one old and one new: the ladder of inference and the large language model Claude. The ladder, for those who don’t know it, is a conversational gambit used in organizational learning and conflict resolution. It guides us to articulate the near-instant climb each of us makes in our head when we think about a complex or controversial issue.

There’s a natural human tendency, faster than we realize, to move from observable data — something that happened, that everyone present would agree happened, that would show up on a recording — up through the rungs of selecting data, interpreting it, and making assumptions about our interpretation, to conclusions that we hold, to beliefs that shape our actions. Two different people, observing the same basic facts, could come to completely different sets of beliefs. Chris Argyris calls this “leaps of abstraction.” The higher we leap, the more we mistake the view from the top of the ladder (our conclusions) for the truth at the base of the ladder. And the more certain we are, the less likely we are to notice our own assumptions.

So in writing “climate change gets worse,” was I guilty of leaping up to a misguided or partisan set of beliefs? I resolved to use Claude to help me examine this.

Large language models excel at this kind of task: assimilating the mass of material written on complex topics, and probabilistically assembling a synthesis of the diverse range of writing on the subject. So I gave Anthropic’s Claude an article on the ladder of inference. I had it read my article on predetermined elements. Then I gave it this prompt:

Then I assimilated the answers, tested my understanding of them, and checked the resource links underlying them, until I felt confident of my own perspective.

The short answer:

“Climate change gets worse” is indeed predetermined. But “worse” doesn’t mean “catastrophic,” not necessarily. Not even given this month’s heat wave.

Nonetheless, the overall temperature of the Earth will continue to rise all through the next 50 years. The effects will be uncertain.

Most likely: Some people, depending on where they live, will be faced with devastating effects. Some people will be faced with severe weather that blows over. And some people won’t be directly affected at all.

By the 2030s: the effects of climate change will be heightened. Most people will probably still be able to ignore it if they choose — for a while, and depending on where they live. The details of this are uncertain.

The longer answer:

The Earth is warming

This is as close to observable data as a planetary measurement gets. Six independent temperature records — kept by NASA, America’s NOAA, Britain’s Met Office, Berkeley Earth, Europe’s Copernicus service, and Japan’s weather agency, built by separate teams using different methods and different stations — all agree that the surface has warmed by roughly 1.1°C since the late nineteenth century.

Skeptics have objected to some of the readings: cities heating up around old weather stations, instruments moved, raw figures adjusted. But the warming shows up consistently and plainly in instruments that cannot be manipulated. Satellites measure it in the lower atmosphere. A global fleet of robotic floats measures it in the heat stored in the oceans. Sea level has risen. Glaciers and ice sheets have lost mass. Spring arrives earlier; species move toward the poles and up the mountainsides. When a dozen unrelated systems move in the same direction, we are looking at data, not at a manipulated artifact.

However: The calculated amount of warming depends on what we use as a comparison date. The precise figure depends on which pre-industrial baseline you choose and how you reconstruct the thinly measured past, which is why careful sources give a range rather than a single number. But there is no dispute over the warming itself. To reject it you must reject the thermometers, the satellites, the ocean floats and the ice all at once. This is not a matter of “consensus,” or of “models,” and it is not a “hoax” — those are ways of deflecting from the data. (It could still be overblown, but that argument comes later, higher up the ladder.)

Greenhouse gases are the cause

The greenhouse gas effect itself is basic physics. Carbon dioxide and methane absorb heat radiating from the Earth. This was demonstrated in the laboratory in the 1850s, and is confirmed today by satellites. The concentration of carbon dioxide is continuously measured around the world. It has climbed from about 280 parts per million before industrialization to more than 420 today. The added carbon is linked to human industrialization by measurements of carbon isotope patterns. Burning fossil fuels is linked to higher levels of carbon-12 relative to carbon-13. The falling level of atmospheric oxygen is another indicator that carbon is being burned, rather than escaping from a warming sea. None of this is modeled. It is observed.

The larger claim — that human emissions are the main cause of the warming we measure, not merely a contributor — is an inference. But its fingerprints rule out the alternatives. If the sun were driving the warming, the whole atmosphere would heat together; instead the upper atmosphere is cooling while the lower atmosphere warms, which is the specific signature of heat trapped below. Nights are warming faster than days, winters faster than summers. That is why the world’s climate scientists, in language negotiated word by word with every government in the room, dropped the hedges and called the human hand in warming unequivocal.

The real uncertainty on this rung is not whether but how much — how much warming each doubling of carbon dioxide finally delivers, a quantity called climate sensitivity. After decades of research, the best estimate is about 3° Celsius, with a likely range of 2.5° to 4° Celsius. That range is significant, and the response need not be smooth: the feedbacks are non-linear, and the warming per doubling may itself rise as the planet heats, so the high end cannot be dismissed. Catastrophe is plausible, but not predetermined.

The warming is causing extreme weather

As I write this on a July evening in New York, my air conditioner is on full blast, and I am sweltering nonetheless. I am tempted to jump up the ladder of inference and blame climate change for causing this. But I can’t. No single storm or flood or heat wave is “caused” by climate change directly.

Instead, I have to talk about probabilities. A warmer world affects the weather by changing the likelihood of extremes. It loads the dice toward certain kinds of extremes. Loaded dice still turn up ordinary numbers most of the time; what changes is how often the extreme ones come up. So the honest claim is not that any one disaster was caused by warming, but that the odds of certain kinds of extreme have shifted.

The confidence level for different causal relationships varies depending on what type of extreme weather we’re talking about. Heat is the strongest case: hotter, more frequent heat waves (like the one affecting Europe and the US right now) are not quite predetermined in the future, but extremely likely. These follow directly from a hotter average, and the science treats the human hand in temperature extremes as established fact. More frequent heavy rainfall is also close to predetermined, because warmer air holds more moisture. Drought is regional and mixed, worsening in some places and murkier in others. Tropical cyclones are a different and more specific case: little sign of more storms overall, but better evidence that a larger proportion of them will reach their most intense categories and carry more rain. For tornadoes, river floods, and wildfires, we can’t say for sure that a warming climate is responsible.

I conflated two things in that earlier article when I called rising financial losses predetermined. Losses are indeed climbing, and that much is predetermined — but most of the climb comes from exposure, not hazard: more real estate and other assets sitting in places that flood, slide and burn. That trend is driven by where we choose to build, not mainly by fossil-fuel emissions. Both are predetermined; only one is about the weather, and the statistics of unprecedented damages fold them together.

Arguments up and down the ladder

I then went back and forth with Claude about some of the arguments people make on various sides of the climate change controversies. It turns out we can place them on the ladder of inference. In doing this, I tried to be careful to treat each argument fairly: to phrase it as its adherents would phrase it, and not mischaracterize it. Then I placed each on the ladder: was it ground truth, the kind of observable data on which we would all agree? Was it an inference, selected from the data? Or was it a conclusion, the result of a leap?

The idea of climate change is a hoax.

Scientists inflate the threat for grant money. “Climate alarmism” is a pretext for raising taxes and imposing control. The underlying data has been cooked.

Some of these assertions can be traced to the 2009 “Climategate” email scandal, which eight separate investigations examined without finding fraud. About a thousand private emails from the Climatic Research Unit at the University of East Anglia were posted online without the senders’ permission. They were embarrassing. One researcher talked of using a “trick” to “hide the decline” in temperatures shown by tree rings. Others discussed keeping contrarian papers out of scientific journals.

Extensive analysis by skeptics confirmed that the climate research being discussed was reliable and conducted in good faith. Some of the scientists were temperamentally defensive and insular, but the criteria for “hoax” were not met. No one who was in a position to fabricate evidence or shut down opposing research had actually done so.

Underlying the hoax claim is a broader assertion about climate science: Institutions reward alarm, and researchers fall prey to the conventional wisdom. This is a belief at the top of the ladder. Even if there were hoaxes, they have clearly been unsuccessful at convincing everyone. If the qualifying factor for a hoax is cherry-picking data to match your chosen conclusion, then the same charge could be aimed at all sides.

Cutting emissions is futile.

This argument starts with the idea that the amount of carbon increase is already baked in from past emissions. No further cuts will do anything to mitigate extreme weather effects.

The underlying question is future emissions — the carbon we have yet to burn — and whether that makes things worse. The temperature keeps rising as long as humanity keeps emitting greenhouse gases. If those emissions stop, then the temperatures will level off. However, the global temperature will not decline on its own; that would take pulling carbon out. So cutting is the opposite of futile: every ton not burned is warming that never arrives.

What we don’t know precisely is how many degrees each ton of carbon buys — the same climate-sensitivity question from before, seen from the other end. We also don’t know at what point the situation gets too critical to ignore, if it ever does.

Some commentators estimate that further emissions won’t matter that much. These include Judith Curry, Richard Lindzen and Roy Spencer, who accept the greenhouse effect but judge sensitivity modest and natural variability large.

Others suggest that further emissions will push us to a tipping point of acceleration. James Hansen, for example, has warned that more emissions, producing more global warming, could push the Atlantic Meridional Overturning Circulation — the system of ocean currents that includes the Gulf Stream — to slow or shut down, disrupting temperatures around the world. That risk is plausible but seriously contested; the early-warning signals are disputed. It’s high impact, but uncertain.

Climate change will be addressed

It’s already being addressed. It’s predetermined that there will be more efforts to mitigate the effects of climate change. But how easy, cheap, and effective will those efforts be? That’s uncertain, and any conclusion is high up the ladder of inference.

Bjørn Lomborg, following the Nobel economist William Nordhaus, calls warming real but manageable, and bets on prosperity and adaptation over expensive mitigation — while degrowth thinkers such as Jason Hickel argue the reverse, that growth itself is the problem.

Does economic growth lead to lower or higher emissions? History suggests both. Richer economies burn more energy and emit more; they also grow more efficient, and some have pushed their own emissions down from a peak. But whether that adds up to real decoupling is just what Hickel disputes — much of the rich world’s decline came from offshoring production, while global emissions kept climbing. Unresolved.

Bill Gates, Steven Pinker and the ecomodernists bank on falling clean-energy costs. Mark Jacobson says renewables can handle all human energy needs. Twenty-one scientists led by Christopher Clack say that some sectors are inherently hard to electrify — cement, steel, aviation — and that a non-fossil-fuel grid still leaves farming, land use, and methane untouched. There are also the various cloud-seeding and carbon extraction schemes being proposed.

Underneath all these ideas are two conclusions that apply not just to climate change but to all human endeavor. The precautionary principle says we should avoid risks that might affect all humanity: fossil fuels, nuclear power, and AI development included. The techno-solutionism principle says that no matter what predicament humanity creates for itself, we’ll be able to invent our way out of it. Both extremes are comforting, if you believe them. Both are flawed, in my opinion—but my opinion doesn’t count for much when we’re looking for evidence and ground truth, because both are high up the ladder of inference.

Scenarios and risk

This is not, in the end, an argument about climate. It is about our collective risk appetite. It is predetermined that climate change will put some things that humanity values at risk. But we don’t know what those will be.

Among the possibilities: vulnerability to extreme weather, the fossil fuel industry, the status quo, the number of climate refugees, the lost natural environments, and the political and economic fortunes of those on all sides of the debates.

We don’t know which threats will become reality. We don’t know which risks are more likely, or which effects will be more grave. We don’t know which mitigation efforts will come to pass. We do know that we all have different risk appetites and, even if we could talk about them openly and dispassionately, we probably wouldn’t reach agreement on which risks are worth taking.

In cases like this, the typical scenario approach is to look for “no-regrets solutions.” There are things to do now that, no matter what happens, will make us glad in the future that we did them. The strongest candidates for no-regrets solutions include using energy more efficiently, hardening infrastructure against extremes, and funding research with transparent adversarial review. These steps pay for themselves whether the warming proves mild or severe. They are already reducing emissions in some geographies.

In my view, the battle over these measures is not about the cost-versus-benefits, but about pride. The greatest barrier is the all-too-human reluctance to admit we might be wrong.

At the same time, one predetermined factor is increasing human demand for energy. No matter what else happens, we will need more electric power and fuel, if only because emerging economies continue to emerge. Any plausible future must assume humanity handles this appetite somehow.

Paul Hawken has done an admirable job of cataloging solutions. Some are no-regrets solutions. Others require sacrifice. History and economics suggest that humanity will end up adopting some of them. We will probably augment our current energy sources with many more non-fossil-fuel options. Not predetermined; but very likely. Also very likely: we will continue to use fossil fuel for quite a while.

So we push on, looking for ways to improve what we do as a species. I suspect this debate will only be resolved in retrospect. Someday, perhaps in the 2070s, humanity will look back on what it did about climate — having decided, through its actions, which risks were worth taking.

Pierre Wack, the scenario pioneer who built Royal Dutch Shell’s celebrated foresight practice, sometimes explained his methods by talking about the Ganges river in northern India. If there are heavy monsoon rains over the Himalayan headwaters, you can tell with certainty that there will be a flood five days later at Allahabad, which is 650 miles downstream. Five days after that, he said, the floods would reach Benares.

“Now the people down here in Benares don’t know that this flood is on its way,” he said, “but I do. Because I’ve seen it! This is not fortune telling. This is not crystal-ball gazing. This is merely describing future implications of something that has already happened.”

Wack and his Group Planning colleagues referred to these certainties as predetermined elements: the parts of the future whose inertia exceeds any plausible intervention. Another scenario pioneer, Peter Schwartz, refers to them as inevitable surprises, because although they are certain to affect us, most of us are not emotionally prepared for them.

Most of us, peering ahead, fix on anxieties and uncertainties that may or may not happen: elections, technologies, and potential crises. We imagine what might happen, and get into the habit of thinking that our fate depends on this contingency. For instance, we pin our hopes on a particular candidate getting into office.

An alternative is to look at the predetermined elements in our world as the playing field. When we recognize the true certainties, we can leap ahead to framing our choices and modulating our expectations. For example: We know it will take a long time to mitigate the effects of the climate crisis, so we invest accordingly in renewable energy. We also know our efforts to manage artificial intelligence need to happen practically overnight, so we work to rapidly build the necessary skills.

There are two kinds of predetermined elements. The first-order trends are basic and happening now. They follow directly from events that already took place — children already born, tons of carbon already in the air, debts already incurred. The second-order ones arise from the combination of first-order forces. Their effects are less predictable, but we can’t avoid the pressures they will place on us.

Taken together, they tell us the world of the 2030s will be markedly different from today and from most predictions being made today. For system leaders, a good list of predetermined elements gives you a start on developing scenarios that help you move to a creative orientation: creating the future you want.

I do a lot of work with scenarios, particularly at New York University’s Interactive Telecommunications Program, where I teach a graduate-level course on the future of media and technology. Here is my list of predetermined elements facing us today.

First-Order Predetermined Elements

The climate crisis gets worse. Enough carbon has already accumulated that the global average temperature will cross 1.5°C above pre-industrial levels by the early 2030s. The intensity of extreme weather, and the frequency of the most damaging events, will increase accordingly. We will see more ancillary effects. For example, the cost of insuring property from weather disasters – storms, fires, and floods – will continue to rise.

Water shortages increase. Enough groundwater is being permanently depleted — fossil aquifers like the Ogallala, the North China Plain, and northwest India do not recharge on human timescales — and enough reservoirs and snowpacks are being drawn down or shrinking to make this a driving force. Chronic water shortages will affect governing regimes throughout today’s arid and semi-arid regions, including the Middle East, North Africa, South Asia, and the American Southwest.

The planet’s biological impoverishment continues. Humanity will continue to encroach on the carrying capacity of the planet. Species extinction will continue to run orders of magnitude above the natural background rate — though most people, living in cities, will register it as a fact reported rather than a loss they feel directly. There are many counter-measures emerging today, and there is a plausible future of long-term ecological recovery (though it is not predetermined). But by the 2030s, if we are not personally affected by food shortages or loss of habitat, it will be because we aren’t paying attention.

The human population balance tips. Enough of the world will have aged, enough birth rates will have fallen below replacement, and enough young people will have been born in sub-Saharan Africa that the species will be simultaneously old and young as never before. Roughly two-thirds of humanity now lives in countries below the replacement rate — China near 1.0 births per woman, South Korea around 0.75. It took less than 100 years, from 1927 to 2022, to expand from 2 billion to 8 billion humans. Now that growth is leveling off. By the mid-2030s, almost all the net growth in the young-adult workforce will be African.

There are several factors involved, all involving deep and seemingly permanent shifts in human relationships, and all permanent; despite the “culture warriors” in many countries demanding that women bear more children, there is no going back. The sociologist Alice Evans traces falling fertility to a centuries-long dissolution of patriarchal kinship: the replacement of arranged marriage by smaller, more autonomous households, and the erosion of the extended family as an economic unit. In culture after culture, women who gain autonomy defer giving birth until they are older, which means fewer children. Even if the childbearing dynamic were to change, which would contradict many other trends, it would take decades to see the aging effect reverse.

The energy system flips. In general, adoption of new technology is determined not by innovation itself but by factors related to demand: the felt need for what the technology does, the relative accessibility of the technology, the cost curve (how rapidly the price drops as use expands), and the capital already committed (including human capital). Today, those factors all favor renewable energy. Solar and wind, innovative batteries, and other manifestations of non-fossil-fuel energy are moving forward rapidly and broadly enough to be poised to become the largest single source of fuel for the global electric power system, overtaking coal and natural gas.

That doesn’t mean that fossil fuels will go away, that any one particular form of energy will corner to market, or that the cost of electricity will lessen. Demand has consistently risen to absorb new supply, and that seems predetermined to continue for some time. In this case, demand includes the computing infrastructure needed for AI servers, but also the technologies needed to manage water shortages, rising temperatures, and rising living standards around the world. Even more energy-efficient AI won’t stop demand from growing; the more energy produced, the more uses people will find for it.

Digital technologies keep maturing. This sea change in human activity began with the semiconductor 75 years ago, and its basic dynamic—continued broadening of computer use with decreased costs at a compounding rate—is predetermined to continue for at least another 10, and probably longer. Because so much technological upheaval has happened so fast, we tend to underestimate the changes still to come. We think of today’s Big Tech companies as destined to be dominant forever, but no individual platform, no matter how well-entrenched, is certain to hold its position. The explosion of large language models in 2023 is just one in a string of similar innovations, and we don’t know which will be truly transformative, in negative or positive ways: Robotics? Prosthetics? Digital fabrication? Individually designed apps? Surveillance? We only know that the pace of change and the sophistication of the tools will increase.

Medicine gets personal. A cost curve steeper than Moore’s Law is at play: sequencing a human genome cost roughly $100m in 2001 and a few hundred dollars today. The first CRISPR therapy won approval in the United States and Britain in late 2023; the mRNA platform, proven at scale during the pandemic, is being turned toward cancer. As in many technologies, the basic research takes time to translate into practical applications. That tipping point is here. Enough personalized medicine capability exists that treatment tailored to an individual’s genetic profile will be increasingly standard in oncology and rare disease by 2031 — still costly, still unevenly distributed, but no longer exotic. This is happening just in time to affect the later lives of tomorrow’s elders; perhaps not soon enough for the people in their 70s and 80s today, but well in time to transform life for many people currently in their 40s.

The fiscal vise. Enough sovereign debt has been issued, and enough of the cost of aging populations has landed on public budgets, that governments across the world will continue to face structural pressure on their finances. The debt itself is predetermined: the bonds are sold, the deficits are structural, the pension and health obligations are fixed by populations already born. What is not predetermined is whether this becomes a crisis or merely a condition, and that turns on interest rates, the general availability of investment capital, and the willingness of creditors to accept loss. Demographic logic, which the historian David Hackett Fischer traced across eight centuries, suggests that as populations shrink they need fewer houses, schools, services, and goods — so interest rates tend to fall. It’s not quite clear yet how those dynamics will play out this time.

Human capability rises overall. This trend is barely recognized, but the global generation reaching adulthood by 2031 will be the most schooled in history. Test scores may have declined in recent years in the US and other Western nations, and there may be legitimate concerns about the quality of school-based education and losing ground to artificial intelligence, but in general, the overall caliber of human skills is rising.

Second-Order: The Predetermined Combinations

The first-order elements are forces in isolation. The second-order elements are what happens when they collide. In making this list, I’ve applied a strict test. The link to two or more first-order predetermined elements must be reliable enough that the result is a pressure we can count on — even when the way people answer it remains open.

Most of these second-order surprises have, at their core, a sea change in the systems and institutions around us. They were developed for a world of young people, robust natural habitats, and prevalent internal combustion engines. It was assumed they would never have to change. Now it’s up to us to change them—and probably not through revolution. Revolution appears to be a young person’s game and there probably aren’t enough young people. Instead, we may have to change our institutions and related systems, dramatically and thoroughly, through non-violent means.

Effects of an aging world

Digital technology advances in fields where workers are scarce, not where they are cheap. One prevalent anxiety — machines coming for jobs — assumes a surplus of employees. The predetermined reality across the aging North and East is the opposite: a shortage. There will be an increase in need for help, probably in ways we don’t yet see, that combines human and digital prowess. Acemoglu and Restrepo have measured the demand: the faster a country ages, the more robots it installs. Germany, Japan, Korea, and now China automate out of necessity, not avarice. Uncertain: whether automation lifts productivity or merely fills the gap — and how the politics land where workers are still plentiful.

The cheap-labor ladder is being pulled up behind Asia. Every modern economy climbed the same ladder: cheap young workers populated the factories, and the factories built the skills and the capital. That ladder needs a reservoir of cheap young labor, and the reservoirs are draining as Asia ages and its wages rise. The last great reservoir is Africa. Uncertain: how Africa navigates this momentous transition, and whether the ladder of industrialization remains the same going forward.

Weight shifts to the old — in the ballot box and on the balance sheet. The voters of the 2030s are already alive, and they skew older every year. By plain median-voter logic, the political center of gravity tilts toward those with the largest stake in present entitlements and asset values. The same demographics concentrate accumulated wealth in older hands while the young rent and service debt. Uncertain: whether societies redistribute wealth and power across generations, let the current imbalance harden, or find a way to expand opportunity for billions of younger people while reducing the damage of today’s ecological footprint.

The dependency load rises, and “who tends the old?” becomes a central question. The ratio of people past working age to people of working age is climbing steeply, and it is locked in place: the future old are already old, and the missing workers were never born. That alone guarantees mounting fiscal and care pressure, whatever the policy. Care work, moreover, resists the productivity gains that lifted manufacturing — it is mostly hands and presence. Its human face is the fastest-growing type of household in the aging world: an old person living alone, often in a city, with no extended family to lean on. Uncertain: who supplies the need for personal care — migrants, machines, family, or no one — and whether medicine compresses the costly years or extends them.

Fewer children, and more poured into each. Gary Becker named the mechanism half a century ago: as families shrink, investment per child climbs — the quantity–quality tradeoff. China’s one-child generation is the vivid case, but the pattern is general. A world of scarcer children concentrates money, attention, and anxiety on each one. Early indicators of change: the number of young adults without cousins, and the use of social media and AI as teaching tools for human interaction. Uncertain: whether that yields a more capable generation or a more fragile, more supervised one.

Pressure to forgive and restructure debt. Enough sovereign and household debt has been issued, against populations aging enough, that the collateral cannot keep growing against a slowing economy. Fischer’s demographic logic points out that a shrinking, aging population eventually demands less capital, and the growth that once outran debt no longer does. Meanwhile, a huge amount of capital will be needed to restore Mother Earth. The pressure to mark down the debt — to forgive, inflate, or restructure — will build predictably. Uncertain: whether that pressure resolves through forgiveness, inflation, default, or grinding austerity.

Geographic changes

The largest cities humanity has ever built will rise in its hottest, poorest places. The defining human movement of the era is not the flow across the rich world’s borders. Most movement is short-range and internal: countryside to city, South to South, as a mechanizing agriculture sheds hands and a shrinking family can no longer work the land. The cohorts are born. The trajectories of Lagos, Dhaka, Kinshasa and their peers toward becoming among the largest settlements ever assembled are about as predetermined as demography gets — and they sit on some of the most heat- and flood-exposed ground on the planet. Uncertain: whether Southern megacities become engines of capability, driven by their own innovative populations, or concentrations of misery dependent on outside support.

Water becomes a problem of energy and capital — and divides the world accordingly. The aquifers are draining, and the cheapest electricity in history is arriving where the sun is strong — both locked. Together they convert water scarcity from an absolute physical limit into a question of whether you can afford the energy and the plant to manufacture water: to desalinate it, recycle it, move it. Coastal, sunny, capital-rich places increasingly can. Inland and poor places cannot. Uncertain: which regions develop their way out—and then claim leading roles in doing the same for others.

The information environment

No shared source of truth — and a rising premium on the verifiably human. When generating convincing content costs nothing, we lose what’s left of our shared and trusted picture of reality. The predetermined flip side: as the fake becomes free and infinite, the un-fakeable becomes scarce and dear. Live presence, embodied skill, provenance, a local reputation that took years to earn — these re-price upward precisely because they cannot be synthesized. Their validity is particularly important for AI systems which are seeking the credibility that comes from citing credentialed sites like Wikipedia. Uncertain: whether institutions of trust are established, or the erosion wins.

The visibility of elite impunity. Universal information colliding with extreme inequality does not in itself make the powerful more accountable. But it makes their conduct more visible. The activities of the “Epstein class” are now seen, in detail, by everyone. Other leaks and releases of information will occur. Uncertain: whether visibility converts into consequence, cynicism, or unrest, and which crimes society decides are forgivable.

AI in the environment. Driven not just by digital innovation, but also by biotech and the need for urban infrastructure, AI will reshape every city environment. Surveillance systems, facial and moement recognition, and light-and-sound engineering will bring responsive technologies outside the virtual space. It will be possible to create streets that resemble Who Framed Roger Rabbit, in which we continually encounter human-made animated systems that know us by name. Uncertain: Who will design these systems, how they will be used, and how people will respond.

The pressure every institution feels

The installed base of human error. The built world encodes the assumptions of the year it was built, and first-order change is now outrunning those assumptions faster than we replace concrete and code. Dubai built almost no stormwater drainage, on the sound assumption that it rarely rains; a warming atmosphere then delivered the rain the city was designed to exclude, most catastrophically in April 2024. Every such embedded assumption is a latent error waiting for the conditions that expose it. Uncertain: which assumptions fail, and how visibly.

The pressure to become a learning organization — even on the corrupt and the authoritarian. The most capable generation in history is reaching adulthood, while the most knowledgeable expertise on human management is available. We are all more connected, better informed, and better able to compare our lives with others (including others of the past) than ever before. The demand for adaptive, learning institutions is therefore predetermined — even where corruption is rampant and the will to reform is dormant. Uncertain: Whether the outmoded businesses and governments of our time will endure, and what would replace them if they change.

A shared incentive against catastrophe. Humanity now occupies a single niche — the world as a whole — and shares, for the first time, a common exposure. As Peter Senge puts it, “All humanity occupies a single niche: the world as a whole. We have no history of doing this successfully.” The interdependence is first-order and already here; what it produces is a common incentive not to push any catastrophe — nuclear, climatic, biological — past the point of no return. Uncertain: Whether we pay attention.

Cheap, deniable force within reach of small actors. The drone and autonomous-weapon cost curve, and its proliferation, are both locked in place. Together they put precise, deniable force in hands that never had it. The same may be true of AI-driven cyberattacks and bioattacks. Uncertain: who wields these forces, to what end, and with what technological, legal, or military restrictions.

The human cost of war rises, and people become targets rather than soldiers. Families are small and each child precious — the quantity–quality world — so low-fertility, aging societies grow acutely averse to military casualties: every soldier lost is an only child, a family line ended, and the tolerance for mass casualties that larger generations could absorb is gone. At the same time, the drone and autonomous-weapon cost curve lets a state or a faction project lethal force without putting its own young people in harm’s way. This changes the nature of war. Edward Luttwak called the casualty-averse version “post-heroic war”; the cheap drone supplies its weapon. Uncertain: whether this deters war (too few children to risk) or unleashes it (cheap, deniable force lowers the threshold), and whether any norm constrains the targeting of individuals.

The limits to certainty

The persistence of the ordinary. Against all the above sits the most underrated predetermined element of all: most people, most days, will live recognizable lives. The school, the clinic, the shop, and the family table endure because institutions change far more slowly than the forces acting on them. This is not complacency; it is the buffer that keeps the surprises survivable — and the reason that system leadership is generally local.

It’s as if we’re all driving down a treacherous highway. We notice the accidents and cars being towed off the shoulder, and the road rage as cars cut each other off. We don’t pay attention to all the drivers who stay in lane, leaving enough space between themselves and the car in front of them. Many of those drivers have experienced past accidents; they don’t want any more. If there were more of them, the road wouldn’t be nearly so scary. Uncertain: the prevailing attitudes and what it takes to bring people to a more system-oriented perspective.

Approaches for System Leaders

The discipline of scenario thinking is a discipline of attention. It tells us where to pay attention. Which predetermined elements affect us most? Which opportunities should we focus on? And which changes do we care about most urgently?

It is humbling and steadying at once: humbling because so much of the future is already decided, steadying because so much of it depends on what we do together.

The predetermined elements provide a working map. The first-order forces — the aging, the warming, the sun, the grid, the genome, the debt — are the ground on which the next decade must be built. The second-order combinations — cities, pressures, robots, possible relief — are where our work takes place.

All three were predicting a crisis during the next few years, between 2010 and 2020.

All three theories had regular crisis points embedded in their cycles. The crises mapped onto history and recurred with regularity. As far as I could tell, however, this would be the first time that the crisis points of all three theories converged — that they would all happen at once.

The crisis came. We’ve all been living in it, even if some of us disagree about what it means or how it will end. Now it’s time to think freshly about where the crisis is taking us and how it might develop.

Here is the slide I showed first at that meeting:

The blue line represented Carlota Perez, the Venezuelan-British economic historian whose 2002 book Technological Revolutions and Financial Capital had become the most rigorous account of the long cycles that drive industrial change. Perez argues that every major technology — from the steam engine to the railway to electricity to mass production to the computer — moves through the same four-phase cycle of roughly sixty years. The current cycle, the information and communications surge, had begun around 1971. By her timeline, it would hit its midpoint crisis sometime between 2000 and 2020. The dot-com crash of 2000 and the financial collapse of 2008 were, in her reading, harbingers of that crisis — a “double bubble,” she called it. What came next had previously been a golden age. Each time, the political and institutional response had been equal to the moment: the leaders of government and capital in the industrialized world came together to forge a period of relative peace and economic growth.

(These are all slides from that same deck.)

This time, the crisis period (which she called “frenzy”) was taking longer to resolve. Indeed, although we didn’t know it yet, it would still be unresolved (and seemingly unresolvable) all the way through 2026 and probably longer. Although as Perez has noted, the crisis point always seems unresolvable until the pieces click together.

The purple line showed a path of change in the prevailing concept of government’s purpose. In his 2002 book The Shield of Achilles, American constitutional scholar Philip Bobbitt argued that the form of the modern state changes roughly every hundred years, driven by epochal wars and the constitutional settlements that follow them. The twentieth-century nation-state, which had won the long war against fascism and communism, was now transitioning to what Bobbitt called the market state — a different kind of polity organized around maximizing opportunity rather than redistributing wealth.

The transition would be marked by political instability, by the rise of populist movements, and by the inability of existing institutions to deliver on the bargains they had made with their citizens. Bobbitt’s timeline put the crisis firmly in the early twenty-first century. This crisis would also be unresolved, and seemingly unresolvable, in 2026.

The third was William Strauss and Neil Howe, whose 1997 book The Fourth Turning proposed that American history runs in roughly eighty-year cycles of four generational moods, ending in periods of profound civic crisis. By their count, the United States was due for its fourth such turning between 2005 and 2025. They had suggested, with some specificity, that it would arrive with a financial shock, deepen through a decade of institutional dysfunction, and resolve only after the country had been forced to remake its civic architecture. That crisis would be in full swing by 2026.

Back then, when I overlaid their timelines in a single PowerPoint deck, the alignment was striking. Perez had her crisis arriving in the first decade of the twenty-first century. Bobbitt had his constitutional transition reaching its critical phase in the same window. Strauss and Howe had their fourth turning beginning around 2005 and culminating around 2025. “You are here,” I added.

At the time, the prediction seemed interesting but theoretical. The financial crisis of 2008 was visible in the rear-view mirror. The political tremors were beginning — the Tea Party had emerged in the United States, austerity politics had begun in Europe — but the larger picture still felt manageable. I gave the talk, the audience nodded, and I went back to editing a magazine.

What I did not expect was how literally the prediction would be fulfilled.

In 2016, the political tremors became a full earthquake. Britain voted to leave the European Union. The United States elected a president whose political appeal was based on attacking the institutions and bargains of the twentieth-century nation-state. Populist movements rose across Europe, in India, in Brazil, and in the Philippines.

For me, the telling moment came in 2017, when Bill O’Reilly interviewed the newly inaugurated Donald Trump. “But Putin’s a killer!” said O’Reilly.

“We’ve got a lot of killers,” said Trump. “What, you think our country’s so innocent?” That was a clear sign we were not heading to Perez’ Golden Age anytime soon. Three years later, the pandemic exposed the fragility of global supply chains, and put every advanced state under stress. It revealed which institutions still worked and which had quietly failed long before.

By 2024, even though the economy had nominally recovered, every Western democracy was struggling with the same set of problems — declining trust, a growing have-versus-have-not economy, fractured information, a sense that the rules had stopped applying — and most of them had elected, or were on the verge of electing, governments that promised to break the existing order rather than repair it.

Now it’s time to look at long-wave cycles again, with a more critical eye. But if they were right about the crisis, what does their convergence tell us about what comes next? And — more usefully — what tools do we have for thinking about the present moment that come from taking the long-wave view seriously?

More Long Waves Than Three

The convergence I noticed in 2010 was not the whole picture. As I have continued to explore this territory over the past fifteen years, I have found a longer list of long-wave theorists, working in different fields, who all describe long-running patterns that affect the world we live in. Some converge on the early twenty-first century as a moment of crisis. Others do not. But all of them describe forces that operate on timescales much longer than the news cycle, and that shape what is possible in the present in ways that are not always visible.

A partial list: The debt cycles described more recently by Ray Dalio, with their longer arc of public-debt accumulation, currency reset, and the rise and fall of reserve currencies. The demographic cycles that drive both inflation and political stability, and that have entered, for the first time in modern history, a phase of pending predetermined global population decline rather than growth. The cycles of crisis and price changes, made vivid again by Covid-19, that historians like David Hackett Fischer have traced across centuries of European history. The cycles of plague and recovery identified by William H. McNeill. The cycles of trust in institutions, which appear to move in roughly thirty-year arcs.

Then there’s the cycle of civilizational collapse. Margaret Wheatley points to research by Joseph Tainter and Sir John Glubb showing that every complex empire falls prey to its own cultural decadence and dissipates. As Peter Senge notes, indigenous cultures have seen this cycle many times: growth, collapse, and rebuild. This time, we’re seeing it happen at a planet-wide scale. Vanessa Machado de Oliveira says that we are hospicing modernity: letting go of the “single story of progress, development, human evolution, and civilization that is omnipresent” and the violence and exploitation that goes with it. The experience of hospicing modernity, she writes, is like “watching a train wreck in slow motion… from inside the train.”

Some of these cycles are tightly connected. Others operate independently. None of them, taken alone, explains the present. All of them, taken together, suggest that the present moment is the result of multiple long-running patterns reaching their critical phases at once. The crisis is not one thing. It is several things, arriving together. Maybe the crisis point of this cycle is now; maybe it’s lasted more than 100 years, since 1919 when Yeats wrote his famous line: The best lack all conviction, while the worst are full of passionate intensity. Or maybe we are doomed, because of cycle after cycle, to relive the world of the worst, time and time again.

Cycles and Arcs

But focusing only on cycles misses something equally important. Underneath the cycles are slower movements that do not repeat. They go in one direction. They have proponents and they have opposition. They overreach and they retreat. But they do not, in the long run, reverse.

There is a difference between a cycle and an arc, and a great deal of the confusion in the current conversation comes from not making it clearly.

A cycle returns to where it started. The seasons cycle. The tide cycles. Perez’s surges cycle, in the sense that the new technology of one surge becomes the mature infrastructure of the next, and is then displaced by a new technology in turn. The form is recurrent.

An arc does not return. It goes somewhere. The expansion of the franchise — from property-owning men to all men to all adults — is an arc, not a cycle. There were reversals along the way, but no serious person today proposes to return the vote to property-owning men. The abolition of legal slavery is an arc. The recognition of the equal moral worth of women is an arc. The lengthening of human life expectancy is an arc.

A good example of an arc, and one of the most profound, is Brian Swimme’s explanation of human history in his book Cosmogenesis. Drawing on his collaboration with Thomas Berry, he says that since the Big Bang, the universe has tended toward greater complexity. The evolution of humanity is part of this arc; indeed, our movement toward complexity accelerated in 1927, with the advent of the Big Bang theory, and in 1964, when that theory was confirmed by the discovery of cosmic microwave background radiation.

Cycles and arcs look similar from inside a single moment. When the franchise is being contested in a particular election, it can feel like a cyclical struggle that could go either way. When the equal worth of a particular group of people is being challenged, it can feel like the whole arc is in jeopardy. The historians and theorists who specialize in cycles tend to see only cycles. The reformers and activists who work on arcs tend to see only arcs. Both are looking at the same world. Both are partly right. The world contains both.

What the current moment requires is a way of holding both concepts at once. Yes, we are in a crisis that several long-wave theorists predicted. Yes, the cycles they described are real, and the convergence of their crises explains much of what we are living through. But underneath those cycles, certain long arcs of progress are continuing to advance, even now, even as they are being contested. The arcs are real too. If Trump was giving voice to the cycles, then Kamala Harris represented the arcs when she said, “We’re not going back.”

The question is which view of the present serves us better: the cyclical view that says the wheel is turning and we should brace for collapse, or the developmental view that says the floor keeps rising even when the wheel is turning, and that the work is to make sure we end up higher than we began.

Three Arcs and their Discontents

Three arcs are continuing to advance, in my reading, even now, and even in places where it does not feel like they are.

The first is the recognition of the equal moral worth of all people. This arc has been visible for at least three hundred years, and longer if you count its religious antecedents. It runs from the Quaker antislavery campaigners of the seventeenth century through the American and French revolutions, through the emancipation movements of the nineteenth century, through women’s suffrage, through decolonization, through the civil rights movement, through the gradual extension of legal protection to people who have historically been treated as less than fully human. It has had massive reversals. It has not stopped moving forward.

The second is the recognition that individuals are entitled to make their own choices about the conduct of their own lives. This arc is younger — perhaps two hundred years — but in many ways it is the deeper one, because it concerns not who has rights but what rights mean. The choice of vocation, of partner, of place to live, of religious belief or none, of expressed identity, of how to spend one’s time and money and energy — all of these were once questions answered by the social position one was born into, and are now, increasingly, questions answered by the person living the life. This arc has also had reversals. It has also, in any period longer than a generation, advanced.

The third is the breaking of the intergenerational chain of harm. This arc is the newest of the three, and the most fragile, and in some ways the most important. It is the conscious decision, made by an increasing number of parents in an increasing number of cultures, not to repeat on their children the violences that were done to them. It is the recognition that childhood does not have to involve corporal punishment, sexual exploitation, emotional abandonment, or the systematic suppression of a child’s developing self. As a mass phenomenon it is perhaps a single generation old. It is not finished. But it has begun.

Each of these arcs has its proponents and its opposition. Each has overreached in particular moments, and the overreach has produced backlash, and the backlash has been mistaken — by the cycle theorists, and by many ordinary observers — for evidence that the arc itself is in retreat. It is not. What is happening is the cyclical pattern of progress and resistance playing out at one level, while the arc itself continues to advance at a deeper one.

Why the Arcs Persevere

The cycle theorists tend to see the populist backlash of the past decade as evidence that the post-war liberal consensus was an illusion. They are wrong, in my view, but wrong in an interesting way. What they are seeing accurately is that the institutional vehicles of the post-war consensus — the welfare state, the international order, the bureaucratic administrative apparatus — are no longer able to deliver on their promises. What they are missing is that the values underneath those vehicles have spread far wider than the vehicles themselves.

The proponents of equal worth, of recognized choice, of the broken chain, are now distributed across every continent. They include people in countries that the West has historically thought of as resistant to these values. They include the young, who in country after country are more committed to the arcs than their parents were. They include people inside the populist coalitions themselves, who arrived there for economic reasons but who have not, for the most part, abandoned the moral framework they grew up inside.

This is why the arcs persevere. Not because their institutional defenders are strong, but because their adherents are serious. They are not going to disappear. The institutions they will use to defend their commitments may be different from the ones we have known. But the commitments themselves are not retreating.

Where Agency Lives

If the cycles are real, and the arcs are real, where does human agency live?

Not at the level of the long wave. We do not get to choose whether the wave arrives. We do get to choose, in the smaller cycles inside the larger ones, what we carry forward from the previous wave and what we let go. Inside Perez’s sixty-year capital cycle there are shorter cycles of technology adoption that move at the pace of years. Inside Bobbitt’s hundred-year constitutional cycle there are election cycles and policy debates that move at the pace of months. Inside Strauss and Howe’s eighty-year generational cycle there are individual choices about how to raise children, what to teach, what to model, what to refuse.

Agency lives at the turning points. It lives in whether the next constitutional order incorporates the gains of the previous one or abandons them. In whether the next technology cycle distributes its benefits or concentrates them. In whether, in our own families and workplaces and communities, the arcs continue to advance through us, or whether we become the place where they stall.

Three long-wave theorists, working in different fields, predicted a crisis around 2020. The crisis arrived. We are living inside it. The institutional vehicles of the previous order are failing. The cycle theorists predict collapse. The decline writers predict apocalypse. Various optimists predict that everything will be fine. None of them is quite right. What is actually happening is more interesting. The cycles are real. The arcs are real. The work is to hold both at once, to see where in the smaller cycles agency lives, and to do that work consciously.

Sources:

Bobbitt, Philip. The Shield of Achilles: War, Peace, and the Course of History. New York: Knopf, 2002.

Dalio, Ray. Principles for Dealing with the Changing World Order: Why Nations Succeed and Fail. New York: Avid Reader Press, 2021.

Fischer, David Hackett. The Great Wave: Price Revolutions and the Rhythm of History. New York: Oxford University Press, 1996. (Thank you, Christopher Crews.)

Glubb, John Bagot. The Fate of Empires and Search for Survival. Edinburgh: William Blackwood & Sons, 1978.

Howe, Neil, and William Strauss. The Fourth Turning: An American Prophecy. New York: Broadway Books, 1997.

Machado de Oliveira, Vanessa. Hospicing Modernity: Facing Humanity’s Wrongs and the Implications for Social Activism. Berkeley, CA: North Atlantic Books, 2021.

Perez, Carlota. Technological Revolutions and Financial Capital: The Dynamics of Bubbles and Golden Ages. Cheltenham, UK: Edward Elgar, 2002.

Senge, Peter. Introduction to the Living Fieldbook (unpublished).

Swimme, Brian Thomas. Cosmogenesis: An Unveiling of the Expanding Universe. Berkeley, CA: Counterpoint, 2022.

Swimme, Brian, and Thomas Berry. The Universe Story: From the Primordial Flaring Forth to the Ecozoic Era — A Celebration of the Unfolding of the Cosmos. San Francisco: HarperSanFrancisco, 1992.

Tainter, Joseph A. The Collapse of Complex Societies. Cambridge: Cambridge University Press, 1988.

Wheatley, Margaret J. Who Do We Choose to Be? Facing Reality, Claiming Leadership, Restoring Sanity. Oakland, CA: Berrett-Koehler, 2017.

Yeats, W. B. “The Second Coming.” 1919. In Michael Robartes and the Dancer. Dublin: Cuala Press, 1921.

Note: This is my first-ever Substack post. I hope to post regularly. The system asks you to subscribe, and subscribing is free. I plan to publish everything I write or edit on your side of the paywall.

I think I know what will determine the growth path of artificial intelligence. It’s not regulation or investment. It’s not oligarchs, Big Tech, or rivalry between the US and China. It’s not AGI, singularity, or the capacity of data centers.

The most critical factor is human demand. AI, for all its military and industrial uses, is still most popular as a consumer product. As generative AI tools (ChatGPT and the like) get cheaper and more capable, they will be embedded deeply into ordinary life. They will evolve to meet what people want.

So what do people want? It’s a quality of interaction with people, at once responsive, knowledgeable, slightly sycophantic, and continually improving its judgment. This quality doesn’t yet have a name—call it human-symbiotic, maybe.1 We know it when we see it. In every plausible future, I think AI systems will continue to develop this quality, and if they do, people will continue to use them.

Scenarios for Human Demand

Each year, I teach a course called the Future of Media and Technology to graduate students at New York University’s Interactive Telecommunications Program. We develop scenarios that describe the way the world might evolve. These are not predictions; nobody can predict the future. Nor are they dystopian scare stories; every plausible future story has both comforting and terrifying elements, because the complexity of human life is present in all of them.

These scenarios are immersive sketches of worlds that might exist in 2032, just seven years from now. Close enough to feel urgent, far enough to allow for genuine transformation.

If you’d like to join us in a conversation about these scenarios, send an RSVP to me at art@artkleiner.com. The session is Monday, Dec 15, at 2-3:30 PM ET (New York time). It takes place by Zoom. There is no charge.

We created these scenarios to help us think about the decisions being made today, which will influence what happens next. We started with the question, “How will humanity evolve?” It’s a highly relevant question for students whose chosen career path (at this department) combines technological innovation, creative arts, and the intention of benefiting humanity. They and their fellow students will be at the front lines of digital evolution, which is moving faster and faster upon us.

Maybe most people want to enhance our bodies and skills, if only because AI (and wearables, super-prosthetics, and biotech) give us the means to do it. If so, we’re headed to Humans 2.0.

Alternatively, the popular demand may be for creativity and personal identity, with inexpensive and responsive AI systems that let us be who we want to be – at least digitally. This future, is called In AI We Trust (Maybe).

In the third scenario, called Intelligent Inequality, AI answers a different kind of demand—not from individuals, but from employers, entrepreneurs and investors. This is the demand for higher productivity. The grim warnings about job loss come true, and people around the world struggle with finding their livelihood.

Let’s look at these futures in a bit more detail, with videos that immerse us in each one – and then let’s think a bit about the implications.

1. Humans 2.0: Embracing Enhancement

In this scenario, technology has consolidated into large conglomerates that offer comprehensive human enhancement. Biotech, wearables, and immersive mixed reality have converged into a single ecosystem. Digital tattoos regulate emotional states and physiological functions. Some employers require implants as a condition of work. Celebrities flaunt their modifications. A new profession of biotech surgeons has emerged to meet demand.

Every walk of life is affected by this, including the motion picture industry.

Producers: Ariel He, Charles Simons, Chris Toh. Copyright © 2025, He, Simons, Toh. All rights reserved. Do not copy or circulate without permission.

Note that the creators deliberately declined to reveal which actor the directors choose. The ambiguity is the point. In 2032, many people will still prefer the raw vitality of unenhanced people. But how long will that last? If enhancement becomes ubiquitous and its costs decline, the unenhanced may become curiosities rather than competitors.

What should we watch for? If you see a credible smart tattoo prototype by 2028, this future becomes more likely by 2032. Similarly, if some jobs start requiring enhancements, that’s another sign. I imagine a future in which wearables and prosthetics evolve together, interacting with robotaxis and street signs come to life.

2. In AI We Trust (Maybe): The Promise and Peril of Decentralization

The second scenario imagines a world where AI power has dispersed. Breakthroughs in model compression have made AI smaller and cheaper. People run their own open-source AI systems, manage their own digital identities, and maintain “float profiles” that generate secondary streams of income. This is a future of personal agency—or so it appears.

The catch is that decentralized systems fail in decentralized ways. The technology doesn’t always work as promised, and when it fails, there is no single authority to blame or fix it.

Producers: Luna Chen, Devlyn Kornegay, Michelle Xu. Copyright © 2025, Chen, Kornegay, Xu. All rights reserved. Do not copy or circulate without permission.

In this future, people rely on AI utterly—for guidance, for logistics, for the management of daily life. AI has broader knowledge and doesn’t judge. But AI also doesn’t care, not in the way humans care about one another. Accountability fragments. Errors compound. The consequences fall on individuals who may have no recourse.

Early indicators include the proliferation of personal AI agents, the decline of centralized cloud computing pricing, and waves of small scandals involving AI misuse of personal data or inadvertently harmful recommendations. If people come to feel like they’re in control of AI, but blame other people for its missteps, this future becomes more plausible.

3. Intelligent Inequality: When Humanity Becomes Data

The third scenario realizes the fear of AI overtaking livelihoods. A small number of people control the technology, and many of the remaining jobs have either been devalued or eliminated. Productivity booms, but GDP growth decouples from employment growth.. AI and data become new forms of capital, concentrated in a few Big Tech companies. In the absence of economic value, humans create social value. They tell their stories. They contribute their data to the models’ growing knowledge base of what it means to be human.

Producers: Han Bao, Yihe Huang, Runqi Zou. Copyright © 2025, Bao, Huang, Zou. All rights reserved. Do not copy or circulate without permission.

The effect of this future has to do with dignity. We see people struggling to maintain their sense of worth – and we see hints that they may accomplish this. But it will not be without a struggle. “When these voices are fed into their machines,” wonders Yonah, “am I still the owner of my stories, or have I become nothing more than a simple idea?” The scenario’s closing question lingers: when AI owns the stories, who still owns the future?

One key early indicator may be the disappearance of entry-level software engineering jobs. If AI can write code, the profession that built the digital economy begins to hollow out from the bottom. Robotics and logistics automation may be other indicators. Perhaps the most telling sign is the reliance on data gathering as a primary occupation for displaced workers.

The strategy we don’t yet have

What unites these scenarios is a shared recognition that we have not yet come to terms with the kinds of actions that AI will require. AI regulation and governance is only a small part of the answer—but governance moves slowly, and there is always a black market. A black market in AI could be devastating. We are also talking about a cultural shift: a transformation in how we understand work, identity, authenticity, and human worth.

We need a strategy for AI – not for companies, but for individuals. A strategy is a series of directed decisions that help us reach a preferable outcome. Our preferred outcome is to have agency: a higher level of influence over the course of our lives. AI tends to give us the feeling that we are in control of it, which accounts for some of its popularity. Easy control, however, is illusory. True control is difficult; it requires us to be accountable for the problems caused by the AI systems we use. All three scenarios suggest we do not have the skills we need to control AI. We need to start building those skills now to be ready by 2032.

The students who created these scenarios will graduate into whatever future emerges. They will build it, navigate it, and—if they are thoughtful—shape it. The rest of us can do the same.

Art Kleiner

Subscribe now