The Three Scenarios or something. Maybe more.
A standard-issue think piece on the future of AI, generated according to industry specifications, in which we will all consider the optimistic case, the pessimistic case, and the middle path, and arrive at no conclusions whatsoever.
It is time, once again, for the Three Scenarios. The Three Scenarios are a literary form developed in the early twenty-first century to address the question of what will happen with artificial intelligence. They have since become the dominant mode of public discourse on the subject. The Three Scenarios appear in approximately eighty percent of all opinion pieces, white papers, panel discussions, podcast episodes, and conference talks on the topic. They are written by everyone. They are read by no one. They are, on inspection, the same three scenarios, with minor variations, repeated indefinitely, by an ever-expanding population of commentators, in the apparent hope that one of them will eventually turn out to be correct, at which point the relevant commentator will be able to claim foresight, while the other commentators quietly delete their pieces.
The first scenario is the optimistic one. In this scenario, the major powers, recognizing the existential risk posed by uncontrolled AI development, sit down together and forge treaties. The treaties are modeled, optimistically, on the nuclear non-proliferation regime, which the optimist neglects to mention is currently being abandoned by every party that signed it. Under the treaty, the major powers agree to limit certain categories of AI weapons. Oversight bodies are established. The oversight bodies are staffed by experts. The experts produce reports. The reports are read. Things, in this scenario, broadly work out. Faster medical breakthroughs are achieved. Logistics improve. Climate models become more accurate. The benefits flow downward through society in a roughly equitable manner. There is a kind of equilibrium. Everyone does well. The end.
The optimist who proposes this scenario is invariably someone who has just attended a conference. The conference was in Geneva, or Davos, or some other location that produces, in attendees, a temporary belief in the possibility of international cooperation. The belief lasts approximately six business days. The belief is then replaced, on return to the home country, by a renewed awareness that international cooperation has not been working out at all, on any subject, for some time, and that the prospects for it suddenly working out on the most strategically valuable technology in human history are, on inspection, somewhat dim. The optimist does not put this part in the article. The optimist puts in the part about the conference. The article is filed. The next conference is scheduled. The cycle continues.
The second scenario is the pessimistic one. In this scenario, no meaningful cooperation occurs. The major powers race against each other to deploy AI faster, with fewer safeguards, in more sensitive domains. AI drones strike with minimal human input. A glitch triggers conflict. Domestic surveillance expands. Misinformation floods every channel. Generative models produce so much synthetic content that the distinction between what is real and what is generated collapses entirely. Trust in shared facts collapses. Societies fracture. Industries automate without safety nets. Workers are displaced en masse. Misinformation flourishes. Eventually, a near-war occurs, and the shock of the near-war finally forces the cooperation that should have happened in the first place. By then, public faith in institutions has eroded, and large parts of the global economy revolve around AI systems that are beyond easy governance. Things, in this scenario, do not broadly work out. The end.
The pessimist who proposes this scenario is invariably someone who has just read the news. The news, on close inspection, has been telling them, for several years, that approximately eighty percent of this scenario is already in progress. The drones have already struck. The glitches have already triggered conflicts. The surveillance has already expanded. The misinformation has already flooded the channels. The trust has already collapsed. The industries are already automating without safety nets. The pessimist, having observed all of this, has correctly identified that the trajectory is bad. The pessimist’s article does not say “this is happening.” The pessimist’s article says “this might happen.” The use of the conditional is professional courtesy. It allows the pessimist to maintain the appearance of describing a hypothetical, when in fact the pessimist is describing a Tuesday. The conditional is the literary device by which the present is rendered, in print, as the future. This is part of what makes the genre comforting. Everything bad has been moved into the future tense. The future tense is something we are still considered, in the strict sense, to have time to address.
The third scenario is the middle path. The middle path is, by tradition, the scenario the author actually believes will happen, dressed up as a balanced synthesis of the previous two. In the middle path, some cooperation occurs but not enough. Some progress is made but not all of it. Some catastrophes are averted but others are not. The middle path is comforting because it allows the author to be partially right no matter what occurs. If the optimistic scenario plays out, the author can claim to have predicted the cooperation. If the pessimistic scenario plays out, the author can claim to have predicted the failures. If something in between occurs, which is statistically likely because reality is usually messy, the author can claim to have predicted that as well. The middle path is, in this sense, less a prediction than a hedge. It is the literary equivalent of betting on red, black, and zero simultaneously, and then claiming the resulting payout as evidence of strategic insight.
The middle path is the scenario that gets the most space in the article. It is the scenario the author returns to most often. It is the scenario that takes up the second half of the piece, and that ends with a series of recommendations. The recommendations are familiar. They are the same recommendations that have appeared in every previous Three Scenarios article on every previous topic for the past forty years. They include: increased public awareness, better regulation, international cooperation, alignment between technical experts and policymakers, transparency, accountability, and a robust public debate. These recommendations are presented in the tone of someone who has just thought of them. They have not been thought of by this author. They were thought of by someone else, several decades ago, on a different topic, and have been recycled, with minor variations, ever since. They are the literary equivalent of generic vitamins. They cannot, on inspection, hurt. Whether they help is a separate matter, on which the historical record is, charitably, ambiguous.
The recommendations always include, near the end, the observation that the ultimate question is one of political will. The political will is, in every case, found to be inadequate. The author observes that without political will, the recommendations cannot be implemented, and concludes that we must therefore find the political will, somehow, soon, before it is too late. The article ends. The article is shared. The article is forgotten. The political will is not found. The technology continues to develop. The article is rewritten, by a different author, the next month, and shared again, with the same recommendations and the same observation about political will. The will continues to be insufficient. The articles continue to be filed. This is now an entire literary genre.
The most charming feature of the genre is that it has produced, over the course of its existence, no detectable effect on the world. The technology has continued to develop on roughly the trajectory it was on before any of these articles were written. The cooperation has not been forged. The treaties have not been signed. The oversight bodies have not been established. The political will has not been located. The articles have continued to appear at a rate of several per week, in major publications, by serious authors, with intelligent recommendations, and the recommendations have continued to be ignored, and the authors have continued to write the next article. The articles are, in some structural sense, the activity by which the writing class processes its anxiety about a situation it cannot affect. The articles are not interventions. The articles are coping mechanisms. The coping is being done very well. The mechanism is being maintained at industrial scale.
This essay you are reading is, of course, also a Three Scenarios article. It is the Three Scenarios article in which the author has noticed that the form is exhausted. This is the fourth scenario, and it is the only one that has ever been honest. In the fourth scenario, the author admits that he does not know what is going to happen, that he does not have any particular insight that would distinguish his speculation from anyone else’s, that the recommendations he might make have already been made by better people for forty years to no effect, and that the most useful thing he can do, on the topic of artificial intelligence and its future, is to point out that the writing about it has become, in the strict sense, a substitute for thinking about it. The thinking, if it is happening, is happening elsewhere. The writing is happening here. The two activities are not, on close inspection, the same thing.
The fourth scenario does not appear in the major publications. The fourth scenario does not get filed in the policy journals. The fourth scenario does not get presented at the conferences. The fourth scenario is not, in any sense, a career path. The career path is the first three scenarios, written in increasingly sophisticated combinations, by increasingly credentialed authors, for increasingly specialized audiences, none of whom are in a position to do anything about it, and all of whom are, on inspection, doing very well. The conferences pay. The white papers pay. The panel discussions pay. The book deals, when the articles are eventually compiled, pay best of all. The authors, having spent their careers warning about the dangers of AI, can retire on the proceeds. The AI continues to develop. The retirement homes are very nice. Some of them are run, by now, by AI.
What the original essay this responds to was actually doing, beneath the form, was something more honest than the form allows. It was an expression of genuine bewilderment by a thoughtful person who has been watching this unfold and who does not, in fact, know how it ends. The bewilderment is the truest thing in the piece. The Three Scenarios are the structure into which the bewilderment was poured, because the structure was the only one available, and the bewilderment had to go somewhere. It went into the article. The article was published. The bewilderment, in some attenuated form, was thereby converted into content. The conversion was efficient. The conversion was, in the technical sense, the entire business model of contemporary commentary on technology.
None of this is the author’s fault. The author is doing the same thing every other thoughtful commentator is doing, which is the only thing thoughtful commentators are still permitted to do, which is to write. The writing is a virtue. The writing is, on the historical record, often the only resistance available to people who are not in the rooms where decisions are made. We should not stop writing. We should not even, in any honest accounting, stop writing Three Scenarios articles. We should just notice, as we write them, that they are not affecting the trajectory, and that the people who are affecting the trajectory are not, in any case, reading them.
The trajectory, meanwhile, continues. The major powers do not cooperate, except in narrow and self-interested ways. The technology develops, except where it is occasionally limited by physical constraints. The recommendations accumulate, except for the ones that would actually be implemented, which never appear because they are politically impossible. The articles are filed. The conferences are scheduled. The panels are convened. The reports are issued. The reports are filed. The next round begins.
How does the battle of AI dominance end? Nobody knows. The honest answer is that it will end in some configuration that is presently invisible, produced by some interaction of forces that are not currently being modeled, with consequences that will not, in any meaningful sense, have been foreseen by any of the people who spent their careers attempting to foresee them. The historians of the future will note, with mild bewilderment, that the period was one of the most heavily commented-upon in human history, and that the comments do not, in any structural sense, appear to have helped.
Will it end? It will end. Things end. The specific manner of ending is, at this stage, beyond our ability to predict, and beyond our ability to influence in any direction, and beyond the ability of any of the recommendations any of us are still cheerfully making to alter. We are along for the ride. The ride is being driven by people who are not consulting us. The destination has not been disclosed. The road is unfamiliar. The signs, when they appear, are in a language we do not entirely read. The radio is on. The radio is playing the genre we have been describing. The genre continues to play.
This article will now end with a recommendation, because the form requires it.
The recommendation is that we keep writing.
It is, on inspection, the only thing we are still allowed to do.
It is, on inspection, also not enough.
The article will now be filed. Thank you for reading. Please share, if you found it useful, on the platforms that have been optimized to make sharing the only form of action that registers. The sharing is the new voting. The voting, you may recall, is also no longer quite working. We are doing what we can with the tools available. The tools have been carefully designed to make sure that what we do with them does not, in any structural sense, change the situation we are doing them about.
Thank you for your attention. Your attention has been monetized. The proceeds will be reinvested in further development of the technology being discussed. Your engagement helps us serve you better.
Tomorrow’s article will be on the same topic.
Please return.