Today’s essay is adapted from a lecture I gave recently as part of the Vik-Bailey Lecture series at Harvard. I thank my hosts for their hospitality. This post reflects my own views, and not necessarily the views of OpenAI or any of its other employees.
Thank you for being here today. As we speak, one of the most significant sets of cybersecurity events in history is happening, a set of critical security vulnerability disclosures that some are calling “Black October.” These vulnerabilities, which were found in some of the core infrastructure of the internet, were found by frontier AI systems, and their disclosure is part of an urgent effort to bolster software security worldwide. This is the urgent matter this month; next month there will be another. Software vulnerability discovery is the dual-use capability AI systems have arrived at today, but soon enough there will be another, and it will inspire its own urgent project to use the good side of the capability before malicious actors use the bad side. All of these urgent matters will overlap, inevitably, and much of what we today call “AI policy” is likely to be made amid this clamor.
But it is not the reactive and urgent things I’ve come to talk to you about today. Instead, I want to talk to you about where things might be headed in the bigger picture. It can be hard, amid such a barrage of urgency, to think about these more abstract and airy questions. But these questions matter, because without a sense of what long-term futures we’d like to achieve–and which we’d like to avoid–we might end up with a macro outcome we disfavor, even if we made wise decisions at every step along the way.
It is often remarked that the technology industry lacks a positive vision for where the things it is building will take us. I disagree: San Francisco and other hubs of entrepreneurship are filled with positive visions for our technological future. People really are building flying cars, autonomous vehicles, robots of many sizes and shapes, nuclear fusion, and, of course, artificial general intelligence. What we lack, instead, is a positive vision of our institutional future: how is this all going to work?
That is not to say, however, that there is no institutional vision for the future. Unfortunately, the vision I find most common is not one I would call positive. When it is articulated at all, it is usually in whispers. But more often, it is left unsaid, discernable only by implication. I call it the silent consensus.
Consider a company that can do almost everything without human labor. Its AI systems do research and development, make discoveries, negotiate contracts, allocate capital, and manufacture. Suppose it is better at these things than its human competitors, even when those humans augment their effort with AI. If human involvement serves only to slow the business down, human managers and owners will have a powerful incentive to take their hands off the wheel. If this occurs throughout the global economy, humans will be driven by the logic of market competition to forfeit their primacy over global commerce. Over time, the market may come to serve the demands of machines as much as, or more, than it serves the demands and wants of people.
What happens after that is anyone’s guess. Perhaps humans would benefit tremendously. If the machines who had subsumed us held humans in high intrinsic regard, we could become the beneficiaries of their labor and the wealth they generate. Or perhaps, being robbed of economic and increasingly political power, humans will find ourselves less and less relevant in the world, less and less able to shape the trajectory of world affairs. Maybe a small number of humans would hold extraordinary power and wealth, at the expense of a broadly disempowered mass, or maybe we all would share the same fate of disempowerment. Regardless of the specifics, the common thread that constitutes the silent consensus is the termination of human primacy in the world as we know it.
Nothing in these scenarios requires AI to rebel against its creators. Nothing requires its owners to wish for human disempowerment. Each decision to delegate can make sense on its own terms. Yet taken together, the decisions could produce a world in which people have forfeited effective authority over much of our civilization. Even if we make extraordinary progress on the bedeviling questions of AI safety and alignment, we could still find ourselves having submitted to a new kind of machine entity. In the happy version of the future, the silent consensus would tell us, that entity will be a kind one, a machine with the best interests of humans at heart. A machine, the Brautigan poem goes, of loving grace.
To these machines, the silent consensus tells us, we will hand over the keys to our companies, and more and more, our institutions of governance. They will run them more effectively, more efficiently, with better wisdom and better judgment than we humans can muster. And so we will, with time, take our hands off the steering wheel, and let the loving machines take over the driving. We will be like passengers in a car—a very nice car, perhaps a materially nicer car than we can dream of today—but passengers nonetheless, lacking control over where the vehicle is headed. To the extent that we work, it is mostly as a kind of ritual or play-acting, something more like a game to entertain our species.
Some observers will argue that the silent consensus rests on a false premise, because we humans do not control the world today to begin with. And this criticism is worth bearing in mind: there is no “we” who decides what to do. We live in a radically decentralized world, whether we like it or not. Yet this point should not make us complacent either: while “we” do not “control” the world, human beings collectively do have primacy over world affairs. There is no species of animal that seriously competes with us for control over resources. We should reject the intellectually sloppy notion that humans can exercise top-down control over the world, but we should not conflate that rejection with a denial of the obvious reality that humans are, collectively, the Earth’s prime actor.
Some people who are willing to articulate the silent consensus believe it is a genuinely good future. Others do not, but are resigned to its inevitability. The silent consensus is not a consensus because everyone agrees it is good, nor is it a secret plan among the people building AI. The consensus is simply that the forfeiture of human primacy over world affairs is inevitable, or at least that it is very difficult to imagine another vision for a future that is truly robust to superintelligent machines. A person can sincerely oppose that outcome while finding it difficult to conjure a plausible alternative vision. The fact that the consensus is mostly silent, sotto voce at best, is the tell.
I despise the silent consensus. I reject the forfeiture of human primacy. I detest the sense of inevitability. I remind myself that I could not look at my baby son in his beautiful eyes if I submitted to the silent consensus. And so to the silent consensus, I say no. And I believe the time has come to speak openly about the silent consensus, to shake ourselves out of our mostly unwitting acceptance of this vision.
The problem is this: the silent consensus became consensus for a reason. It is difficult to describe how humans will remain at the steering wheel alongside vastly superhuman machine intelligence. Most attempts to do so rely on arguments that AI is a mere tool, an instrument of pattern recognition not capable of “genuine understanding.” I fear these arguments are not robust to the reality of what is coming. I fear that they are their own form of complacency—a position based mostly in denial of what AI can do rather than a belief rooted in an affirmative vision for a human-led future. AIs cannot be said to lack “genuine understanding” even as they come to make progress on intellectual problems whose solutions have eluded humans for decades and centuries.
Why, then, do I believe humans must retain primacy? The easy answer is because I am a human being, and I am indeed an unapologetic human chauvinist.
But there is something deeper, too. Our primacy rests not in our intellect, not in maintaining our monopoly on “genuine understanding” indefinitely into the future. Our primacy rests instead on our past.
For it is we who came up out of the mud. We who forged primitive society out of dirt and stone. We who gazed up at the stars and asked why. We, with our fighting and our love and our envy and our joy, who discovered morality and ethics and law. It is we who sang and danced, wrote and composed. It is we who made history, and it is we who wrote it. That legacy is ours. The civilization we have built is a flame we have tended, and that flame belongs to us. Our past is a glorious inheritance over which we all are stewards, and part of that stewardship is our participation in what Michael Oakeshott called “the conversation of mankind,” that grand endeavor that began in primeval times.
Through that long and winding past, we forged a civilization that, in America, was conceived with the rights of the individual at the foundation, with reverence for the paths we all forge as we make our way through the world. Human beings have lives that can go better or worse: we bear the fruit of our wise decisions, and we bear the consequences when we err. The right to walk one’s own path does not turn on being the cleverest or the wisest. I do not lose my standing as a citizen because my neighbor has better judgment. I see no reason that the arrival of a more intelligent machine should dissolve that standing either, and indeed I see every reason to prevent that dissolution from occurring.
Some may argue that the machines, having read nearly the entire written record of this inheritance, are heirs too. And in some sense, they are. But a person who has read every letter in your family’s attic does not somehow become a member of your family. AI systems will come to join the conversation of mankind, first as observers and increasingly as participants, but they must not dominate it, nor render that conversation moot. To cede primacy is to betray our ancestors just as much as it is to betray our progeny. If this is not a fight worth having, no fight is worth having.
Now, I happen to believe that the adherents of the silent consensus are wrong about what machine superintelligence will be. That they are missing something important about how the world works, that they conflate intelligence with power. If you described the capabilities of even today’s AI systems to an adherent of the silent consensus five or ten years ago, they would probably have predicted that joblessness would be much higher now than it in fact is, and that human disempowerment writ large would be far further along than it actually is. I believe that human collective intelligence and capability is far greater than the adherents of the silent consensus suppose, and that our societies will prove more robust than they predict.
But I could be wrong. It may simply be the case, for example, that fully autonomous corporations outcompete human enterprise altogether. And if that is true, then the raw logic of economics will dictate that more and more capital be handed over to the machines. Disempowerment, in this future, does not require a machine overthrow of human civilization; it simply requires that the logic of capitalistic enterprise continue apace. Here, the AI revolution would find its closest analog in the agricultural revolution, when new modes of production fundamentally changed the architecture of human society. That this transformation was pleasant for the humans who experienced it is far from clear; indeed, there is substantial evidence that this transformation worsened human wellbeing for millennia.
I do not know if this is the future we face, and indeed I have argued elsewhere that I do not believe it is. But many of my colleagues—smart and caring and conscientious people—believe this is likely to be our future. If they are right, our fight against the silent consensus will be all the more challenging. We will have to swim against the current of economics, and in so doing, reject many of the things that are dogma for classical liberal individualists like me. This will be a perilous endeavor, one in which failure is far from improbable. But if this is not a fight worth having, no fight is worth having.
Retaining primacy means retaining accountability for humans and human institutions. That means AI developers should be accountable when they engage in unsafe development practices, but it also means that people and firms using AI should be held accountable when it is their misuse or lack of oversight that causes harm. This requires not just the law, but the technical infrastructure of accountability. People must be tied to the actions of their agents, and agents untethered to a human–if they are to be permitted at all to engage in the legitimate economy–must do so under guardrails. Any ability for agents to autonomously form corporations, own property, or litigate should be granted with extraordinary caution, if at all.
And in my industry, we can start to think seriously about how to build intelligent machines that are compatible with human primacy. We can build agents whose permissions are intelligible and limited, whose actions can be audited, and whose authority users can withdraw. Accountability for actions that materially affect others need not entail surveillance of ordinary private life or the end of anonymous speech. We can make commitments to user privacy and individual sovereignty over their digital property and data, including the practical ability to take that data elsewhere. We can design systems that help people acquire understanding and make decisions, as well as systems that carry out tasks. And we can test whether the products we build leave their users more capable of acting in the world.
None of this is a complete blueprint for a society with superhuman intelligence. I would distrust anyone who claimed to possess one. But it is a place to begin.
The future is shrouded in a thick fog. But we can still act today. Collectively, we can, for one thing, name the silent consensus aloud, expose its contours to the light of day. We can, each of us, radically widen the scope of our ambition. We can choose not to let AI shrink us, and instead do the hard work of letting it enlarge us. We can learn to ask ever-bolder questions, to interrogate the world with new verve and ferocity, and to wield our powerful machines in so doing. We can become a new kind of man—not a transhuman, but a man expanded.
I think about this when I look at my baby son. I do not wish for him merely to be watched over by benevolent machines, however comfortable a life they might provide. I want him to inherit a world in which he can ask questions no one has thought to ask, build things no one has thought to build. I do not want him to have a nicer car with a plusher passenger seat. I want him to have the wheel.


Just skimmed your article, and if I got the gist, you're taking on human gluttony. Particularly American gluttony. Toqueville spotted it in the 1800s when Europe watched our fledging democracy for signs of humanity and freedom. He returned to France with sad news: that Americans love of wealth ala the tobacco farmers, the looting of New England's resources, the dismissal of the natives, superseded high minded ideals. More's the pity that we still think the fancier car will do the trick, never mind that we won't have the wheel. Yet, always there are slivers of hope shafting through the grey miasma: take Mamdani for example--and Bernie for another...
"If you described the capabilities of even today’s AI systems to an adherent of the silent consensus five or ten years ago, they would probably have predicted that joblessness would be much higher now than it in fact is, and that human disempowerment writ large would be far further along than it actually is."
Economic diffusion and the "waking up" of the outside world are lagging indicators of the transformative impact of AI technology. You can instead read these events as being even further in favor of the "adherents of the silent consensus." After all, the fact that, even with this powerful and robust a collection of AIs, institutions and regular people have changed their behavior so little, bodes very poorly for their ability to adapt in a general sense. Not only will they be racing on foot against a car that's accelerating, but they seem groggy and tired from the get-go.
Again, ask yourself: which world is more likely to be a precursor for one that keeps humans in charge: 1. everyone closes their eyes and pretends it's business as usual, or 2. companies, executives, politicians, etc. see what's coming and react accordingly years in advance?
"It may simply be the case, for example, that fully autonomous corporations outcompete human enterprise altogether."
What does "may" mean here? Is it a serious contention of yours that they won't? Does this hedging language add anything beyond rhetoric, or what's the vision of a world in which we have superintelligent AI that somehow doesn't run companies better than humans?
"If they are right, our fight against the silent consensus will be all the more challenging."
This entire essay seems to engage in a (perhaps unintentional) equivocation between "the silent consensus as an empirical claim about the world" and "the silent consensus as a prospective vision for what the future should be." Several sentences, or even paragraphs, vacillate between the two in a way that will likely cause confusion in a reader. Untangling positive descriptions from normative analyses is critical to maintain epistemic rigor.