Mark Zuckerberg is the most recent contributor to a shelf that is beginning to sag under the weight of its tomes. Indeed, at this point it is almost easier to point to the tech titans who haven’t written a manifesto than to list the ones who have. The genre includes Sam Altman’s The Intelligence Age and
There is something very peculiar about all of this.
Traditionally, manifestos have been written by outsiders, by iconoclasts seeking to overturn the established order. The best illustration is perhaps the most famous manifesto ever written. When Karl Marx wrote The Communist Manifesto with Friedrich Engels, he was living in exile in Brussels, his activities heavily curtailed and under surveillance; barely a week after the manifesto was published, he was arrested and expelled from Belgium. More recently, the
But the authors of the AI manifestos are not outsiders banging on the gates. They run some of the richest and most powerful institutions in the world. They command enormous pools of capital, employ thousands of people, and have unusually direct access to governments. Manifestos are the weapons of the powerless, and these men are anything but.
So what is going on? Read the texts together and an answer starts to emerge—one that is both basic and important. In brief, it is this: for all their power, the authors do not control what happens next. They still need something from the rest of us. And the manifestos reveal what that is.
Zuckerberg begins his essay by claiming that superintelligence is a given in the next few years. Amodei says that AI’s “creation was probably inevitable the instant humanity invented the transistor, or arguably even earlier when we first learned to control fire.” And we have reached the point, he also says, where “the idea of stopping or even substantially slowing the technology is fundamentally untenable.”
The funny thing, however, is that if something is truly inevitable, it is unnecessary to write manifestos about it. When was the last time you encountered a manifesto about the sunrise? When did you read rich men—and it is men—waxing lyrical about the untold promise of our local star?
Never, right? And this tells us something important. In the words of Lady Macbeth, the manifesto writers protest too much. These manifestos profess inevitability precisely because AI’s development is not* *inevitable. And to understand why it is worth doing this, it is helpful to distinguish between two senses in which AI might be seen as inevitable.
The manifestos trade on an ambiguity between two senses in which AI might be inevitable, which I will call weak and strong inevitability.
Weak inevitability means something like the following: AI is not going back in the box. Existing capabilities, technical knowledge and commercial incentives will not simply vanish.
But the manifestos routinely move from that claim toward something much larger—superintelligence, enormous infrastructure buildouts, rapid deployment across the whole economy. These are the claims of strong inevitability, and they are very much *not *inevitable. They require the continuing mobilization of chips, energy, land, capital, regulation and, above all, customers; in a word, they remain subject to human choice.
Now, weak inevitability is more or less undeniable. But strong inevitability is simply untrue—even if we do in fact end up in one of the worlds that the manifestos describe, it is not currently inevitable that we must end up there. What has already been discovered may be impossible to reverse, but the scale, pace, and success of what gets built around it remain very much open.
And, as I’ve argued before, the future still—thank heavens—depends on us, on human choices. The discourse of strong inevitability attempts to obscure that fact. The interesting question is why its proponents are trying so hard.
There is a historical parallel that illuminates why the advocates of strong inevitability are becoming increasingly shrill, and it is one that the manifestos themselves keep reaching for: nuclear technology. Andreessen speaks of nuclear power as a silver bullet, and says that stopping “the large-scale deployment of civilian nuclear power” was “perhaps the most catastrophic mistake in Western society” of his lifetime. Amodei puts nuclear first among technologies that “work well in a technical sense” but were stunted by “regulations or misplaced fears.”
Now, nuclear technology is not like Roman concrete. Humanity did not lose the knowledge required to use it. Nuclear reactors continue to run today. But the trajectory of nuclear use is very different to what it could have been. Rather than a blanket and widespread adoption of nuclear technology, we have something much more limited and fragmentary. For instance, in the 1960s the Atomic Energy commission predicted that “more than 1,000 reactors would be operating in the United States by the year 2000”; as of March, the actual number was 96.
It is not difficult to see the similarity to AI—a technology of unparalleled power, restricted and heavily constrained because of social objections to its danger. For the writers of the manifestos, and for everyone whose books, balance sheets, and IPOs have been staked on the strong inevitability thesis, this would be a disaster. But a society’s decisions should not be governed by the goal of preventing disaster for the kind of people who write tech manifestos.
Given that there is nothing inevitable about the exact shape of the future deployment of AI, the natural question is who gets to decide what that shape should be.
In a democratic society, the answer is not mysterious. Decisions about the shape of the economy, the concentration of power, and the disruption of hundreds of millions of livelihoods are exactly what democratic processes exist to decide.
In an ideal world, elections would turn on these questions, legislatures would reflect with intelligence and moral seriousness on the various interests at stake, and the public sphere would witness vigorous debate conducted with rigor, honesty, and mutual good will. And all of this would be done within a framework of free and independent institutions, rather than institutions beholden to all manner of different interests.
Sounds good, doesn’t it? And also completely unrealistic. That would be a nice world, but it is not the one we are living in. So while it is important to call for democratic renewal and to work toward it, it is also important to recognize that this renewal is not going to arrive on the timeline the technology is moving on.
And this, it turns out, places special responsibility on business leaders.
The AI manifestos come from individuals and institutions with profound commercial interests. The frontier labs, for instance, are companies that need customers, partners, and organizations willing to reorganize themselves around their products. The enormous bets that VCs, PE, hedge funds, bankers, and institutional investors are making on AI depend on commercial success—on companies buying what the AI vendors are selling.
This means that claims about the pace and scale of the buildout and deployment across the economy run through procurement departments, IT budgets, and board approvals. And this means that business leaders have more influence over the development trajectory of AI than most politicians, and certainly more than most individual voters.
Now, let me be clear. Procurement decisions are not substitutes for democratic decision-making. But in our non-ideal democratic situation, we have to accept that they are also not neutral choices. Deploying by default, on the vendor’s timeline and under the vendor’s framing, means voting for some particular version of the future.
And so, whether we like it or not, this means that business leaders face choices with substantial social implications. And leaders must be sure that they are actively making those choices rather than simply accepting the choices that the writers of manifestos are pushing them to make.
In a word, they must make sure that they are choosing the pace and the terms of the development that they are paying for.
None of what I have said is an argument against the development of AI, let alone a veto. For a start, the weak inevitability claim is true: the technology is here, and much of it is genuinely useful. Deciding to move fast, integrate deeply, and sign the standard terms may well be the right call for your organization.
Further, I am an optimist about the power that AI has to unlock human potential. Indeed, it is precisely this optimism that has made me write this piece: the potential of AI is so great that we should decide how to implement it, rather than letting the writers of manifestos do it for us.
And right now, because of the particular and peculiar conditions of our moment, it is business leaders who are on the front lines of making those decisions. They should make them intentionally.