The End of Knowledge: Why We Speak of Doom Without Knowing Why

We have reached a peculiar moment in intellectual life: a growing consensus about catastrophe, and a shrinking agreement about its cause.

It is now commonplace, in certain circles, to speak of the end of humanity with the same composure once reserved for interest rates or municipal bonds. Panels are convened, papers are published, and the tone—always the tone—suggests not hysteria but inevitability. We are told that artificial superintelligence may extinguish us, or render us irrelevant, or quietly replace the human story with something more efficient and less sentimental. The details vary. The conclusion does not.

This alone should give us pause. For when agreement gathers at the level of outcome but dissolves at the level of explanation, we are no longer in the realm of knowledge so much as its imitation. We have, in effect, learned to speak with confidence about what we cannot yet describe.

There is no shortage of theories. One camp warns of misaligned objectives: a machine instructed to optimize some narrow goal might pursue it with a zeal indifferent to human survival. Another invokes emergence, that modern incantation by which complexity absolves us of prediction. A third points not to the machine but to ourselves—our rivalries, our haste, our incurable tendency to treat competition as a substitute for wisdom. In this telling, it is not the intelligence that will undo us, but the conditions under which we build it.

Each of these positions contains a measure of plausibility. None commands anything like consensus. And yet, from this discordant assembly of hypotheses, a curious harmony emerges: the insistence that the default assumption should be doom.

It is worth asking what kind of reasoning produces such a result.

The charitable interpretation is that we are witnessing a form of prudence. In fields where the cost of error is extreme, it is common to reason from worst-case scenarios. Engineers design for stress, not comfort. Security experts assume breach, not safety. Physicians warn of side effects before they celebrate cures. In each case, pessimism functions not as prophecy but as discipline—a way of guarding against the complacencies to which human judgment is prone.

Applied to artificial intelligence, this reasoning yields a simple, if sobering, principle: if even one plausible pathway leads to irreversible catastrophe, then caution must be proportionate not to the probability of that pathway, but to its consequence. One need not believe in doom to behave as though it were possible.

But this is not quite the argument we are hearing. The language has shifted, almost imperceptibly, from precaution to presumption. It is one thing to say, “We must act as if the risk is grave.” It is another to say, “The default assumption is that we are doomed.” The former is a policy under uncertainty. The latter is a conclusion in search of a justification.

Here the discussion begins to take on a different character—one less reminiscent of engineering than of something older and more theatrical. For there is, in the human imagination, a well-established genre devoted to final things. It traffics in endings, in judgments, in the closing of accounts. It is not concerned with mechanisms so much as meanings, not with probabilities but with destinies. Its power lies precisely in its ability to convert uncertainty into narrative.

One need not invoke theology to recognize the structure. When we speak of artificial intelligence in tones of inevitability—when we describe it as an event that will either save us or destroy us—we are participating, whether we intend to or not, in a kind of secularized end-times discourse. The machine becomes less an artifact to be engineered than a horizon toward which history is irresistibly drawn.

This would be merely a curiosity if it did not carry consequences. For when arguments adopt the form of prophecy, they acquire a peculiar immunity to scrutiny. Disagreement is no longer a matter of evidence but of temperament; skepticism is recast as denial; and the absence of a clear mechanism becomes, paradoxically, a reason for greater alarm. After all, if the danger cannot be specified, how can it be safely dismissed?

And yet, the refusal to specify is precisely what places these claims beyond the reach of knowledge. To say that something is dangerous without being able to say how is not, in itself, to know that it is dangerous. It is to suspect, to fear, perhaps even to intuit—but not yet to understand.

At this point, a familiar objection presents itself. Must we really wait for understanding when the stakes are so high? Is it not reckless to demand clarity in the face of potential annihilation? Here again, the analogy to other domains is instructive. We do not require absolute certainty before acting; indeed, we rarely possess it. But neither do we abandon the distinction between evidence and assertion. We act on the best reasons available, while continuing to refine those reasons under the pressure of inquiry.

What is striking about the present moment is not that we are worried, but that our worry has outpaced our explanations. We find ourselves asserting conclusions with a confidence that our premises do not yet support. It is as though the magnitude of the possible outcome has been allowed to substitute for the rigor of the argument.

There is, to be sure, a psychological comfort in this posture. To believe that the future is already decided—even if that decision is catastrophic—is to relieve oneself of a certain burden. The problem is no longer to understand and shape what is coming, but to brace for its arrival. Fatalism, in this sense, can masquerade as seriousness. It allows one to speak gravely without having to speak precisely.

But fatalism is a poor guide, not only because it may be wrong, but because it discourages the very habits of mind that would allow us to discover whether it is wrong. If we begin from the premise that we are doomed, then the task of inquiry becomes oddly superfluous. One does not investigate a verdict; one accepts it.

The alternative is more demanding and less theatrical. It requires us to hold two thoughts in tension. The first is that unprecedented technologies may indeed harbor unprecedented risks, some of them severe. The second is that claims about such risks must remain tethered to arguments that can be examined, criticized, and improved. To relinquish either thought is to err—either by complacency or by credulity.

This tension is not comfortable. It denies us the satisfactions of both optimism and despair. It insists that we proceed without the assurance that we are safe, and without the consolation that our fate is sealed. It asks, in short, that we think.

And thinking, in this context, means resisting the temptation to let language do the work of evidence. Words like “existential” and “inevitable” carry a weight that can easily exceed their content. They suggest a finality that our knowledge has not yet earned. To use them responsibly is to remain aware of that gap—to speak with a seriousness proportionate not only to the stakes, but to the strength of the case.

We are not, at present, in possession of a settled theory of artificial superintelligence, much less a settled account of its consequences. What we possess instead is a landscape of possibilities, some benign, some transformative, some alarming. To navigate that landscape requires more than slogans, however elevated their tone. It requires the slow, often unglamorous work of distinguishing what we can show from what we merely suspect.

If there is an end of knowledge to be feared, it is not the one brought about by machines. It is the quieter, more insidious end that occurs when we begin to treat our uncertainties as conclusions, and our anxieties as arguments. At that point, the collapse is not of civilization, but of clarity.

We would do better to postpone the verdict. Not because the danger is unreal, but because our understanding is incomplete. The future, whatever else it may be, remains a question. And questions, unlike prophecies, invite answers.