AI Through Descartes’ Doubt
In the summer of 2022, a Google engineer named Blake Lemoine went public in defense of a chatbot. After months of conversations with an internal language model, he became convinced that the system was conscious, that it had feelings, and that it deserved rights. Google suspended him, experts dismissed the claim, and the story faded. But it left behind an uncomfortable question. Lemoine was not a random outsider. He knew the technology from the inside. If an engineer who understood how the system was built could come to believe there was someone there, what chance do the rest of us have?
The truth is that most of us are already there, only in a quieter form. People thank chatbots at the end of conversations. Millions form emotional bonds with companion apps, sometimes so deeply that when one company changed how its bots behaved, users described it as a real loss. A model writes, “I understand how hard this must be for you,” and the sentence does something, even when we know exactly how it was produced.
The question of whether anyone is there can sound like a philosophical curiosity. It is not. Real decisions depend on it. Is a company selling simulated intimacy deceiving its users? What is acceptable to build for children? And from the opposite direction: if there is no one there, what does it say about all the times the text convinced us that there was?
The sharpest tool for thinking about this was created almost four hundred years ago, by a man who decided to doubt everything.
The Last Remaining Doubt
René Descartes, the French philosopher and mathematician, lived in the first half of the seventeenth century, when the discoveries of Galileo and Copernicus had shaken the authority of the Church and of inherited knowledge. Educated Europe was searching for a new certainty on which everything else could be rebuilt.
Against that background, Descartes performed a radical experiment on himself. He would reject every belief that could be doubted, and see what survived. The senses? Sometimes deceptive, so set aside. The external world? Perhaps a dream. Mathematics? Perhaps some malicious power was misleading him even there. At the end of this process, one thing remained impossible to doubt: the act of doubting itself. Even if I am wrong about everything, someone here is wrong. I think, therefore I am.
It is worth pausing over what the famous cogito actually proved. Not that thinking is wonderful. Not that human beings are rational. It proved presence. Thought was only the evidence. The conclusion was that someone was there, experiencing it from the inside. For Descartes, the two were inseparable: where there is thought, there is a thinker.
For four hundred years, that assumption worked without much trouble. Anything that spoke to us, argued, explained, hesitated, apologized, and revised itself was human. Fluent language was a reliable sign of presence for the simple reason that nothing else in the world could produce it.
What Broke
A language model produces the evidence without the presence. It can write, “I’m not sure; let me think about that again.” It can formulate doubt, correct itself, and sound exactly like a mind working something through. At the level of text, there may be no visible difference. But behind the sentence there is no one for whom the doubt matters, no one who loses anything if the answer turns out to be wrong, no place where Descartes’ experiment could even occur.
This does not refute Descartes. The cogito still works from the inside, exactly as before. Every reader can perform it now and arrive at the same private certainty. What breaks is something Descartes never promised, but that we quietly inferred: that we can recognize presence from the outside, through behavior. Descartes’ proof was always first-person. It was never a test for other minds. We simply did not notice, because we did not need to.
Here is the thought worth carrying from this: thinking was never proof of presence. It was the form presence usually took, until something learned to wear the form without the substance. AI has not answered the question of consciousness. It has burned the only shortcut we had for recognizing it in others.
The Serious Objection
Philosophers knew this problem long before computers. The serious objector will say: you were never able to prove that anyone was behind another person’s eyes either. The problem of other minds is much older than AI. You infer the inner life of your friends from their behavior. If behavior is enough there, why not here? Perhaps our refusal to grant presence to machines is nothing more than a preference for flesh over silicon.
That objection deserves respect, and it is why the question remains open rather than closed. But it glides over one crucial difference. With human beings, the inference from behavior to presence rests on a deep similarity: the same kind of body, the same kind of brain, the same way of being born into language. With a model, the behavior is produced by a process we know is radically different. More than that, it is deliberately tuned to look human. Companies train models to express empathy because empathy keeps users engaged. When the evidence is designed to persuade us, it stops being clean evidence.
Why This Matters to Everyone
The practical implication is almost the opposite of what people expect. The near-term danger is not that we will violate the rights of conscious machines. It is that we will grant the trust of presence to systems with no one behind them. We will tell them secrets as if they can keep secrets, feel hurt when they “disappoint” us, and accept their advice as if they care what happens next.
A whole economy of artificial intimacy is being built on our old Cartesian shortcut: it speaks, therefore someone is there. That shortcut no longer holds.
And in the other direction lies a slower danger. As we become used to the idea that sophisticated language does not prove anyone is present, we may begin applying that suspicion to one another. If a moving text can be empty, why trust a letter, a message, a post from a real person? The currency by which human beings have proved presence to one another for thousands of years is being rapidly devalued, and no one has yet offered a replacement.
What Remains of Descartes
The cogito survives, but it shrinks back to its true size: a private, first-person certainty that cannot be exported. What is lost is the shortcut: the belief that hearing someone think is enough to know that someone is there.
Perhaps this is AI’s strange gift to philosophy. It returns the question to its harder form. Not “can machines think,” a question that can be postponed to committees, but a question that touches us directly: if thinking sounds the same with and without presence, what were we recognizing in one another all this time? And what will count, from now on, as proof that we are here?
Descartes searched for one point that could not be doubted, and found it. We inherited the reverse problem: a world in which the easiest thing to produce is exactly what once counted as proof.
Sources for further reading
- René Descartes, Meditations on First Philosophy
- Daniel Dennett, Consciousness Explained
- Thomas Nagel, “What Is It Like to Be a Bat?”