Hilary Putnam

In the [YOU] on AI Field Guide

The cycle holds that what we see in the machine is decided by what we already believe about ourselves. Putnam's gift is to make us suspicious of those beliefs. He showed that our most natural assumption about the mind—that it is a self-contained inner thing, that meaning and selfhood live inside us, that fluency is understanding—is false even for human beings. Carry that corrected picture to the machines and the questions improve: not "does it sound like it understands?" but "what is it connected to, and is the connection the right kind?"

His externalism lands directly on the cycle's anxieties about large language models. A model has consumed more text about water than any human could read in a thousand lifetimes, yet it has never tasted water, never been thirsty, never stood at a lake. On Putnam's account this is precisely the situation in which reference is in doubt—the system is a master of the symbol whose contact with the substance is, at best, third-hand. The fluency is real; whether the words reach the world is the open question, and he supplies the apparatus for asking it without licensing an easy answer in either direction.

He stands in the cycle's gallery as the patron of disciplined uncertainty. Where the boosters declare the machines conscious on the strength of their fluency, and the skeptics declare them empty with a confidence the arguments do not license, Putnam models a third thing: build the strongest case on each side, and be ready—as he was, against his own life's work—to be moved by whichever the evidence favors. The orange pill, in this series, is the refusal to take the comfortable answer; Putnam took the uncomfortable one even when it cost him his own theory.

Origin

Born in Chicago in 1926 and trained in the rigorous postwar tradition of analytic philosophy, Putnam was a serious mathematical logician before he was a philosopher of mind, which is why his arguments have the hard edges of proofs rather than the soft contours of opinion. He held chairs at Princeton, then MIT, then Harvard, where he taught for thirty-five years. He was also, for a time in the late 1960s, a committed anti-war radical, an affiliation he later called a mistake; in his last decades he returned, deliberately and movingly, to the Judaism of his ancestry. The pattern is consistent: Putnam took positions all the way, and then, when the evidence turned, he turned with it.

His most consequential idea was that the mind is to the brain roughly as a program is to the computer running it. In papers from "Minds and Machines" in 1960 to "The Nature of Mental States" in 1967, he argued that what makes a mental state the state it is, is not its physical composition but its functional role—the pattern of causes that produce it and the effects it produces in turn. Pain is whatever plays the pain-role, in any creature or contraption. The model he reached for was the Turing machine, and the picture he drew severed the mind from any particular biology, opening a materialist path that did not tie thought to any one kind of physical thing.

Then, in 1988, in a book pointedly titled Representation and Reality, the founder of functionalism set out to refute it. The reversal grew from his own work on meaning: externalism—the thesis that meaning is not in the head—had shown that mental content depends on the world and the linguistic community, not on anything internal, and so functional organization could not determine what a thought is about. There is almost no parallel for this in the history of philosophy: a thinker dismantling, in public and in detail, the monument he had spent his youth erecting. He died in March 2016, before the systems that now dominate the conversation existed in their current form.

Key Ideas

Functionalism and multiple realizability. The same mental state can be realized in physically different systems—pain is felt by octopus and human alike, so it cannot be identical to any one brain state, only to a functional role. This is the argument that buried the identity theory and that, more than any other, licenses the idea of a machine mind. But multiple realizability establishes a possibility, not an actuality: that a mind could be made of silicon does not mean that any given model is one.

The brain in a vat. A brain wired to a supercomputer that feeds it an entire simulated life cannot coherently think "I am a brain in a vat," because its word "vat" was acquired through stimulation and refers only to vat-images, not real vats. The supposition is self-refuting. A text-trained model is, in a suggestive sense, a brain in a vat—its only contact with the world runs through symbols—and the argument poses, rather than settles, whether its words reach the world through the human authors who wrote its corpus.

Twin Earth and semantic externalism. On a planet identical to ours except that its "water" is XYZ rather than H₂O, your molecular twin means something different by the word, though you are internally identical. Cut the pie any way you like, meanings just ain't in the head. The model's mastery of how "water" is used is word-to-word information; whether it knows what water is requires word-to-world relations that usage alone cannot supply—the rigorous form of the symbol-grounding problem.

The division of linguistic labor. Most speakers cannot tell gold from fool's gold; they mean "gold" correctly by deferring to experts and to the world through a social chain. This keeps the externalist verdict honest: grounding can be indirect, and a model is plugged into that same chain at one further remove, through the embedded speakers whose words anchored its training text. Whether that inheritance is thick enough to count is genuinely open—the right apparatus for arguing it, not a license to assume the answer.

The honesty of changing one's mind. Putnam built the dominant theory of mind and then let the strongest objection win, even when it was fatal to his own reputation. This self-refutation models the cognitive virtue the AI debate most needs and most rarely gets: construct the strongest case against whatever you currently believe, take it as seriously as the case for, and revise when the arguments require.