Putnam Against Putnam

In the [YOU] on AI Field Guide

The cycle is built around a refusal of the comfortable answer—the orange pill is the willingness to see clearly even when seeing clearly costs you something. Putnam Against Putnam is that willingness in its purest philosophical form. The AI field is polarized into camps that have, in effect, pre-committed: one has decided that scaling will deliver genuine intelligence and reads every advance as confirmation; the other has decided these systems are sophisticated mimics with no inner life and reads every failure as proof. Both treat their position as an identity to be defended rather than a hypothesis to be tested. Putnam's career is a standing rebuke to this entire mode of argument.

The lesson is methodological, not merely a counsel of humility. Putnam's reversal happened because he had built both the theory and the argument that refuted it, and he could not, in honesty, hold both. The application is to construct, as deliberately as possible, the strongest case against whatever one currently believes about machine minds—and then take that case as seriously as the case for. If you believe these systems understand, you owe yourself the most forceful version of the argument that they do not, and you must be prepared to be moved by it; if you believe they are empty, you owe yourself the strongest argument that the boundary between sophisticated function and genuine mind is thinner than you assume.

The concept also carries the cycle's stance toward certainty itself. The questions Putnam wrestled with—what meaning is, what thought is, whether the mental reduces to the physical—are at the frontier of what inquiry can reach, and the machines are new and changing under us, which makes the questions harder still. The appropriate posture is not confidence but disciplined uncertainty: strong views held provisionally, as the best current account rather than the final word, with a clear sense of what would change one's mind. That combination of substantive commitment and genuine revisability is the rarest thing in the discourse, and it is the spectacle this concept preserves.

Origin

Putnam built functionalism in the 1960s and watched it become close to the consensus in philosophy of mind. The theory said a mental state is fixed by its functional role—its pattern of internal causes and effects—and it remains the implicit creed of much of the AI research community. The instrument of its undoing was his own Twin Earth argument, which showed that meaning, and therefore mental content, is not fixed by anything internal: your molecular twin, internally identical to you, means something different by "water" because the world he is embedded in is different.

From this Putnam drew the conclusion that two systems can share every functional state and still differ in what their thoughts mean—so functional organization cannot determine mental content. He pressed a second line as well, turning multiple realizability against its maker: mental kinds are themselves multiply realizable over functional kinds, so identifying the mind with a program fails for the same reason identifying it with a brain state had failed. He had, in effect, refuted himself twice over with a single body of work.

It would be a mistake to read this as a retreat into mysticism, and the distinction matters for how the concept applies to AI. Putnam did not embrace dualism or claim the mind is made of special non-physical stuff. His objection was conceptual: the vocabulary of computation, however powerful for describing brains as physical systems, is simply the wrong vocabulary for capturing what we mean by mind, because meaning, reference, rationality, and justification are normative and world-involving, not facts about syntax. Later he softened into a "liberal functionalism," allowing that the right functional capacities, suitably embedded, matter—but he never returned to the clean, self-contained machine-table picture of his youth.

Hilary Putnam