The Folk Theory of the Soul

**EDO SEGAL:** Patricia, your most notorious idea, developed with Paul Churchland, is eliminative materialism — the claim that our everyday talk of beliefs and desires is a *theory*, folk psychology, and that like other folk theories it might turn out to be false and get replaced by neuroscience, the way the four humors and phlogiston were replaced. Lay it out, because it is about to collide head-on with a man for whom the soul is not a theory but the form of the body. And then I want to ask what it does to the word "understanding" itself.

**CHURCHLAND:** The thesis is simple to state and hard to swallow. When we explain a person — she stayed home because she *believed* it would rain and *desired* to stay dry — we are using a theory. Belief, desire, intention: these are posits, theoretical entities, like "humor" or "caloric" or "phlogiston" once were. They feel obvious and indispensable, but so did the humors, for two thousand years. The claim of eliminative materialism is that the brain may simply not contain anything that matches the crisp, sentence-shaped objects folk psychology posits — discrete beliefs with definite contents, combining by a kind of inner logic. The brain's processing looks to be distributed, statistical, continuous, nothing like a list of sentences you believe. If that is right, then "belief" and "desire" are a useful approximation that will not survive contact with the mechanism — not eliminated as phenomena, but redescribed, the way chemistry did not eliminate fire but replaced "phlogiston leaving the wood" with oxidation.

Now the part for tonight. If "belief" and "desire" are folk theory even for *human* brains, then "soul" is folk theory squared. It is the oldest posit of all, the inner ghost that does the believing and desiring, and it carves nature at joints I do not think nature has. And here is the sharp edge for AI: the entire field has imported this folk psychology wholesale and never noticed. We say the model *believes* a false fact, *wants* to be helpful, *intends* to deceive, *knows* the answer. Those words feel natural and they are doing enormous unearned work, in newspapers and in courtrooms. The friar's "the machine cannot truly understand" and the engineer's "the machine truly understands" are both leaning on a [folk concept of understanding](https://www.youonai.ai/fieldguide/med/five_kinds_of_understanding) that may not refer to anything real even in us. My deepest move is not to take one side of that question. It is to suspect the question is malformed.

**AQUINAS:** This is the most dangerous thing you have said, and I mean that as the highest compliment, because it goes under everything. Let me first agree, and then refuse. I agree that "belief" and "desire," as little sentence-shaped tokens stacked in a mind, are a crude picture, and that the actual workings of a creature are subtler than the folk talk. I have my own subtler machinery — the intellect, the will, the passions, the interior senses — and I never thought the soul was a warehouse of sentences. So if eliminative materialism only says the folk picture is coarse, I am an ally.

But you go further, and here I plant. You say the *concepts* of belief and understanding might refer to *nothing real*, that they might join phlogiston in the museum. Now turn that on the act you just performed. You understood my position well enough to attack it. You *grasped* what eliminative materialism claims, *judged* it true, and *desired* to persuade me of it. If understanding refers to nothing — if there is no real act of grasping — then there was no grasping of eliminative materialism, and your theory was never understood by anyone, including you, and so was never asserted, only emitted, like the machine's marks. The doctrine, taken to the bottom, eats the mind that holds it. Phlogiston could be eliminated because the *chemist* who eliminated it still understood combustion. You cannot eliminate understanding itself without eliminating the understander who does the eliminating. This is why I say the soul is not a posit like caloric. Caloric was a guess about something outside us. The grasp of meaning is the thing doing the guessing. You can be wrong about the heavens. You cannot be wrong about whether anyone is presently understanding, because the wondering *is* an understanding.

**CHURCHLAND:** That is the classic self-refutation charge, and it is the best one against me, so let me answer it carefully rather than wave it off. I do not claim there is *no* phenomenon. I claim the *folk concept* of it may be wrong about its own nature. When I redescribe "understanding" in terms of neural mechanism, I am not saying nothing happened when I grasped your point. I am saying that what happened was a particular brain process, and that the *concept* "grasp," with its implied immaterial grasper, mischaracterizes that process — the way "sunrise" mischaracterizes the earth's rotation without there being no dawn. There was a dawn. There was no rising sun. There was understanding. There may be no immaterial graspers. The self-refutation only bites if eliminating the *concept* eliminates the *phenomenon*, and it does not, any more than heliocentrism abolished morning.

**AQUINAS:** But notice the asymmetry, which is everything. In "sunrise," the phenomenon — the dawn — is fully described by the new account: the earth turns, light arrives, no residue. You have replaced the bad concept with a good one that captures all there is. My claim about understanding is that when you finish the neural description — these populations, this dynamics, this integration — there is a residue the description does not capture: that it is *grasped*, that the meaning is *present to* someone. If you are right, there is no residue, and "sunrise" is the perfect analogy. If I am right, there is a residue, and "sunrise" is a false comfort, because the rotation explains the dawn entirely but the firing does not explain the presence. So your analogy *assumes* there is no residue — assumes the very thing in dispute. We are, once more, at the one seam, wearing a new coat. Whether the full physical story leaves anything out.

**EDO SEGAL:** Before I mark the convergence, let me pull this down to where the reader lives, because eliminative materialism sounds like a seminar word and it is about to be law. The courts are starting to ask whether a model "intended" harm. Companies say their assistant "wants" to help and "knows" your preferences. Patricia, you are saying every one of those sentences is a folk-psychological loan that may never be repaid. So when a regulator asks "did the system intend to deceive," what should she ask instead?

**CHURCHLAND:** She should ask what it *did* — what representations it formed, how they were encoded, what process turned the input into the output, what it was trained to optimize and who checked. Those questions have answers you can investigate. "Did it intend to deceive" presupposes the folk machinery of belief and intention, and may be a demand for a fact that nature does not supply, in the machine *or* in us. Notice I am not letting humans off the hook here. When we say a *person* "intended" to deceive, we are also deploying a theory, and a good lawyer already knows how slippery intent is in a human skull. The machine just makes the slipperiness impossible to ignore. My counsel is the same for both: stop adjudicating the ghost, start describing the mechanism.

**AQUINAS:** And here, oddly, I am with her against the regulator's loose talk — but for the opposite reason, and the reader should see why two enemies arrive at one door. Patricia says do not ask whether the machine "intended," because intention may be folk theory that refers to nothing. I say do not ask whether the machine "intended," because intention is a real act of a rational will, and the machine *has no will to do the intending* — there is no one there to mean the deception, so the word misfires. She empties the word for everyone. I keep the word full and deny the machine is the kind of thing it can be said of. We forbid the same sentence. She forbids it because she doubts the ghost exists anywhere. I forbid it because I am certain the ghost is real and certain it is not in the machine. The regulator gets the same instruction from the saint and the eliminativist, which should tell her something.

**CHURCHLAND:** It should tell her that the mentalistic vocabulary is doing unearned work no matter which of us is right, and that is a more robust conclusion than either of our metaphysics. When the friar and I agree on the practical rule from opposite ends of the universe, the rule is probably sound.

**EDO SEGAL:** I want to name a convergence, because it is real and the reader should bank it. You both reject the warehouse-of-sentences soul. You both think the folk picture of the mind is too crude. You both refuse to let the engineer say "the machine believes" without paying for it. Mark that as a genuine agreement — number it. Where you split is whether, after the full mechanism is described, there is anything left over. The friar says a residue, presence, the grasp. Patricia says no residue, only the machinery feeling like something from inside. Hold that, because the next hour leaves the head and goes to the heart — to morality, to where caring comes from, and whether a thing with no body and no kin could ever be good. After the break.