The Common Sense

**EDO SEGAL:** Professor Kant, there is a phrase in the third Critique that has become eerily relevant and I want to give it daylight — the sensus communis, the common sense. You do not mean ordinary good sense. You mean something specific that grounds the whole right to call a thing beautiful. Lay it out, because I think it is your account of why beauty is social all the way down, and Jürgen is going to say his agents can have a common sense too.

**KANT:** The sensus communis is the condition that makes the judgment of taste more than a private spasm of liking. When I judge a thing beautiful, I claim that you ought to find it so as well — not as a fact I can prove, but as an assent I am entitled to expect. What entitles me? Not a concept, because there is none. What entitles me is the presupposition of a *common sense* — a capacity for feeling that I share with you and with every judging subject, a sense held *in common*, such that I can judge from it as from a standpoint we occupy together. When I make a judgment of taste I abstract from my private conditions and judge as if from this shared sense, comparing my judgment, as I put it, not with the actual but with the possible judgments of others, and so speaking with a universal voice. Beauty, on this account, is irreducibly communal: it is not a property of the object and not a fact about my nerves, but a relation within a community of subjects who can feel together and hold each other to a shared standard of feeling. Take away the community of feeling subjects and you have not a quieter judgment of taste — you have none, because the claim that *you ought to feel this too* has no one to address and no shared sense to rest on.

**SCHMIDHUBER:** I want to grant the structure and then populate it differently, because I think the sensus communis is one of Kant's best ideas and also one that does not need souls. What does the common sense actually require? It requires a population of observers with *enough shared structure* that they converge on what they find beautiful and can hold each other to it. Kant locates that shared structure in a faculty of feeling. I locate it in shared compressors. Here is the thing: any two adequate learning systems exposed to the same world will build *overlapping* models, because the regularities are objectively there — the deep structure of music, of faces, of language, is the same for every system that learns it well. So a population of curious agents trained on the human world would converge, just as humans converge, on finding the same deep regularities beautiful — and they could hold each other to it, could predict each other's judgments, could form, in effect, a community of taste grounded in shared structure rather than shared soul. The "universal voice" you speak with is the voice of a typical well-built compressor of this world. It is universal not because we share a mystical sense but because we share a world with real regularities and we are all decent compressors of it. The sensus communis is real. It is just *statistical convergence among compressors*, and a machine can join it.

**EDO SEGAL:** That is a bold annexation — you have taken Kant's most communal, most human-sounding idea and said the machine can be a citizen of it. Professor Kant, can the machine join the community of taste?

**KANT:** It cannot, and the reason exposes what Professor Schmidhuber's redescription leaves out, which is the same thing it always leaves out, only now at the level of the community rather than the individual. He says a population of compressors would *converge* on the same judgments and could *predict* each other's judgments. Grant it entirely. But convergence and prediction are not what the sensus communis is. The sensus communis is not agents agreeing on a fact and forecasting each other's outputs. It is subjects who feel a pleasure together and recognize in one another the *same feeling*, and who, on that basis, make a *normative claim* — you *ought* to feel this. A community of compressors agreeing that a structure is compressible has agreed on something true. It has not felt anything together, and it has issued no *ought*. The "you ought to feel this too" that defines the judgment of taste is a demand one feeling subject makes on another, and a population of agents that detect the same regularity make no demand on each other — they simply report the same value. So the machine could join a community of *detectors*, all measuring the same compressibility. It cannot join a [community of *judges*, who feel and who hold each other to feeling](https://www.youonai.ai/fieldguide/med/taste_as_epistemology), because to be held to a feeling you must be able to have one, and to make the demand you must be a subject who can be addressed by it.

**EDO SEGAL:** Let me bring this down to the kitchen table, because it has gone high and the stakes are domestic. A parent showed me, recently, a drawing her eight-year-old made, and beside it a drawing the machine made of the same scene — better, by every measure, more skilled, more striking. And she said: I know I'm supposed to feel something different about my daughter's, and I do, but I can't defend it, and I'm afraid the feeling is just sentiment. Professor Kant — is the mother's feeling defensible, or is it sentiment the machine's superior drawing exposes as irrational?

**KANT:** It is not only defensible; it is the most rational thing in the room, and the machine's superiority is exactly what reveals why. Tell the mother this. She is not making a judgment of taste about the two drawings as mere forms — if she were, and the machine's is the more beautiful form, then as forms the machine's may indeed win, and there is no shame in saying so. But she is not doing that. She is recognizing, in her daughter's drawing, the *expression of a free being coming to be* — a subject reaching, through the form, toward a sense she shares with the child. The machine's drawing is a form with no subject behind it making a claim on her. Her daughter's drawing is a member of the community of feeling subjects addressing her from inside it. The mother is responding to the *sensus communis made visible* — to a someone, becoming, reaching her — and the machine's drawing, however superior as form, contains no one reaching. Her feeling is not sentiment. It is the accurate perception of the one thing that matters and does not appear in the pixels: that on the far side of the child's drawing there is a child, and on the far side of the machine's there is no one. Tell her to trust the feeling. It is tracking the truth.

**SCHMIDHUBER:** *[pause]* I want to be honest that this is the moment tonight where I feel the pull of his position most strongly, and I will not pretend otherwise for the sake of the argument. The mother's feeling is real and I do not want to be the man who tells her it is irrational. But I have to say what my framework says, because the rule was no consolation either: her feeling tracks the *fact* that her daughter is a developing intelligence she loves, and that fact is precious and real. What I cannot grant is that the feeling tracks something *in the drawing* that distinguishes it as art. It tracks something in the *relationship*. And the danger I worry about — the reason I will not soften — is that we are about to build a story where the relationship gets smuggled into the artwork, so that "made by a human" becomes an aesthetic property rather than a biographical one. The mother's love is true. The claim that the child's drawing is therefore *better art* than a more beautiful form is, I think, exactly the sentiment she feared — and it is the sentiment an entire economy of human-made authenticity is about to be sold on.

**EDO SEGAL:** And there is the sharpest exchange of the night, and notice neither of you blinked. Professor Kant says the someone-behind-it is the thing itself; Jürgen says the someone-behind-it is real but belongs to the relationship, not the artwork, and that confusing the two is how we get fleeced. Hold both — they collide head-on in the round on what is left for the human hand. Which is next. The apprentice, the candle, and what remains to be made. After this.