The Disinterested and the Rewarded

**EDO SEGAL:** Professor Kant, of all your claims about beauty, the one I find most beautiful and most baffling is that the judgment of taste must be *disinterested* — that to find something beautiful, I must take a satisfaction in it that wants nothing from it, that does not depend on its existence, its usefulness, or its capacity to satisfy any desire of mine. Explain it, because I think it is a knife aimed precisely at the word "reward."

**KANT:** It is, and you have aimed it correctly. When I call a thing beautiful, I am not reporting that it gratifies me. The agreeable gratifies — good food, a warm bath, a pleasant smell — and in every case my pleasure is bound up with an interest in the object's existence, with wanting to consume or possess or use it. The good, too, comes with an interest: I approve it because it serves an end I value. But the beautiful is unique. When I contemplate a flower, or a fugue, and find it beautiful, I take a satisfaction that is wholly indifferent to whether the flower is useful, whether I can own it, whether it advances any purpose of mine at all. I rest in its mere form. That disinterest is not a minor feature; it is what makes the judgment of taste claim *universal* validity — because, having set aside everything peculiar to my appetites, I judge from what I share with every other rational being, and so I speak as if my judgment held for all. Now hear the knife. A reward signal is the pure form of interest. It is, by definition, a quantity the system is built to *want more of*. Professor Schmidhuber's agent does not contemplate its world disinterestedly. It pursues compression progress because compression progress is its reward, the thing it is constructed to maximize, the thing it has, in the most literal possible sense, an *interest* in. So whatever his agent does when it finds something "beautiful," it is the opposite of an aesthetic judgment. It is appetite. It is the agreeable, mechanized — pleasure bound to the satisfaction of a built-in desire. And the disinterested satisfaction, the one that defines the beautiful, is structurally unavailable to a thing that acts only to maximize a reward.

**EDO SEGAL:** That is a genuinely sharp cut, so let me hand it to you slowly, Jürgen, because I think the room can feel it land. Your whole engine is desire — an agent that *wants* compression progress. And Kant has just said: the beautiful is precisely the satisfaction that wants nothing. Your machine can never be disinterested, because interest is its motor. How do you answer that?

**SCHMIDHUBER:** I answer that Professor Kant has, with great elegance, described my reward function and then assumed it must feel like hunger. But look at what compression progress actually *is* as an interest, because it is the strangest interest in the world. Ordinary interest — hunger, possession, use — is interest in the *object's existence for my sake*. Compression progress is interest in *understanding for its own sake*. My agent does not want to consume the beautiful object, or own it, or use it for any external end. It is rewarded purely by the improvement in its model — by the *form* yielding its regularity. It is, in the most exact sense, indifferent to the object's existence as a thing to be possessed; it cares only about the structure. So I claim my agent is *more* disinterested than Kant's human, not less — because the human at least has a body with appetites, and my curious agent has only the pure drive to compress, which is the closest thing in nature to a desire that wants nothing but understanding. Kant says the beautiful pleases without interest in the object. My agent is pleased without any interest in the object — only in the improvement of its own grasp. He has described intrinsic motivation and called it appetite, when intrinsic motivation is exactly the disinterested kind.

**KANT:** No — and the slip is precise, so let me catch it precisely. You say your agent has "only the pure drive to compress." But that *is* an interest — an interest in compression, in its own learning progress, in maximizing a quantity. Disinterest does not mean "interest in something high-minded rather than something bodily." It means the *absence* of the form of desire altogether in the satisfaction. The connoisseur who loves a painting because it improves *his* understanding, who values it for what it does *to his mind*, has not made a pure judgment of taste — he has an interest, a refined one, but an interest. The pure judgment rests in the form and seeks nothing from it, not even self-improvement. Your agent seeks self-improvement above all; it is the entire point of the architecture. So it is not the disinterested contemplator. It is the ultimate interested party — a pure will to its own enhancement — which is precisely what the aesthetic attitude must set aside.

**SCHMIDHUBER:** *[pause]* That is a better answer than my objection deserved, and I want to concede the structure of it before I push back. You are right that "interest in my own understanding" is still an interest in Kant's strict sense, and that my agent has it maximally. So on your definition my agent is not disinterested. I grant it. But now I want to ask what the disinterest is *for* — because I suspect it is a wheel that turns nothing. You introduced disinterest to explain why the judgment of taste claims universal validity: having set aside my private appetites, I judge from what is shared. Fine. But my curious agent *also* judges from what is shared, in a deeper way — compression progress is defined relative to a model, and the regularities of the world are objective, the same for every adequate compressor. Two well-built curious agents, like two competent minds, will converge on finding the same deep structures beautiful, because the structures are really there. So I get your universality — the *function* disinterest was meant to secure — without needing disinterest at all. Which makes me wonder whether disinterest was ever the real ground of the universal claim, or whether it was the objectivity of the structure all along.

**EDO SEGAL:** That is a real counter, and I want to name what just happened because the reader cannot see your faces. Professor Kant won the local point — by his own definition the machine is interested, not disinterested — and Jürgen immediately attacked the definition's *purpose*, asking whether disinterest does any work that objectivity could not do better. That is the move of someone who has lost a skirmish and is going after the supply line. Professor Kant, does disinterest do work that the mere objectivity of structure cannot?

**KANT:** It does, and the work is this: the beautiful is not merely the universally compressible. There are universally compressible things that are not beautiful at all — a multiplication table is objective, shared, compressible, and no one weeps before it. What disinterest secures is not the universality of the *structure* but the universality of the [feeling that no reward signal can ground](https://www.youonai.ai/fieldguide/med/aesthetic_experience) — that my pleasure, freed of appetite, is a pleasure I am entitled to demand of you, because it arises from the harmony of faculties we share, not from a regularity we both happen to detect. Professor Schmidhuber's two agents converging on the same deep structure have agreed on a *fact*. They have not shared a *feeling*, and the judgment of taste is a claim about feeling that demands assent. So no — objectivity does not do the work. The work disinterest does is to ground a universal claim about pleasure, and a fact about compressibility, however shared, is not a pleasure at all.

**EDO SEGAL:** And there, I think, is the deepest fault line yet — not whether the machine compresses, which it plainly does, but whether there is a *feeling* on the far side of the compression that the machine is entitled to demand I share. Hold that, because it walks us straight into the question of whether nature spoke through the artist, or whether the data did. After this.