Can AI have taste?

Whether AI can have taste depends on what we mean by taste. If taste is the capacity to reproduce patterns derived from existing preferences, AI can approximate it. If taste is a living point of view that risks itself on choices, it cannot.

Taste as pattern

A model trained on what humans have preferred can predict what tends to be preferred. In that limited sense, it exhibits a kind of taste — statistical, derived, backward-looking. It can sort and suggest with surprising accuracy.

Taste as a living point of view

But the taste that matters in art is not pattern-matching; it is the willingness to choose against the pattern when something matters more than being liked. That taste is inseparable from a self that has stakes. A model has no stakes, and so it has no taste in this deeper sense — only a convincing imitation of one.

In essence

AI can reproduce patterns derived from taste, but it cannot possess taste as a living point of view that risks itself on choices. The first is statistical; the second requires stakes. The question is not whether AI simulates taste but whether simulation is the same as having it.

Related questions

Can AI choices be better than human ones? By some metrics, yes. But better by preference-pattern is not the same as taste that takes a risk.

Will this change? Capabilities will grow. Whether stakes can be engineered is a different, unresolved question.