Even if you can't fully predict how a work will turn out, you still have control over your artistic processes in a way that the AI user is lacking. Even AI engineers often struggle to figure out what makes their models make the decisions that they do.

But don't forget that this is a question that exists in both philosophical and practical aspects. Philosophically "what is art" is a very nebulous thing to pin down. Practically, if AI users are allowed to copyright their output, they can use it for "plagiarism laundering" so to speak, by ripping off artists' entire collections, training AI on it, and then selling works that are clearly based on those other artists' even if non-identical. This is not something current copyright accounts for, but current copyright was made for a world with printing presses and photocopiers, not one with AI.

In most experimental work, the artist does make a direct contribution to some key elements of the work, for example framing or background. Which is all that's necessary, you can still obtain copyright over something that is only partially under your control.

If an artist gives up all direct control over an experimental work - such as the infamous monkey selfie - then I think they should no longer be able to copyright it.

After you've spun enough brushes or popped enough balloons, the results will be fairly predictable. And some elements, for example the color of paint in the brushes/balloons, would be under full control.

Even if the final result is not completely predictable, an artist only needs to establish that a significant part of it is a form of creative expression.