[__MatrixMan__]

Since we enjoy art, we have a lot of it around. That means that there's a wealth of training data available. If we're lucky, the techniques that we develop teaching AI to mimic what we like to do will also apply to teaching AI to do things that we don't like to do, or things that we can't do.

That may be a stretch, but it's the most charitable explanation for OpenAI's behavior that I can come up with. Art as practice for something else.

Perhaps later we'll be generating things like fungal species which leech lithium from the depths and bring it to the surface for extraction. That would be truly useful. But there's not very much training data for such a thing, so we're not focusing on it until we get our techniques sorted out in other domains.

Also, you'd have to be a double-specialist to work on such a thing. Both AI and Micology. Those are hard to find and take a long time to train. Much easier to find is the double specialist in AI and things-that-look-cool.