The point I’m trying to make is that this evolving code part is capable of creativity, or at least a very good imitation of it.

That’s the main thing distinguishing old chess/go bots from the new generation, which has become way, way stronger. The old bots essentially just did depth searches and then evaluated positions based on spoon fed human knowledge. This was a big hurdle for Go bots in particular, because depth searches are extremely computationally difficult with a board that large.

The new generation instead, plays millions of games against itself. It randomly changes its strategies over time. If it wins, it tells itself, “hey I won! Maybe that is worth remembering”, slightly changes it’s code accordingly and continues building from there.

These type of bots are capable of coming up with completely new strategies on their own. Again - not just through search trees, that’s completely infeasible for a game like go - but by modifying its own code incrementally until it knows how to play the game. And similar things can happen here, even to a lesser degree. Go/chess have the advantage of having very clear outlines of what “good” is - if you win the game, good, have your cookies continue just like that, sport. For essays etc… it’s a bit more vague - the best we have is user feedback, and you need some separate intelligent code to generate “feedback” on its own. But in this manner, it does something that is, imo, akin to “creativity”