I'm wondering if there is a way to use Machine Learning algorithm to optimize games like Screeps.com, which is a game that you control your game by writing JS code. Letting the algorithm write human readable code might be too challenging, but optimizing some aspect of the game should be possible, like the best scale up route optimization using re-enforced learning.

I was actually thinking about this a couple of days ago. In regards at least as to how we would use:

A) NLP to process what is required
B) an internal system that would construct code out of established parsers

On a language like JS it would be very interesting to see since it does have certain similarities to a language like lisp. It would be more interesting to see it in lisp since code is data and data is code.

@SevenDeadlyBugs Teaching AI to code sounds very interesting. Maybe throw all the GitHub JS repository's .js file at the AI and let it learn? It might be able to train the model to write syntax correct code, but how do we let the model write meaningful code?

@AleCx04 Lisp is a very interesting case. Being able to 'update' itself, Lisp was seen as the language for AI back then. But I think if we can make AI work on one language, it shouldn't be much harder to make it work on another. Given that we have enough code base to train the AI.