I definitely support.
Originally Posted by
Tyspolir
--- At this point it feels like CA already accepted they can never get the AI right (which to be fair, is extremely difficult, especially to hardcore fans who have very high expectations).
I think it depends. I am no expert, not even close. But I have some basic experience in programming myself and some knowledge of the subject, so I will give my speculation regardless of you want it or not
The simplest learning algorithms are easy to program, and some of them WILL reach perfection, eventually. Especially those that imitate the evolutionary process: https://en.wikipedia.org/wiki/Evolutionary_algorithm
The thing is, I have a strong feeling the time required would increase exponentially as the game would become more complex, and Total Wars are, in the end, very complex games. Perhaps some learning algorithms can be efficient and relatively fast, but I don't know enough of them to give any definite examples.
I think in practise, learning algorithms can be used to improve AI in TW, but they would probably be requiring a simplified version of the game in order to produce usable results in a proper time. At least a foolish AI that ignores some features can be an improvement over plain stupid AI that does nothing properly without cheating. And the process would need to be cut at some point anyways, because playing against a god-tier foe is not fun for most people.
Here I mean that the AI would not necessarily need to be learning while in the end product, but the learning algorithm is used as a tool for creating the AI while in development.