Dinglir
Posts: 620
Joined: 3/10/2016 Status: offline
|
Unfortunately, I don't think you should hold your breath for a Machine Learning (ML) based AI in WitE. First, the technical side. In order to train a ML bases AI, you would need quite a lot of training material. In games such as Chess, Go and Texas Hold'em (poker) such material is readily available and all of those games have seen a ML based AI developed. So for a wargame to have a ML based AI you would need to release the game, have a lot of people play it (while collecting data) and then set the ML based AI to train. Next, one of the absolute strengths of a game such as WitE is that the rules are continously changing. For an example, the HQBU's rules have changed significantly over the last year or so. So an ML based AI trained on the old rules would make some pretty silly mistakes in a game with the new rules. Another reason why it is not so simple to just do an "intelligent" game AI. Finally, I am wondering if the strategies used in a game such as WitE are simply not to advanced for anyone to train an AI quite yet. With a success criteria of being in Berlin by 1945, the Soviet ML based AI would be swamped with options to such a degree that training it would be virtually impossible. Second, the business side. As ML development is still in it's infancy, there are simply just a lot better business opportunities to throw ML resources at before anyone would get around to games such as WitE. Next, the training would be a problem from a business perspective as well. Obviously, you can't release a "blank" AI. For the first many games, such an AI would start the game as the Axis by moving half it's units in the direction of Berlin. Not really sound from a business perspective, as noone would buy such a game. In short, the game would still need a scripted AI when released. The challenge would be to build a ML based AI on top of the scripted version in such a way that it could slowly replace the scripted version. That would then reveal the next problem. If such an ML based AI was finally trained, noone would be able to beat it. Wite is a game won by the side that makes the least mistakes. A ML based AI would slowly become better to such a point that it could never be beat (provided that training material etc is there). As for writing a generic ML based AI that could play any game, I just can't see it happening. It would be like writing a piece of computer code that could do whatever task you set it. You would still have to code in victory conditions and "AI allowed moves". But even so, the internal algorithms would be general and not necessarily suited to the specific game. I just don't see it happening. In short, I would guess it will still be a decade or so before games such as WitE will se a functioning ML based AI. However, the development of ML is going so fast that I might be surprised.
_____________________________
To be is to do -- Socrates To do is to be -- Jean-Paul Sartre Do be do be do -- Frank Sinatra
|