I know I’ve been asking alot ai questions today, but using logic, I will use zenith go 6 as a example, What if I put that to play games against itself, would it improve or what? This sounds like a stupid question even from my point of view since I doubt it, but I got to confirm it, In simpler terms, Would zenith go 6 improve if it played a huge amount of games against itself, for example if i go on zenith go 6 and put it to play against itself and repeat would it improve over time or is this just a stupid question and it wouldn’t?
Let's put it this way.
When I said I was new I meant ‘‘Very new’’ to the point i barely know what machine learning is. I am merely using logic as leela zero alphago played against itself to improve, the difference is a neural network but I don’t know for sure.
No, it will not improve.
OK, let’s compare it with evolution, because the principle is similar. Artificial intelligence does not learn from just playing against itself or others, but it learns by making several copies of itself, each copy being slightly different, and then determine which one is the best. Just like in nature, each individual is a little different, and the ones most suitable to ‘win’ (i.e. Reproduce) gets more chance to be selected for the gene pool of the next generation.
However, in nature this variation between individuals comes forth from randomness, while in artificial intelligence there are some tricks to steer the network’s next generation in the most efficient direction (through a process called back propagation). In that regard it is perhaps more similar to artificial selection, such as in breeding.
Since simulating games by Lizzie or Zenith does not involve any of this selection process, the programs do not learn anything.
3 Likes