The algorithms are improved, and there are more network types, and better understanding of real neurons, but I think you are mostly correct. I built a program

The algorithms are improved, and there are more network types, and better understanding of real neurons, but I think you are mostly correct. I built a program that played tic-tac-toe in -94, using a combination of traversing a problem tree while evaluating the positions using a neural network. To my understanding, this is basically the same approach used to develop a Go player, only that it took a month training a minimal network to do something useful at all at the time on my small Amiga 500...