Matt and I watched all the games with great interest, since we’re both in the field of computer science, and Matt is also a Go enthusiast.
Go is a fascinating game to see an AI do so successfully because the way that humans play Go is not very systematic. Strategies to win are unintuitive, and the consequences of any given move are not immediately obvious. Expert human Go players become good by playing lots of matches and learning to recognize various situations that arise. Through familiarity with different board configurations they learn what is best to do in each situation. They learn to develop a “feeling” for the board. This is not really the kind of thinking that computers are typically good at.
Also, unlike many games that AIs play, it cannot be solved via “brute force” – that is, by calculating every possible move it could make, and every possible response its opponent might make, etc. because the “problem space” is very large, meaning that there are too many possibilities for the AI to consider in a timely manner. This is a limitation of our current hardware. Quantum computing might change this.
The AI was trained the way an expert is trained – by playing lots and lots of games. An AI can play a huge number of games much faster than a human can, which is where computing power became an advantage here. It also doesn’t get tired and make mistakes after hours of concentration like a human does.
As for the AI uprising, I’m not worried.