Some background may be in order since I have just gotten back into chess recently. I put considerable time and effort into the game in my late 20's through early 30's. I'm not a very strong chess player but have always had an interest in the game and its history. More recently the role of computers and artificial intelligence in chess has really caught my attention. Starting with Garry Kasparov's defeat (unofficial) at the "hands" of IBM's Deep Blue in 1997, chess programs were quietly built into very powerful, if a bit, well, mechanical, chess "engines". As recently as last year, the most powerful of these engines was the open source "Stockfish" series.
Then last year a "new kid" entered the computer chess arena. A couple of matches were set up to test the abilities of a new AI (Alpha-Go) that had already proven its mettle in the ancient Asian game of "Go" when it bested Lee Sedol in head on competition for the title of World Champion. After some updating the chess version of Alpha-Go was named Alpha-Zero.
Stockfish lost. Out of 100 games Alpha-Zero won 28 and lost zero. The rest were drawn. What is most amazing about this is that Alpha-Zero had no database provided at all. Its chess programming was limited to the "rules of the game". It also is about a thousand times slower than Stockfish (60 to 90 thousand moves per second versus SF's 60 to 90 million). The new AI was trained using the reinforcement learning approach and a heuristic method known to programmers as the Monte Carlo Tree Search. That is, it played against itself and learned chess strategy basically from scratch. In 4 hours it was outplaying the best Stockfish algorithm. They gave it another 4 hours of practice (just for good measure) and then "threw them in the ring" together. Like its predecessor Alpha-Go had done in Go, Alpha-Zero has chess playing humans amazed at its creative play and the resulting games are being studied around the world. Arguably, in 8 hours, self-taught, Alpha Zero became the strongest chess player in history, human or otherwise. A common objection, when this comes up, is, "Well, someone had to program the computer to be able to out-compete the human." Fair enough, I'll come back to that in a second.
This story is far from over as both camps, Stockfish vs the Alpha series AI (and its open source clone "Leela Chess Zero" or LCZero), have improved and continue to slug it out. The thing which stands out the most to me, in all this, is that the reinforced learning approach didn't require massive databases of historical games or opening theory. The AI basically developed all of that by practicing against itself for a few hours. Let that sink in. The AI then proceeded to come up with new strategies that haven't been seen before. And it continues to improve its own play. It is this black box aspect of these new AI entities that has me both excited and terrified. Yes, I get that this is not artificial general intelligence. It's "just a game", albeit one that has challenged mankind for centuries. But the programmers don't really know what's going on under the hood.
Until recently, I felt that many old school chessies, myself included, had all but lost interest in chess since it appeared that machines now dominated it. Then it seemed that a renewed interest was sparked, particularly in very young players. Computers and AI now seem to be viewed less as competitors and more as tools for study. Very powerful tools, at that. I also think the currently reigning (human) world champion Magnus Carlson may be responsible for a lot of this renewed interest. He is doing a commendable job as a "chess ambassador".
I'm less interested in competitive play these days. I want to help spark an interest in chess in very young players. There may be an avenue for me to teach (actually, I prefer the verb "coach") chess to elementary school and high school students at local schools, possibly via GT classes.