https://deepmind.com/blog/alphazero-shedding-new-light-grand-games-chess-shogi-and-go/
If you visit chess24.com, you can watch GM Matthew Sadler’s YouTube commentary on selected AlphaZero games. Amazing stuff!
The analogy of Hubert Dreyfus’ old books to “God of the Gaps” seems striking.
remarkable what Alphazero is doing. Hopefully it can eventually be applied to medicine,agriculture,the economy etal
The few games made available show Stockfish playing a very constrained game and odd moves that a very low rated player might make randomly. Computer vs. computer games are often mismatches of computing power, time limitations, and heuristics. More convincing would be games at standard tournament time controls against players rated above 2750 or a test of the program against a series of 2200’s, 2300’s, 2400’s, 2500’s, and 2600’s to compare differences in pattern recognition, tactical accuracy, strategic differences, and endgame technique. Human players often play in unique ways that can be very challenging (frustrating?) for computer programs. There may be stylistic differences that AlphaZero would find more difficulty coping with. Its AI abilities may for now still crash the car or send an astronaut spinning off into space.
I look forward to AI solving practical problems after they get finished trying to figure out human decision-making processes within a limited realm of possibilities. Will AI understand empathy, love, hate, jealousy, persistence in a hopeless position, kindness, or optimism? Will AI eventually understand why we do foolish things like play chess, go, or shogi rather spend our time on productive activities that generate tangible benefits mostly for a small group in an economy? Passions drive our reason as well as our ability to create. What drives AlphaZero besides humans trying to make money from its algorithms? Pray that AI never becomes sentient. The results may not go well for us in a troubled world.
Not sure what games you saw. On youtube Danny King analyses a Dutch Leningrad that Alpha used to beat Stockfish that is incredible.Likewise he showed a Caro kann where Alpha makes a weird pawn sac right out of the opening and eventualy grounds Stockfish down.If you haven’t seen them check them out
The rules of chess, shogi, and go can all be defined exactly, so transferring this knowledge to the real world will be hard. Hassabis sees protein folding as the big short-term real-world application. Some interesting links: