Amazing Coincidence in the Matches of Deep Blue and AlphaGo — Move 37, Game 2

Figure 1. Positions in Deep Blue’s game (left) and AlphaGo’s game (right).

Deep Blue vs Garry Kasparov 1997 and AlphaGo vs Lee Sedol 2016 were two critical moments in the history of artificial intelligence. Deep Blue vs Garry Kasparov 1997 was a six-game chess match, where Deep Blue won the match 3.5–2.5. AlphaGo vs Lee Sedol 2016 was a five-game Go match, where AlphaGo won the matches 4–1. The matches of Deep Blue and AlphaGo were twenty years apart. However, there was one interesting coincidence in these matches— both computers made their most amazing moves on Move 37 in Game 2. Figure 1 shows the positions in the games.

• Deep Blue (white) played 37. Be4 that seems to be a human move that surprised Garry Kasparov and many players/critics.
• AlphaGo (black) played 37. P10 is a non-human move that surprised Lee Sedol and many players/critics.

Let’s walk through some details about these two moves.

Deep Blue vs Garry Kasparov, Move 37, Game 2, 1997

Figure 2. Comments on The Chess Mind and YouTube.

Deep Blue (white) played 37. Be4. This move was controversial back then because a chess computer usually preferred a material advantage (37. Qb6 where the white queen forks the black pawn at B5 and the black bishop at D6) rather than a positional advantage (37. Be4 where the white bishop controls the center and supports its own pawns at D5 and F5.)

From the technological perspective, many artificial intelligence tasks can be regarded as optimization problems that improve/maximize/minimize the values of some objective functions. In the case of chess, objective functions could include gaining a material advantage by capturing the opponent’s pieces, achieving a positional advantage by controlling the center or aligning pawn structure, protecting the king against checkmate at all costs, and so on. In earlier days, a chess computer usually preferred a material advantage over a positional advantage simply because the former case was much easier for formulation, calculation, and programming. But, a material-over-positional strategy was more predictable in general. Deep Blue’s move 37 seemed to break such a strategy which made people wonder if Deep Blue bent the rules by itself.

Garry Kasparov and DeepMind’s CEO Demis Hassabis discussed his match with Deep Blue and his thoughts on the future of artificial intelligence in the following video. Here is an interesting comment about this video: “Back then we had the problem of engines cheating using humans. Now we have the problem of human’s cheating using engines.” All joking aside, Kasparov’s comments on human-machine collaboration are certainly a masterpiece not only for the chess community but also for the public.

Video 1. Garry Kasparov and Demis Hassabis, 2017.

AlphaGo vs Lee Sedol, Move 37, Game 2, 2016

Figure 3. Comments on DeepMind and Quora.

AlphaGo (black) played 37. P10. When this move appeared on the screen, many players/critics thought the stone had been put down in the wrong place. AphaGo’s strategy here seemed to limit the space of white on the right while developing the space of black in the center. But such a move is far beyond what a human player can master.

Figure 4. Comments on Yahoo.

As shown in Figure 4, AlphaGo achieved both goals, i.e., limiting the space of white on right and developing the space of black in the center, eventually. (Note: Figure 4 is an animation with six screens. If you see a still image, please check the pages in Yahoo.)

Video 2 shows people’s reactions to Move 37. Video 3 shows a summary of Game 2. Video 4 is AlphaGo Documentary. These videos are excellent sources for understanding the move and the overall story.

Video 2. Move 37 reactions and analysis.

Video 3. AlphaGo vs Lee Sedol, Game 2–15 Minute Summary.

Figure 4. AlphaGo Documentary

Rematch?!

There was no rematch under Deep Blue vs Kasparov and AlphaGo vs Lee Sedol.

IBM retired Deep Blue just months after its matches with Garry Kasparov. Deep Blue had never matched against any other players. No research nor progress on Deep Blue had been published since.

DeepMind also retired AlphaGo. But, DeepMind had established a great legacy around AlphaGo.

• match against European Go Champion Fan Hui — 5:0
• match against South Korean professional Go player Lee Sedol — 4:1
• match against 60 players online — 60:0
• match against Chinese professional Go player Ke Jie — 3:0
• gifts to fans around the world — self-play games and teaching tool
• papers have been published: AlphaGo, AlphaGo Zero, AlphaZero, etc.

Rematch? Nope! Thinking in this way … Deep Blue vs Kasparov 1997 was World War One. AlphaGo vs Lee Sedol 2016 was World War Two. Do you really want a rematch?!