On May 11, 1997, Garry Kasparov, chess grandmaster, bowed to the Deep Blue computer, from the American company IBM. Rarely, in the history of technologies, has a “leap forward” been so visible and commented on.
Since then, the computer has continued its great strides and if the computer sometimes shows a little time for reflection when we face it in chess, it is only to reassure us, us humans. The blow was calculated instantly. “Chess or Othello programs are extremely strong, with a computational advantage that dominates the human being and makes the chance of beating them very unlikely”recognizes Quentin Cohen-Solal, computer scientist and specialist in learning and games.
→ REREAD. In chess, is man doomed to lose against computers?
If we do not permanently lose against the machine, it is only because we have added obstacles to it or we have deliberately designed it to be underperforming. Even go, a game considered the prerogative of humans, has seen IT prevail. In 2016, the AlphaGo program beat Korean Lee Sedol, one of the best players in the world.
Cooperative and incomplete information games
Are we exceeded on all the plateaus? “Games with incomplete information, those where not all the pieces or cards are visible, hold up well”, answers Véronique Ventos, specialist in artificial intelligence (AI) and bridge player. On a chessboard or a goban, only the opponent’s strategy is unknown. In bridge or poker, on the other hand, the opponent’s cards are hidden, in addition to his strategy.
Another area of weakness: cooperative games. By design, most AIs are black boxes, they do their job very well but can’t explain how. Difficult to play with a machine that refuses to communicate. At last, “an AI finds it very difficult to reuse its experience of a game in another similar gameexplains Quentin Cohen-Solal. A human being who knows how to play checkers will play chess more easily. This is not the case with AI, which stores a lot of experience but no knowledge. »
→ CHRONICLE. The AI that makes AI
In the future, engineers and researchers are mainly working on making programs less energy-intensive. To succeed in beating the human, AlphaGo had mobilized several hundred scientists, millions of simulated parts and an energy consumption at least 30,000 times greater than that of Lee Sedol’s brain.