Google's DeepMind has developed an artificial intelligence, referred as 'Student of Games', that can outsmart human players in a multitude of games, such as chess, Go, and poker. This development is seen as a critical advancement towards a more comprehensive artificial intelligence capable of executing any task with superhuman proficiency.
Google's AI excels in strategic games
Google DeepMind's artificial intelligence, known as 'Student of Games' (SoG), has demonstrated its prowess in a range of games, including chess, Go, and poker, that necessitate distinct strategic approaches. This AI is deemed a significant step towards a generalized artificial intelligence that can perform any task with superior efficiency. The lineage of SoG can be traced back to two projects, DeepStack, which was the first AI to beat professional poker players, and AlphaZero, another DeepMind's creation, which has outperformed the best human players in games like chess and Go.
Adaptable AI learns from both game types
SoG begins as a blueprint for learning games and then hones its skills through practice. Unlike DeepMind's previous AI, AlphaZero, which was adept at perfect-knowledge games, SoG can adapt to both perfect and imperfect-knowledge games, making it broadly applicable. In an imperfect-knowledge game, players are unaware of the state of all other players, such as their hands in poker, while in a perfect-knowledge game like chess, both players can see the position of all pieces at all times.
AI outperforms both human and machine competitors
SoG has been put to the test on various games, including chess, Go, Texas hold'em poker, and Scotland Yard, along with Leduc hold'em poker and a custom-made version of Scotland Yard with a different board. It managed to outperform several existing AI models as well as human players. It is believed that SoG could learn to play other games with the same proficiency, demonstrating its wide-ranging ability.
Despite SoG's impressive performance in various games, critics argue that there is still a long way to go before an AI can be considered generally intelligent. Games, they contend, are controlled settings with clear rules and behaviors, unlike the complexities of the real world. Thus, while SoG's capabilities are remarkable, its application in real-world scenarios remains to be seen.