Google DeepMind Creates AI-Powered 'Student of Games' to Beat Human Players

Several AI tools have been developed over the years, some outpacing humans in various fields. But what if we have an AI that can beat humans in various games? This is precisely what Google's DeepMind has in store for us.

It collaborated with various organizations to create an AI system called Student of Games (SoG) capable of mastering and beating humans at various games, Tech Xplore reported.

Unlike traditional game-playing AI, SoG isn't limited to one game. It excels in chess, Go, poker, and more, as it can teach itself how to play virtually any game.

Student of Games

The SoG system showcases remarkable proficiency in defeating human opponents across various games and has the unique ability to learn and adapt to new games.

Machine learning and artificial intelligence (AI) have evolved significantly over the past 50 years, leveraging human-generated data to train computer systems. One notable application of this technology is in board and parlor games.

Teaching a computer to master a board game and enhancing its capabilities to outperform humans marks a significant stride in AI. The researchers noted that computer systems designed for board games typically specialize in a single game, like chess, creating artificial intelligence tailored for specific tasks.

However, the creators of SoG have developed a versatile, intelligent system that can play a range of games requiring different skill sets. Games can be broadly categorized into those with perfect knowledge, where both players have complete information, and those with imperfect knowledge, like poker, where players only have partial details.

According to the researchers, SoG demonstrates excellence in both types of games, displaying proficiency across various gaming scenarios and substantiating its capability to outperform expert human players.

SoG Beats Human Players

The AI system has successfully defeated its opponents in Go, chess, Scotland Yard, and Texas Hold 'em poker. What sets the Student of Games apart is its ability to teach itself to play virtually any game, hinting at its potential to excel in a wide range of gaming scenarios.

The study's abstract introduces SoG as a general-purpose algorithm that amalgamates previous AI approaches, seamlessly combining guided search, self-play learning, and game-theoretic reasoning.

"Student of Games achieves strong empirical performance in large perfect and imperfect information games - an important step toward truly general algorithms for arbitrary environments," the authors wrote.

The study not only highlights the achievements of the SoG in chess, Go, and poker but also establishes its superiority over existing agents in imperfect information games like Scotland Yard.

Collaborators of Google DeepMind include EquiLibre Technologies, Sony AI, Amii, and Midjourney. The findings of the AI research team were published in the journal Science Advances.