Unlocking Creativity in AI: How Tom Zahavy’s Diversified Chess Program is Changing the Game

The Covid-19 pandemic forced many people to stay at home in early 2020, leading to a resurgence of interest in chess for computer scientist Tom Zahavy. Zahavy, who had played chess as a kid and was inspired by Garry Kasparov’s matches against IBM’s Deep Blue, delved back into the game during the lockdown. He watched chess videos, including “The Queen’s Gambit,” and focused on solving chess puzzles rather than improving his game.

Zahavy’s renewed interest in chess led him to explore the limitations of chess programs, particularly in solving complex puzzles like those created by mathematician Sir Roger Penrose. These puzzles revealed that while computers could defeat human players, they struggled with certain types of challenging problems. This sparked Zahavy’s curiosity and professional interest in exploring creative problem-solving approaches in artificial intelligence.

As a research scientist at Google DeepMind, Zahavy collaborated with colleagues to develop a new AI system that combined the strategies of up to 10 different programs, including DeepMind’s AlphaZero. This diversified system showed more creativity and skill in tackling challenging puzzles like the Penrose puzzles, highlighting the potential of teamwork among AI systems.

The team’s approach of using diverse AI systems to tackle complex problems goes beyond chess and has implications for various fields. By encouraging AI systems to consider a wide range of strategies, researchers like Zahavy and his colleagues are paving the way for more creative problem-solving in artificial intelligence.

The concept of diversity in AI systems is not only a step towards addressing the generalization problem in machine learning but also mirrors the benefits of collaboration seen in human endeavors like music composition. While there are challenges to overcome, Zahavy’s work demonstrates the potential for AI systems to think creatively by exploring a multitude of options. Ultimately, the research suggests that creativity in AI could be a matter of computational power and the ability to consider a diverse range of solutions.