It took about 50 years to develop an AI system capable of playing chess to the level of a grandmaster. Now, programmers have developed a new AI player, not one with the goal of winning chess games, but rather learning how to lose them.
Researchers from Cornell University developed the AI chess player, known as Maia, to learn more about how humans make mistakes and predict human error. The hope is that once perfected via chess, Maia will be able to predict human fallibility and become better at assisting and negotiating with humans.
One potential application for the technology is in the medical field. A medical training system that anticipates human error can train doctors and medical staff to also anticipate those common mistakes.
Maia was developed using an open-source clone of Alpha Zero, a revolutionary AI program created by the Alphabet subsidiary DeepMind. The program was modified to reinforce when it accurately predicted human behavior. This modification of AI to replicate human behavior, rather than perform at the highest possible standard, opens up a whole host of new avenues for AI applications.
In addition to medical training, the program could be used in chess itself to help players prepare for matches by going up against an AI system that perfectly replicates their future opponent’s playing style. It could also be used in negotiation practice or interview preparation.