Training with AI: Evidence from chess computers

Abstract

We suggest that AI can help decision-makers learn; specifically, that it can help them learn strategic interactions by serving asartificialtraining partners and thus help them to overcome a bottleneck of scarce human training partners. We present evidence from chess computers, the first widespread incarnation of AI. Leveraging the staggered diffusion of chess computers, we find that they did indeed help chess players improve by serving as a substitute for scarce human training partners. We also illustrate that chess computers werenota perfect substitute, as players training with them were not exposed to and thus did not learn to exploit idiosyncratic (“human”) mistakes. We discuss implications for research on learning, on AI in management and strategy, and on competitive advantage.