Strategic interactions between humans and artificial intelligence: Lessons from experiments with computer players

Artificial intelligence (AI) is starting to pervade the economic and social life. This renders strategic interactions between humans and artificial agents more and more common. At the same time, experimentalists have increasingly employed computer players as a tool to provide a deeper understanding of strategic interaction in general. What can this strand of research teach us about the strategic behavior of humans interacting with AI? I review 162 experimental studies using computer players. I find that behavior and the outcomes of strategic interaction often change when humans interact with computer players rather than other humans. In particular, humans usually adapt to computer players even absent detailed prior information about them, they often act more selfishly and more rationally in their presence, and they apply different cognitive processes. Moreover, while humans are able to exploit computer players following simple algorithms, they are also vulnerable to exploitation by more sophisticated ones. Conversely, computer players may also enhance efficiency. I suggest several directions for future research.