Hybrid Systems of Humans and AI Agents

Human societies are characterized by complex networks of social connections. These formal or informal relationships - which may range from friendly to hostile, from altruistic to competitive, from salubrious to insalubrious - may advance or compromise both individual and public welfare. To this web of connections, we are increasingly adding machines. Humans increasingly interact socially with software agents (e.g., in phone calls to service providers such as airlines, or with online bots) or physical machines (e.g., autonomous trucks that share the road with us). An important scientific challenge is thus to understand how such machine technology might be deployed - deliberately or incidentally - to promote beneficial outcomes at the individual and collective level. The impact of artificial intelligence (AI) agents will be felt in social dilemmas as disparate as cooperation, coordination, evacuation, navigation, sharing, or information flow (as with the spread of fake news or hatred online). In this line of current work, we investigate "hybrid systems" of humans and AI agents ("bots"). We explore how bots placed within online (and offline) social groups, can affect individual behavior and group performance. Can simply programmed agents, when added to social systems, help people to overcome collective action problems of diverse sorts, and avoid social traps? In short, here, we are interested not so much in creating super-smart AI to replace human cognition, but rather in creating what we call "dumb AI" to supplement human interaction. We create and test new models of such collective action problems, and invent new sorts of bots that might help overcome them.