Social Preferences Towards Humans And Machines: A Systematic Experiment on the Role of Machine Payoffs

Abstract

There is growing interest in the field of Cooperative AI, that is, settings where humans and machines cooperate. By now, more than 160 studies from various disciplines have studied how people cooperate with machines in behavioral experiments (see review by March, 2021). Our systematic review of the experimental instructions reveals that the implementation of the machine payoffs and the information participants receive about them differ drastically across these studies. In an online experiment (N = 1198), we compare how these different payoff implementations shape people's revealed social preferences towards machines. When matched with machine partners, people reveal substantially stronger social preferences and reciprocity when they know that a human beneficiary receives the machine payoffs than when they know that no such “human behind the machine” exists. When not informing participants about machine payoffs, we measure weak social preferences towards machines. Comparing survey answers with those from a follow-up study (N = 150), we conclude that people form their beliefs about machine payoffs in a self-serving way. Thus, our results suggest that the extent to which humans cooperate with machines depends on the implementation and information about the machine's earnings.

Keywords: Machine behavior, cooperative AI, human-computer interaction, social preferences

JEL Classification: A12, C13, C18, D03, O33