Enhancing oneself vs. Undermining opponent in human-AI competitive contexts

It remains unclear how humans and Artificial intelligence (AI) should interact socially in competitive contexts. We investigate how competitive strategy preferences shift when humans are framed as facing AI versus human opponents, focusing on the trade-off between self-enhancing and opponent-undermining behaviors. A total of 85 participants engaged in a competitive decision-making task in which they repeatedly chose between an option that increased their own points and another option that decreased their opponent's points under three probabilistic reward conditions with different expected values. Participants were randomly assigned to either an AI or human opponent group, but the point outcomes following participants' choices were generated by the same algorithm in both groups. Results revealed that while expected value significantly influenced decisions, participants exhibited a general preference for self-enhancement. Crucially, in the equal expected value condition, where the two options were matched in expected payoff, participants in the human opponent group selected the self-enhancing option significantly more often than those facing an AI opponent. Computational modeling further showed a higher learning rate in the human opponent group under this condition, indicating heightened sensitivity to feedback. These findings suggest that humans display more prosocial and less antagonistic competitive strategies when perceiving their opponent as human. The study highlights the profound impact of opponent identity on social decision-making in competitive contexts and points to a form of “AI label effect” in human–AI interaction.