Who is responsible for self-AI or others-AI collaboration? The effect of power and task outcome in responsibility attribution.

Humans increasingly collaborate with Artificial Intelligence (AI), raising critical questions about responsibility attribution for task outcomes: when a collaboration succeeds or fails should the human be held responsible or AI? Prior research in non-AI settings has indicated that the responsibility attribution may differ depending on the task outcome (success vs. failure), the attribution targets (self vs. others), and on power (high vs. low). The current experimental study extends this line of inquiry to human-disembodied AI collaborative tasks (i.e., computer algorithms), investigating how these factors shape attribution patterns. The results indicate that, when collaborating with AI, participants with a low sense of power are more likely to accept responsibility for successful task outcomes relative to failures, whereas those with a high sense of power do not exhibit this self-serving bias. Furthermore, when observing others collaborate with AI, participants, regardless of their sense of power, attributed responsibility for both successes and failures to these others rather than AI. The implications for the field of human-AI interaction are discussed.