Diversifying AI Recourse Boosts Willingness but Risks Cognitive Overload

Algorithmic recourse systems provide users with counterfactual action plans to reverse unfavorable AI decisions, like loan denials or job rejections. A new study from researchers at Kyoto University and elsewhere, accepted at IJCAI-ECAI 2026, investigates the psychological trade-offs of diversifying these recourse sets. The team, led by Tomu Tominaga, ran a between-subjects controlled experiment with 750 participants, manipulating both the diversity and size of recourse sets. They measured psychological benefits (e.g., willingness to act, perceived transparency) and costs (e.g., cognitive load, negative emotions).

Results show that diversification improves benefits for small sets without incurring additional psychological costs. For large sets, however, diversification makes cognitive load more salient, reducing or even reversing the positive effects. The authors conclude that naive diversification can burden decision subjects and call for new methods that incorporate human cognition and psychology. This work underscores a critical gap in current AI recourse design: more options aren't always better—the cognitive cost of evaluating diverse counterfactuals must be managed carefully.