Can a human and an AI system together be more accurate than either working alone? That question anchors a new arXiv preprint, which defines this outcome as synergy or complementarity. The abstract frames it as a formal problem: specify the conditions under which the combined accuracy exceeds both individual accuracies.
According to the abstract, the paper develops formal conditions for this kind of complementarity and then examines empirical evidence. The authors' starting point is that such synergy is "notoriously difficult to achieve," even in settings where one might expect collaboration to help.
The preprint is listed as arXiv:2610.27965v1. Because the abstract is truncated, the specific conditions and evidence are not detailed in the announcement. The key contribution, as described, is to make the vague idea of human-AI complementarity precise enough to test.