This paper investigates whether humans and artificial intelligence can cooperate so that their combined accuracy exceeds that of either agent alone, a phenomenon referred to as synergy or complementarity.
We introduce a formal framework that models human‑AI interaction through reliance patterns, which capture the joint correctness structure of human and algorithmic decisions.
Within this framework we derive necessary and sufficient conditions for synergy: at least one party must be able to identify and correct the errors of the other, a mechanism we call Human Correction and Algorithmic Correction.
These theoretical results lay a foundation for designing knowledge‑based decision support systems, where interaction policies must dynamically integrate heterogeneous sources of expertise.
Analyzing seven clinical case studies reveals a “dominance trap”: a strong algorithmic advantage can paradoxically suppress synergy by diminishing opportunities for human corrective intervention.
Additional experiments show that synergy can persist even in the presence of cognitive biases such as automation bias, provided that human and AI errors are functionally de‑correlated.
In summary, the work offers a formal basis for key concepts in the human‑AI systems literature and delivers actionable guidance for building interaction‑centric AI systems that foster complementarity.
Review