Complementarity in Human-AI Collaboration: Concept, Sources, and Evidence

TL;DR

This work provides researchers with a comprehensive theoretical foundation of human-AI complementarity in decision-making and demonstrates that leveraging these sources constitutes a viable pathway towards designing effective human-AI collaboration, i.e., the realization of CTP.

Abstract

Artificial intelligence (AI) has the potential to significantly enhance human performance across various domains. Ideally, collaboration between humans and AI should result in complementary team performance (CTP) -- a level of performance that neither of them can attain individually. So far, however, CTP has rarely been observed, suggesting an insufficient understanding of the principle and the application of complementarity. Therefore, we develop a general concept of complementarity and formalize its theoretical potential as well as the actual realized effect in decision-making situations. Moreover, we identify information and capability asymmetry as the two key sources of complementarity. Finally, we illustrate the impact of each source on complementarity potential and effect in two empirical studies. Our work provides researchers with a comprehensive theoretical foundation of human-AI complementarity in decision-making and demonstrates that leveraging these sources constitutes a viable pathway towards designing effective human-AI collaboration, i.e., the realization of CTP.

Figures (12)

• Figure 1: Asymmetries in decision-making between human and AI as sources of complementarity.
• Figure 2: Illustration of the principle of human-AI complementarity based on different (in)correct decisions that human and AI can make.
• Figure 3: Complementarity potential ($CP$) split into inherent ($CP^{inh}$) and collaborative ($CP^{coll}$) components for a single instance with better human performance (left) and better AI performance (right) in respect of $T^\ast=AI$. $l_D$ denotes the instance-specific loss with $D\in\ \left\{H,AI\right\}$ with a lower loss indicating better performance for the same instance.
• Figure 4: Illustration of (theoretical) complementarity potential and (realized) complementarity effect for a hypothetical situation extending the introductory example in \ref{['fig:illustration_human_AI_complementarity']}.
• Figure 5: Illustration of (theoretical) complementarity potential ($CP$) and (realized) complimentarity effect ($CE$) in respect of different loss scenarios for a single instance---assuming, without loss of generality, that the AI performs better overall ($T^\ast=AI$). $l_D$ denotes the instance-specific loss with $D\in\ \left\{H,AI,I\right\}$---with a lower loss indicating better performance for the same instance.