The role of human-like AI in effective human–machine communication

Abstract

This study explores how different evaluation metrics shape empirical findings on the communicative performance of AI agents. Using crisp-set Qualitative Comparative Analysis (csQCA) applied to a sample of 45 empirical studies, we find that human-likeness does not function as a universal driver of effectiveness. Instead, its influence appears to be context-dependent, becoming relevant in certain research settings and design configurations. Our analysis suggests that more consistent explanations of communicative success are found in combinations of contextual factors, agent typologies, and evaluative criteria, rather than in any single feature. In particular, in domains that emphasize precision and technical functionality, the absence of human-like attributes does not necessarily impede performance. These findings contribute to ongoing discussions about AI-agent design by pointing to the value of a contextual and conjunctural approach, and invite further examination of the conditions under which anthropomorphic features may or may not support effective interaction.