Not yet human-reviewed
This source record was written automatically by the research pipeline and has not yet been reviewed by the maintainer or integrated into a topic page. Treat it as a provisional summary and verify claims against the original source.
Uncovering the dynamics of human-AI hybrid performance: A qualitative meta-analysis of empirical studies
Citation. Göndöcs, D., Horváth, S., & Dörfler, V. (2025). Uncovering the dynamics of human-AI hybrid performance: A qualitative meta-analysis of empirical studies. International Journal of Human-Computer Studies, 205, 103622. https://doi.org/10.1016/j.ijhcs.2025.103622
Summary. This qualitative meta-analysis synthesizes empirical human-AI collaboration studies published from 2018 through 2024 to characterize the factors shaping hybrid performance. It identifies a framework of AI, human, interaction, and task-related factors and maps their interdependencies rather than treating collaboration performance as an additive outcome. The authors propose that the framework and dependency graph can support context-specific design and monitoring of collaborative AI systems.
Key claims
- The review fully analyzed 206 empirical studies and identified 24 performance factors across AI, human, interaction, and task-related categories.
- Qualitative pairwise coding produced 326 directional links among the factors, indicating that hybrid-performance effects can be interconnected and non-linear rather than isolated.
- The dependency graph highlights reliance on AI, decision-making authority, and cognitive load among the influential factors and is proposed as a tool for diagnosing and improving a particular human-AI collaboration context.
Evidence & limitations
- The study uses PRISMA-guided searches across six databases, qualitative thematic analysis, and dependency graphs; its evidence is a secondary synthesis rather than a new causal experiment.
- The authors note that qualitative meta-analysis can lose original-study context, compound interpretation bias, and combine heterogeneous methods and theoretical perspectives; reviewed primary studies also often used domain-specific, small or unbalanced samples, simplified tasks, and short-term use.