A Unified Systematic Framework for Human-AI Collaboration: Technologies, Evaluation Metrics, and a Tiered Taxonomy

Abstract

The rapid development of artificial intelligence has positioned Human-AI Collaboration (HAIC) as a promising paradigm for integrating the complementary strengths of humans and AI systems. However, due to its interdisciplinary nature, existing studies often focus on isolated aspects within specific topics or applications, lacking a unified system-level abstraction. In this paper, we propose a unified HAIC framework grounded in a multi-agent system, encompassing five core technological pillars: situation assessment, social understanding, shared decision and control, human-AI interaction, and collaborative learning. Within this framework, we conduct a literature synthesis to consolidate recent advances across these dimensions. Building on these insights, we introduce a system-level evaluation methodology that integrates multi-dimensional metrics to assess HAIC effectiveness. Finally, we propose a tiered HAIC taxonomy characterizing capability evolution from basic assistance to symbiotic HAIC. Together, these contributions establish a structured, extensible foundation for designing and evaluating next-generation HAIC systems.