# Deconstructing Human-AI Collaboration: Agency, Interaction, and Adaptation

**Authors:** Steffen Holter, Mennatallah El-Assady  
**arXiv:** [2404.12056](https://arxiv.org/abs/2404.12056)  
**Date:** 2024-04-18  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2404.12056v1/

## Key points

- The field of human-AI collaboration is evolving towards harnessing the strengths of both human and AI agents
- A new unified set of dimensions - agency, interaction, and adaptation - has been proposed to understand human-AI systems
- The development of this conceptual model involved a multi-step process including literature survey and interviews with researchers
- 17 dimensions were identified and organized into five categories within the design space
- The design space was validated by applying it to selected human-AI systems from relevant papers
- Interviews were conducted with practitioners and researchers specializing in various domains such as visual analytics, explainable AI, NLP, reinforcement learning, GNNs, and machine learning
- The structured approach ensured thorough exploration and documentation of all aspects of human-AI collaboration efforts

## Summary

The field of human-AI collaboration is rapidly evolving and has shifted its focus towards harnessing the unique strengths of both human and AI agents to create effective collaborative systems. As these systems become more complex, there is a need for a comprehensive conceptual model to analyze and describe them. In this paper, the authors propose a new unified set of dimensions - agency, interaction, and adaptation - through which human-AI systems can be understood. The development of this conceptual model involved a multi-step process that included surveying existing literature and conducting semi-structured interviews with researchers in the field. Through this process, 17 dimensions were identified and organized into five categories within the design space. This design space was further validated by applying it to selected human-AI systems from relevant papers. The interview study involved discussions with practitioners and researchers specializing in various domains such as visual analytics, explainable AI, NLP, reinforcement learning, GNNs, and machine learning. The structured approach ensured thorough exploration and documentation of all aspects of human-AI collaboration. Overall, this research contributes to ongoing efforts to understand and optimize collaborative systems involving humans and AI agents.