AI Agents in Science: A Call to Study Human-Agent Systems
A recent paper published on arXiv (2608.14667) posits that AI agents integrated into scientific teams should be analyzed as human-agent systems (HAS), with the focus being on the interaction between humans and agents. The authors argue that existing studies emphasize the independent functions of 'AI Scientists,' neglecting the collaborative nature of scientific work. Their literature review and empirical findings reveal potential dangers of utilizing agents without considering human-agent interactions, such as a decline in the diversity of scientific exploration. Case studies illustrate how scientists and AI can enhance each other's abilities. The authors advocate for further research using the HAS perspective to create mathematical models aimed at improving human-AI collaboration in scientific endeavors.
Key facts
- Paper arXiv:2608.14667 is titled 'Position: AI Agents in Scientific Teams Should Be Studied as Human-Agent Systems'.
- The paper argues that studying AI Scientists as human-agent systems is underexplored and undervalued.
- Current work focuses on autonomous capabilities of AI Scientists, overlooking social aspects of scientific teamwork.
- Deploying agents without accounting for human-agent dynamics introduces near-term risks, including reduced diversity of scientific inquiry.
- Real-world case studies show that scientists and agents can augment each other's capabilities.
- The authors call for new research using the HAS lens to develop mathematical frameworks for human-AI synergy.
- The paper is based on literature and empirical analysis.
- The source is arXiv, a preprint server for scientific papers.
Entities
Institutions
- arXiv