News Release

AI can do the task. Can it keep humans in charge?

A National Science Review article proposes a unified framework for human-AI collaboration in which a human-factor agent is a key component. The framework aims to preserve human agency, improve mental alignment, and support human well-being

Peer-Reviewed Publication

Science China Press

The unified HAIC framework and its five technological pillars.

image: 

The unified HAIC framework and its five technological pillars. The task agent interacts with the task environment, while the human-factor agent models and interacts with human users. Collaborative learning supports the joint adaptation of both agents.

view more 

Credit: ©Science China Press

A review accepted by National Science Review proposes a unified framework for designing and evaluating how human and AI agents work together. The authors bring together research from several fields to address a practical question that arises wherever people use AI, from healthcare to transportation and manufacturing. What should an AI system be able to do as a partner, and how can researchers tell whether the partnership is working?

The framework distinguishes two roles within the AI system. A task agent handles the task and responds to changes in the environment. A human-factor agent focuses on the human user, interpreting their intentions and supporting communication. The review organizes research around five core technologies that help these roles work together: situation assessment, social understanding, shared decision and control, human-AI interaction, and collaborative learning.

The authors propose evaluating collaboration across several dimensions. Alongside task performance, their approach considers mental alignment, or how closely the person and the AI share an understanding of the task goal and the next step, as well as user engagement and trust. For example, an AI system may help someone complete a task faster. A fuller evaluation would also ask whether the person understands what the system is doing and can step in when needed.

The review describes levels of collaboration ranging from basic AI assistance to human-AI symbiosis, where people and AI continually coordinate and adapt to each other. These levels help researchers identify the capabilities an AI system needs for a given application. For example, a voice assistant that interprets a user’s intent illustrates Level 2 (intent-responsive collaboration). A factory robot that coordinates assembly tasks with workers illustrates Level 3 (shared decision and control collaboration). At Level 4 (proactive interaction collaboration), a home care robot can anticipate when someone may need help and offer it before being asked.

The authors are affiliated with the Institute of Automation and the Institute of Software, both part of the Chinese Academy of Sciences, as well as the University of Science and Technology Beijing, Northwestern Polytechnical University, Xi’an Jiaotong University and Zhejiang University.


Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.