Cross-Cultural Evaluation of Bias–mitigating Social Robots in Managing Robot Moderated Human–human Interactions

Summary

This project sought to explore the design and feasibility of a bias-mitigating robot for fair human-human interaction moderation and to strengthen a collaboration between Swansea and Kyoto Universities, initiated with support from the Japan Society for the Promotion of Science (JSPS). At the beginning of this project, we set out three objectives that extended our prior work on bias mitigating robots: (1) finalise the design of an LLM-powered fairness seeking moderator robot (2) transfer of expert domain knowledge in ROS from Japanese partners to UK project team and (3) evaluation of fairness seeking robot on
improving fairness in human-human interactions. To date, we have developed a novel autonomous, multi-agent robot that detects and mitigates dominance bias in real–time human–human interactions within a Japanese cultural context. We are seeking funding to finalise our empirical evaluation and replicate the study in the UK. It will help us explore how cultural norms influence perceptions of fairness and dominance, and evaluate the ethical acceptability of using bias-mitigating robots in various contexts.

Outputs include:​

• Cross-cultural datasets ​
• Submission to ACM HRI​, 2027
• Policy brief

Collaborating with Kyoto University.

Activities and Achievements

During visits to our collaborators in Japan, we have successfully completed the design and knowledge transfer of a fairness seeking robotic system that autonomously monitors human-human interactions for signs of bias and then acts to try to mitigate them. We also conducted an initial pilot of the experimental methodology in English. In consultation with our partners – experts in designing non-verbal social robotic behaviours – we also spent considerable time reflecting on and refining the bias mitigation intervention sequence our robot performs when delivering fairness seeking interventions: taking inspiration from micro-aggression and inter-group dialogue (IGD) literature, our robot will now follow an approach to bias mitigation grounded in interdisciplinary research, merging computer and behavioural sciences techniques to detect and act on bias within human-human interactions.

Work continues on analysing Japanese language transcripts and replicating the work in the UK to enable examination of cross-cultural influences on the presence and identification of bias in free-form interactions.

Impact

Our approach to bias identification and mitigation may be applied to professional settings, to mitigate the risk in real-time of negative affects influencing decision-making performance and decision-maker dignity, in human-human interaction scenarios.