UKRI AI CDT in Dependable and Deployable AI for Robotics (CDT-D2AIR)

Stuart Gow

Project
How to Build and Evaluate Empathic but Psychologically Safe Social Robots

Summary
Given the importance but also the significant risks associated with this area, I am investigating how we can build and evaluate empathic social robots. The challenges are perhaps twofold, firstly what are the most effective ways to develop computational models that can properly use multimodal perceptions (speech, body language etc) to evaluate and model the affective states of  a person engaging with a social robot and then for the robot to respond in a human-like and empathic manner.  Secondly, how can potential solutions be evaluated to ensure they are performant but also that they are psychologically safe and do not cause harms such as over dependency or inappropriate influence. There are many examples of existing conversational agents causing psychological harm and an empathic, embodied robot increases the risk and impact of such harms.