Embodied conversational agents for health: a comparison of empathic versus neutral dialogue

Games technology can be used to build Embodied Conversational Agents (ECAs) that tend to have a humanlike appearance and exhibit humanlike verbal and non-verbal behaviours. When it comes to digital health, ECAs can provide vital support to patients by being more reachable through their smart phones or other digital gadgets like IPad/computers. To encourage users to follow the advice given, the use of empathy and development of a working alliance is recommended. While researchers have looked at expression of empathy via non-verbal cues and certain characteristics of the patient, this study focuses on making ECAs more effective by using human-like empathy expressed during conversation through relational cues. Relational cues are the empathic utterances that build a more long term alliance among the patient and the therapist, for example, the choice of words in social dialogues or being polite during a conversation. The goal of this study is to measure the working alliance with a neutral versus an empathic ECA. Furthermore, we evaluate the impact of the relational cues on the change in the behaviours recommended by the agent. The main findings of the study establish that empathic Alex is able to change the behaviours of the users more than the neutral Alex. The likability and working alliance with the virtual coach is rated high by most of the users. Future work has been derived through the feedback received from the users revolving around having more personalised and advanced content as well as more interactivity.