An Empirical Study of a Pedagogical Agent as an Adjunct to an eHealth Self-Management Intervention: What Modalities Does It Need to Successfully Support and Motivate Users?

Prior research has shown that the more patients know about their disease, health, and lifestyle the better the health outcomes are. Patients who are suffering from either a physical disease with mental consequences or from mental illnesses can contribute to their own feeling of mental well-being by following evidence-based online, self-guided therapeutic interventions. These self-guided therapeutic interventions during which there is no contact with a care provider have shown high effectiveness. However, users (patients) of self-guided eHealth interventions have difficulties fulfilling the entire trajectory as is mirrored in high non-adherence rates. Users have reported a need for support, that is traditionally provided by human care providers. This study investigates the opportunities from within the technology to increase its support level toward the user. We deployed a pedagogical agent acting as an adjunct to a self-guided positive psychology psycho-education intervention. This agent provided instructions and user support in between and explicitly not during the online learning modules as to mitigate the risk of distraction. By setting up a between-subjects design and deploying three versions of a pedagogical agent (also known as Embodied Conversational Agent), varying the features of animation, speech, and visibility we investigated whether users felt more supported than by a fourth text-only control condition. All four conditions provided similar task-related support and emotion-related support to the user. Our results showed that our pedagogical agent made users feel guided and supported with respect to fulfilling their tasks. However, no effects were found of emotion-related support resulting in higher user motivation and an improved learning experience. Significant effects of visibility and voice were found, but animation of our pedagogical agent had no effect. On the feedback outcome variable, we found a gender effect. Male participants graded the visible Embodied Conversational Agent (ECA) higher than female participants and graded the non-visible ECA lower than female participants. In our view, ECA’s should not necessarily be deployed with the ambition to compete with the profound human potential to deliver support and guidance. Exploring ECA capabilities merits further attention, from the stance that the technology itself can support users and potentially make them adhere.