The potential of embodied conversational agents to stimulate adherence to eHealth interventions: steps towards a more dynamical approach

Studies on Embodied Conversational Agents (ECA’s; ‘robots on screen’) have been conducted for more than two decades since the term was formally introduced in the late nineties of the previous century. Even after all these years, ECA’s hold the promise of more productive Human-Computer Interaction, but to date their results are still mixed and inconclusive.
Within this thesis it has first been investigated whether ECA’s can contribute to improving user adherence to self-guided eHealth interventions. Secondly, main methodological issues with regards to ECA research have been addressed and an alternative approach has been proposed.
To do so, this thesis used insights from a different domain; the Dynamical Systems Perspective (DSP). DSP is a class of mathematical equations that describe time-based systems with properties such as complexity and non-linearity and can be simulated through Agent-Based Models. The principle of rapport is common to both the ECA and DSP domain and created the bridge we needed. Note that rapport is the experience of harmony between conversational partners. It is associated to the notion of ‘being in sync’.
This thesis demonstrated that the simulations run with the Agent-Based Model (ABM) were associated to fluctuations with regards to user motivation. The eHealth experiments run with the ECA showed small positive effects due to the ECA’s task-related support.
By comparing behavioral states of eHealth user and ECA and assessing their degree of mirroring across multiple modalities (e.g., facial, lexical, postural), the state of the rapport building process can be gauged on a short-term and real-time basis. Noteworthy, these measurements can be taken during the eHealth intervention. Likewise, the fluctuations in both the user’s motivational state and the user-ECA rapport state can be assessed. As an approach, it is recommended to compare the dynamical patterns that arise from these short-term data taken during the experiment with the static questionnaire data taken after the experiment.
It is expected that this combined dynamical-statical approach will give the ECA in eHealth domain a new impetus. This way, the evidence for effective ECA’s can gain strength in order to live up to their long-standing promise.