Research by Michelle van Pinxsteren - Radboud University
Research by Michelle van Pinxsteren - Radboud University
Introduction
In the past decade, the deployment of conversational agents (chatbots, avatars, robots) has become increasingly prevalent. Improved capabilities to respond to users, driven by advances in artificial intelligence and, in particular, natural language processing, have enabled their use in various service contexts. Their unique ability to mimic human communicative behaviors (HCBs) can provide benefits for users and service providers, yet knowledge on how to use this ability effectively is still lacking. Although anthropomorphism theory and the CASA paradigm have provided a basic understanding of how relational variables such as trust and liking can be affected by conversational agents’ use of human-like communicative behaviors, the reality is more complex. Therefore, this dissertation aimed to investigate the effects of the use of HCBs by conversational agents on relational outcomes and how these HCBs can be implemented into service encounters considering users’ individual needs and the service context.
Findings
The results of this dissertation support the notion that a one-size-fits all approach for the implementation of HCBs in conversational agents does not exist. Both the literature review and the generative design study show that users largely agree on how conversational agents should look. Users need to be able to identify with a conversational agent and find it important that a conversational agent has a warm appearance that conforms to certain social norms. However, users’ needs concerning conversational agents’ verbal and nonverbal behaviors are more idiosyncratic. In particular, users express different preferences for the frequency, timing, and applicability of social-oriented verbal (e.g., empathy, small talk) and nonverbal behaviors (e.g., nodding, emotional expressions). These needs seem to vary as a function of the service context, the phase of the service interaction, and individual users’ needs, such as experiencing discomfort. The experiment showed that it is best to implement such behaviors in a way that they are adaptive to the user in every turn of the interaction. Finally, the experiment validated that social oriented nonverbal behaviors are particularly important when the user experiences discomfort.
Implications for Service Designers and Managers
For service designers and managers, this dissertation offers a blueprint for improving service interactions involving conversational agents. In particular, we describe users’ latent needs for both appearance and verbal and nonverbal behaviors of conversational agents. More importantly, we recommend practitioners to carefully investigate the individual needs of their users, the structure of their service interactions, and other context-related factors that could impact users’ needs. Yet, this dissertation does not only provide a blueprint for service designers and managers but is also applicable in other contexts. Due to the rapid technological developments and the COVID-19 pandemic, the use of conversational agents has found new applications. For example, avatars or virtual agents are increasingly used as ‘virtual influencers’, marketing products and services to millions of followers. In addition, conversational agents are becoming more prevalent in domains such as mental health, coaching, and motivating patients, where the ability to build relationships with users is perhaps even more important than in many ‘standard’ service contexts. Incorporating the findings of this dissertation into the design of these conversational agents can therefore be expected to bring considerable societal benefits.
Get in touch to receive the full report.