Therapeutic Encounters: Social and Ethical Implications of AI Chatbots in Mental Healthcare

Digital technology plays an increasingly significant role in mental healthcare, with the global field of digital mental health expanding rapidly. More than one billion people worldwide experience mental health issues (WHO 2022). Accelerated by the COVID-19 pandemic, recent advances in artificial intelligence (AI) have raised hopes for improving access to mental health services while easing financial pressures on healthcare systems. In this context, chatbots – often powered by AI – have attracted particular interest. While the use of chatbots in mental healthcare can be traced back to the 1960s, recent developments in AI have significantly advanced their capabilities and broadened their application. Driven largely by start-up companies, an increasing number of mental health chatbots are now available to users.

Although academic interest in mental health chatbots is growing across various disciplines, the field is new and rapidly evolving. This presents challenges to existing research approaches and calls for inter- and transdisciplinary collaboration. To explore these complexities, the symposium brought together international researchers as well as chatbot designers and developers to discuss different aspects of what the use of chatbots in mental health entails. The particular focus was on how chatbots are currently being used to provide clients with forms of talk therapy.

Sociotechnical Imaginaries in Developing Chatbots: Assisting or Replacing Professionals?

Sebastian Merkel from Ruhr University Bochum focused on the role of 'sociotechnical imaginaries' (Jasanoff / Kim 2015) in shaping the design and development of mental health chatbots. Drawing on interviews with chatbot designers and developers, he explored how shared visions of desirable digital futures influence the ways in which technology is deployed to address the global mental health crisis. These imaginaries not only inform concrete design decisions but also shape broader expectations about the future role of chatbots in mental healthcare. While many developers see chatbots primarily as supportive tools that complement traditional therapy, others envision a more disruptive scenario in which chatbots could potentially replace human therapists altogether. The findings highlight how sociotechnical imaginaries structure both the technical and conceptual development of digital mental health technologies. Moreover, they raise important questions about how such technologies are reshaping notions of care, responsibility, and therapeutic relationships in digital contexts.

Managing Conversational Trouble: Communicative Competence of Chatbots

The third part of the workshop shifted the focus toward chatbots as communication partners, raising the question of whether these technologies are altering our fundamental understanding of communication itself. Ole Pütz from Bielefeld University presented a talk on interactions with chatbots based on large language models (LLM), examining both their technical foundations and the dynamics of their interaction with users. His presentation offered insights into the underlying technologies, including how such chatbots are trained. Beyond the technical aspects, Pütz focused on how users engage in ‘repair work’ during interactions with chatbots – a process that involves addressing misunderstandings or breakdowns in communication. He argued for a shift in perspective: rather than focusing on machine intelligence, we should pay attention to communicative competence. Drawing on empirical examples from conversation analysis, Pütz demonstrated that when chatbots manage conversational trouble, they do so not by genuinely ‘understanding’ the situation, but by responding to feedback provided by human users.

Developing a Framework for ‘Artificial Communication’

In his talk, Benjamin Marent from the University of Sussex (UK) argued that chatbots are changing the nature of what we think of as communication. He presented interview data with developers to highlight the design assumptions and challenges that guide the creation of mental health chatbots. Based on developers’ experiences of, and observations about, therapeutic encounters with chatbots, Marent expanded on Niklas Luhmann’s thesis that computer technology can become a functional equivalent of (human) consciousness and can participate in communication. Drawing on sociological systems theory, particularly the recent work of Elena Esposito, he suggested a framework for the kind of ‘artificial communication’ between humans and conversational AI applications. Marent argued that such a framework will be essential to understanding chatbot ‘behaviour’, reaping its benefits and minimising its harms.

Balancing Control and Autonomy in Digital Mental Health Tools

Stefan Lüttke from the University of Greifswald presented the second practice pitch focusing on the design and development of CADY, a chatbot created to support young adults experiencing depressive symptoms. He provided detailed insights into the motivation behind the project and its development process. The presentation echoed a recurring theme of the symposium: the challenges in mental healthcare and the potential of technology as a response. Lüttke emphasised that CADY was developed with the aim of offering accessible, low-threshold support in a situation where professional mental health resources are limited. In addition, he shared findings from several studies conducted with CADY. One key area of research focused on user-controlled personalisation. Results showed that hybrid settings – in which users could customise their chatbot experience – were more effective in reducing depressive symptoms than settings without personalisation. Users consistently preferred a combination of chatbot interaction and autonomous usage, reflecting a need for both control and autonomy in digital mental health tools. Moreover, participants expressed a desire for competent and meaningful recommendations, underlining the importance of perceived credibility and relevance in chatbot responses.

Conclusions and Outlook: AI as ‘Quick Fix’ for Systemic Failure?

Concluding observations on the symposium and the discussions were provided by Cornelius Schubert from TU Dortmund University. He outlined the importance of examining how digital mental health innovation intersects with social, economic, and clinical dynamics. Drawing on critical innovation studies, Schubert questioned the assumption that technological solutions are inherently beneficial, critiquing the reliance on ‘quick fixes’ in response to systemic healthcare deficits. The focus of tech start-ups and policymakers on closing care gaps without addressing structural causes of mental health can be scrutinised as part of neoliberal logics inherent in AI innovations. Concerns were raised about how AI reshapes therapeutic depth, clinical authority, and patient expectations, as well as the ‘invisible infrastructures’ influencing user experiences. Looking forward, the need for research on the roles of chatbots – therapy, support, or expertise – was emphasised, alongside exploring agency within human-AI interactions and the implications of digital care.

Photos: © College UA Ruhr / M. Wachtling