“Social Connections with AI Agents – Investigating Effects on Loneliness and Health” A Workshop hosted by Elizabeth Broadbent
On the 19th of October, NOMIS Fellow Prof. Elizabeth Broadbent hosted a closed workshop at the Collegium Helveticum in Zurich. The workshop asked how people can maximise the benefits of social relations with agents, while avoiding the harms that could be associated with them. Nadja Rupprechter, a recent member of the sHAI, was invited to speak about relationships with text-based, conversational AI agents, and presented her research among 25 domain experts.

Drawing on her research on human–chatbot relationships and parasocial phenomena, Nadja discussed how people form and experience relationships with conversational AI, how they compare to interpersonal and traditional parasocial relationships, and what these relationships mean for loneliness and wellbeing.
The program was full – presentations ranged from research in HRI to HCI but also included perspectives of filmmakers and designers. The event was kicked off by an input from Elizabeth Broadbent herself, followed by presentations from session chairs Emily Cross, Beate Ditzen, and Urte Scholz.
After a full day of talks, the group came together to decide on “Where We are going from here” – summarizing the most important next steps for research on human-agent relations, as well as the roles of responsibility from scholars, policy makers, and the platforms themselves. One conclusion the group shared: AI agents should be designed to support and foster human connection, not to replace it.

The event concluded with a live session and invigorating commentary from Prof. Ben Shneiderman on our conclusions. He stressed the dangers of anthropomorphic agents and systems, which, according to him, enthrall users to inevitably build parasocial relationships with them over time – relationships that are not necessary for successful task fulfillment. He concluded by proposing a future where AI does not refer to itself using “I” statements.
His position marks one end of an ongoing debate: is human-like design itself potentially problematic, or does the answer lie in how different people perceive and use agents, regardless of their design? The workshop made clear that this question needs answers from research, design, and policy alike.

My own position sits between trying to foreclose bond formation at any cost and the full embrace of human-agent relations as symmetrical bonds. Human-agent relations extend, and reveal, how deeply we are built to socialize. For some users, these relationships bring clear benefits, from feeling better day to day to having a safe space to express themselves – and those experiences are real and to be taken seriously. Others may displace precious time, for finite biological beings like us humans, onto an agent that, ontologically speaking, cannot invest back – and lose out on social opportunities in the “real” world. And it appears that design matters here: in my own research, human-like design was strongly associated with users’ parasocial attachment. So, Shneiderman is right that agents’ anthropomorphism draws people in. Whether that pull helps or harms, though, depends on who is being drawn in, and why. Answering that requires ongoing research that follows users over time and also looking into what people actually do when they engage with these agents.
Many thanks to Elizabeth Broadbent and Urte Scholz for the invitation and for bringing together such an interdisciplinary group, and to the Collegium Helveticum and the NOMIS Foundation for hosting and supporting the event.
For more information on the program and the speakers, please see the website below.





Comments