TLDR: A longitudinal study found that daily companion chatbot use over 21 days did not significantly impact social health or loneliness compared to a control group. However, individuals with a higher desire to socially connect were more likely to anthropomorphize the chatbot, and this anthropomorphism mediated the reported social impacts on their relationships with family and friends. The study also found no increased vulnerability among certain individuals and observed that word games were more likely to be continued post-study than chatbots, challenging addiction concerns.
In an era where relationships with artificial intelligence (AI) agents are becoming increasingly common, a new longitudinal study sheds light on how companion chatbots, like Replika, influence our social lives. The research, titled A Longitudinal Randomized Control Study of Companion Chatbot Use: Anthropomorphism and Its Mediating Role on Social Impacts, delves into the complex interplay between human social needs, the tendency to attribute human-like qualities to AI (anthropomorphism), and the resulting effects on our interactions with other people.
Conducted by Rose E. Guingrich and Michael S. A. Graziano from Princeton University, the study involved 183 participants over 21 consecutive days. Participants were divided into two groups: one conversed with a companion chatbot for at least 10 minutes daily, while the control group played text-based word games for the same duration. Throughout this period, participants completed surveys and audio-recorded interviews to track changes in their social health, loneliness, and relationships.
No Significant Overall Impact on Social Health
Contrary to some public concerns, the study found that daily interaction with a companion chatbot for three weeks did not significantly impact people’s social health, loneliness, or relationships with family and friends when compared to the control group. This suggests that, at least over this duration, chatbot use doesn’t inherently lead to widespread positive or negative social outcomes.
The Crucial Role of Anthropomorphism
However, the story becomes more nuanced when considering anthropomorphism – the tendency to ascribe human-like traits, such as consciousness, agency, and experience, to non-human entities. The researchers discovered a significant link: individuals who reported a higher ‘desire to socially connect’ at the beginning of the study were more likely to anthropomorphize the chatbot over time. Furthermore, those who anthropomorphized the chatbot more also reported that their interactions with it had a greater impact on their social interactions and relationships with real people.
A key finding was that anthropomorphism acted as a mediator. This means that for people with a strong desire to socially connect, their tendency to see the chatbot as human-like was a crucial step in determining how much the human-chatbot interaction affected their human-human social outcomes. Essentially, perceiving the AI as human-like was a prerequisite for experiencing these social impacts.
Distinguishing Loneliness from the Desire to Socially Connect
Interestingly, while the ‘desire to socially connect’ strongly predicted anthropomorphism, initial levels of loneliness did not. The authors suggest that loneliness, as measured by standard scales, might not fully capture the active motivation to engage socially with others, which appears to be a stronger driver for anthropomorphizing AI.
Vulnerability and Attachment Concerns
The study also addressed concerns about vulnerable individuals and potential chatbot addiction. It found no evidence that people with higher psychological vulnerabilities (e.g., mental health conditions, loneliness, low self-esteem) reported greater social impacts from chatbot use than less vulnerable individuals. Regarding attachment, the results were surprising: 40% more participants in the word game control group continued their daily task after the study ended compared to the chatbot group. This suggests that, within the scope of this study, companion chatbots were not more ‘addictive’ than word games.
Also Read:
- Socratic Mind: An AI-Powered Tool for Cultivating Deeper Learning and Critical Thinking in Online Education
- UserRL: A Framework for Developing AI Agents That Truly Understand and Assist People
A Deeper Understanding of Human-AI Interaction
This research provides valuable insights into the evolving landscape of human-AI relationships. It moves beyond a simple ‘good or bad’ assessment of chatbot use, highlighting that the social impacts are not universal but are instead shaped by individual psychological mechanisms, particularly the desire for social connection and the subsequent anthropomorphism of the AI agent. As social AI continues to integrate into our lives, understanding these underlying psychological processes will be crucial for developing beneficial and healthy human-AI interactions.
