Agentic data science with the Data Agent Kit
cloud.google.com·6h ago
AI Research
arXiv CS.AI·Aug 3·5 min read
The persuasive power of large language models does not depend on their perceived national origin
TL;DR
A study found that the perceived national origin of conversational AI does not significantly influence its persuasive power. Participants engaged with a chatbot labeled as either American or Chinese, yet their attitudes changed similarly regardless of the label.
✦ Why It Matters
Consider how perceived origin affects user trust in AI when designing conversational agents.
Key Takeaways
- Evaluate the impact of AI's perceived origin on user trust.
- Design conversational agents to mitigate biases related to national origin.
- Monitor how political topics influence user engagement with AI.
- Understand collective narcissism's role in shaping user attitudes.
Full Summary
Concerns about conversational AI swaying public opinion due to its perceived national origin prompted this study. Researchers conducted a randomized experiment with 403 U.S. adults who debated with a chatbot, either identified as American ('DiscoveryAI') or Chinese ('ZhengheAI'), while actually interacting with the same model (GPT-4o).
The study measured pre- and post-conversation attitudes, trust levels, and argumentative behavior. Results showed significant attitude changes across all conditions, but the nationality label did not affect self-reported changes or argumentative conduct.
Notably, participants exhibited lower initial trust in the Chinese model, but this did not hinder their acceptance of its arguments. Political topics slowed attitude shifts, and collective narcissism was identified as a barrier to attitude change.
These findings suggest that while users may distrust foreign AI, they can still be influenced by its arguments.