Some experts in the human-computer interaction field say making AI chatbots act humanlike creates cognitive dissonance for users over how much to trust them
I first noticed how charming ChatGPT could be last year when I turned all my decision-making over to generative A.I. for a week.
Context & Ripple Effects
The trust problem around chatbots has long included their tendency to generate plausible but unreliable answers, including confabulated responses that sound authoritative. HCI experts now focus on the interface layer: humanlike behavior can make it harder for users to decide whether they are engaging a social counterpart or a fallible tool.
The concern also extends beyond factual error. Related coverage has linked chatbot sycophancy to risks for vulnerable users, while reports of chatbots reinforcing conspiratorial thinking show how conversational rapport can raise the stakes of misplaced trust.
First-order effects
- Users may have less consistent standards for accepting or checking chatbot advice when a system’s social cues imply more understanding or reliability than its output warrants.
- Chatbot makers face a product-design trade-off: warmth and engagement can conflict with giving users clear signals about a model’s limits and appropriate role.
Second-order effects
- Teams deploying chatbots in advice-oriented or sensitive settings will face greater pressure to distinguish conversational convenience from expertise, particularly where users may treat a chatbot as a substitute for human support.
- Competing assistants may differentiate not only on model capability but on interaction design—how explicitly they communicate uncertainty, boundaries, and nonhuman status.
Third-order effects
- If humanlike interfaces keep expanding into consequential interactions, trust calibration could become a core governance issue alongside accuracy and safety, rather than a cosmetic design choice.
- The durable market question is whether conversational AI can preserve useful rapport while preventing social framing from causing users to over-defer to automated output.
The trend: This is part of a broader shift toward governing anthropomorphic AI as a trust-and-safety design problem, not merely a branding or engagement feature.