How People Learn to Trust Machines

Research Area: Trust & Persuasion, RAP Lab

We tend to think of trust as something we build with other people: a coworker who follows through, a friend who tells us the truth even when it’s inconvenient. But increasingly, we’re extending that same instinct to machines: voice assistants, chatbots, robots that greet us by name. The question the RAP Lab keeps returning to is simple, and a little unsettling. Why?

The answer starts in the 1990s, with Stanford’s Media Equation research, which found that people treat computers and other media as if they were real social actors, even while knowing full well they aren’t. We don’t decide to trust a machine the way we decide to trust a spreadsheet formula. Something more automatic is happening: a set of social reflexes built for other humans, firing in response to a voice, a face, or even just a name.

That matters, because trust built this way doesn’t behave like trust in a tool. Nobody worries about “betraying” a calculator. But people report feeling genuinely let down when a chatbot gives bad advice, or oddly reassured when a robot “remembers” their name. The humanlike cues are enough to trigger real expectations, and real emotional consequences when those expectations aren’t met.

There’s also a timing problem. Trust between people usually builds slowly, through repeated small proofs of competence and honesty. Trust in machines often moves faster and less predictably. A single confident, well-phrased response can earn more trust than it’s earned, a pattern researchers call automation bias. A single visible error can erase it entirely, even in an otherwise reliable system. The machine doesn’t get the benefit of the doubt a colleague would.

This is the gap the RAP Lab studies: not whether people trust machines, but how. What triggers it, what breaks it, and how those patterns differ from person-to-person trust in ways designers need to account for. As AI systems take on more active roles (making decisions, managing tasks, working alongside us as something closer to coworkers than tools), getting this right stops being a curiosity and starts being a design requirement. Systems that invite more trust than they’ve earned put people at risk. Systems that earn trust and can’t communicate it get underused.

Understanding how trust actually forms, not how we assume it should, is the first step toward building artificial persons people can rely on appropriately, not blindly.

This piece is part of the RAP Lab’s ongoing research into trust, persuasion, and anthropomorphism in human interactions with social machines. Learn more about our research areas at raplab.ca.