Humanlike design of AI voice assistants helps users disregard security risks, study finds
LAWRENCE — The term “robotic voice” describes a flat and mechanical speech pattern that lacks the natural cadence of human communication.
But in this burgeoning age of artificial intelligence, that type of voice is becoming a thing of the past. Instead, humanlike qualities are increasingly considered the norm for when machines converse with people.
“With human-to-AI interaction, many parts are just like human-to-human interaction. So if the person you’re talking to is more attractive or more friendly, there is more trust in the relationship,” said Andrew Chen, Ronald G. Harper Professor at the University of Kansas School of Business.

“If an AI system has a voice that is more friendly or has an appearance that is more attractive, you also tend to trust it and forget about the security risk.”
Chen’s article, titled “Sounds Good and Feels Better — How Anthropomorphism Influences Artificial Intelligence Voice Assistant Use,” finds that anthropomorphism increases perceived enjoyment significantly, which in turn reduces both perceived security risk and the consumer’s resolve to stop using the product. The article appears in Information & Management.
“Anthropomorphism by definition is assigning humanlike qualities or characteristics to a non-human entity. So those qualities — appearance, voice or even personality ‚ are now part of AI systems. We commonly see these in things like Apple Siri or Amazon Alexa,” Chen said.
Artificial intelligence voice assistants (AIVAs) are often preferred by customers for practical reasons.
“The main benefit of AI voice assistants is simply convenience; it is always ready for you to interact. You get answers and suggestions without going onto your computer,” Chen said.
Privacy remains a concern for those hesitant to adopt the technology. However, anthropomorphism does not reduce the perceived security risk directly. Instead, the relationship operates primarily through perceived enjoyment, he said.
“It’s very difficult for the user to assess what’s behind the technology or system. That’s a disadvantage. But if you talk to people or something humanlike, you probably feel more comfortable,” Chen said.
To study this, his team culled survey data from 277 AIVA users. This was analyzed using structural equation modeling, supplemented by robustness checks and post hoc analyses. The researchers specifically looked at how AI systems affected two paths: affective and cognitive appraisal.
“Many times we use emotions or affected information as the heuristic to judge what’s ‘behind the box.’ So if the design can increase the affective appraisal, that becomes the heuristic to look at the technology behind it. If you feel comfortable or you feel attached to the AI system, you reduce the cognitive appraisal part and kind of forget about it,” he said.
Chen noted how past research typically views the two paths as parallel. But most interestingly, he said, what he found is the affective appraisal will interact with the cognitive appraisal.
“The richness of those anthropomorphisms of the AI system is not linear to the effect of the cognitive appraisal. It’s actually quadratic. So when people feel more comfortable, they reduce the risk perceptions or reduce cognitive appraisal even more,” he said.
A KU faculty member since 2006, Chen researches human-computer interaction. He is now actively working on multimodal ways to improve such AI interactions.
Chen said he often uses AIVAs in his day-to-day life.
“I have Amazon Echos in my houses. But some people are not comfortable with them. I had a friend come in, and he saw the Echo and said, ‘Can you unplug it?’ He was worried about those AI systems recording things,” Chen said.
The professor said he believes his latest research has plenty of practical implications.
“We hope designers of the AI system can look beyond function. It’s not like you should tell an Apple engineer, ‘Well, function for Siri is not important.’ It obviously still has to work. But we would like them to spend more time and effort on the hedonic part of the AI system so people feel more comfortable interacting with those systems,” Chen said.
He also said that tactic creates somewhat of a slippery slope for this pervasive technology.
“AI systems operate on IQ, or intelligence quotient, and EQ, or emotional quotient. But not many people know there’s a third one called SQ, or spiritual quotient. AI is really good at the IQ part. They’re actually smarter than human beings in many cases. But they have a long way to go on the EQ part and especially the SQ part,” Chen said. “And we probably shouldn’t train AI to get too good at either of those.”
Chen co-wrote the article with Yixiu Yu and Sisheng Lianga of Central Michigan University, Haoqiang Jiang of California State University Channel Islands and Rui Sundrup of the University of Louisville.