fizkes/Shutterstock
Human faces provide the most information-dense stimuli that people encounter in daily life. It takes just around 100 milliseconds—less than half the blink of an eye—to decide whether to trust the person in front of you.
But that snap judgment may be wrong for two reasons, according to two studies conducted in the laboratory of Chicago Booth’s Alexander Todorov. First, it may be swayed by social stereotypes that develop and get passed down through multiple generations, regardless of whether they’re accurate. Second, it may be misled by a smile.
Snap judgments about whether to trust someone probably evolved for quick threat assessments, and it’s important to understand how they work, write Todorov and the researchers in the first study led by Booth postdoctoral scholar Stefan Uddenberg. Appearing trustworthy helps people get elected, get loans, and get more lenient criminal sentences. And the social stereotypes that influence a person’s perceived trustworthiness increasingly inform artificial intelligence–driven facial-recognition systems, they say.
To probe the development of these stereotypes, they took an approach similar to the game of “telephone.” They showed participants computer-generated faces designed to vary along a preestablished continuum of perceived trustworthiness. Paired with each face was an amount of money, unknown to the participants, that the fictitious character had supposedly shared with a partner in a trust game. A larger amount would signal a higher level of actual trustworthiness.
Researchers will be able to generate synthetic but photorealistic faces that can be tuned along sets of perceived attributes.
Looking for a Trustworthy Face? There’s a Photo Database for ThatThe participants were asked to guess the percentage of the money that each character had shared. In a learning phase, each guess was followed by a display of the amount that the researchers had attached to that face. After that, the subjects guessed the percentages without receiving any feedback. Those values then informed the learning phase for a second generation of participants, and so forth. After multiple rounds, the researchers find, judgments of trustworthiness converged around simple positive assessments of facial characteristics, regardless of the initial level of trustworthiness assigned to each face. Even if the first participants learned that a certain happy, attractive character had shared less than they had expected, participants over time nevertheless indicated that they trusted the character to share. This shows that “participants held a strong bias,” the researchers write.
A key element to the stereotypes underlying trustworthiness is attractiveness, known for a century to scientists as the “halo effect.” Todorov worked with another group of researchers to demonstrate that in the absence of the halo effect, faces that seem happier or more approachable also appear trustworthy.
This team ran three experiments on what makes a face seem trustworthy if attractiveness is removed from the equation. They used the same system as the first team for generating artificial faces but tweaked the “trustworthy” faces to make them less attractive—and didn’t alter them to make them smile or frown. Then they showed the faces to dozens of online participants.
Faces seen as happier and more approachable rated as more trustworthy even if they weren’t deemed attractive, they find. When the researchers carried out the same experiments using machine learning in place of human participants, the computers and the humans concurred: faces that seemed happier and more approachable were perceived as more trustworthy.
“The findings clearly show that emotional expressive behavior is an important cue for perceived trustworthiness,” the researchers write. Thus stereotypes and a smile can lead you to judge someone trustworthy, even when the person isn’t.
Researchers put a twist on a classic experimental design.
How Some Experiments Use Emails to Control for Systemic Bias
The answer affects financial planners, caregivers, and many other surrogate decision makers.
Why We’re So Bad at Making Choices for Others
Knowing a better iPhone is coming may ruin your experience of the one you have.
Do You Really Need to Upgrade Your Phone?Your Privacy
We want to demonstrate our commitment to your privacy. Please review Chicago Booth's privacy notice, which provides information explaining how and why we collect particular information when you visit our website.