The new study shows that healthy people easily recognize when those around them become increasingly untrustworthy—and they react, appropriately enough, by pulling away. But researchers found that the same wasn’t true for those who have significant levels of anxiety.

People who are anxious, the study concludes, continue to trust and invest in people who display increasingly untrustworthy behavior.

“We know from previous research that learning and uncertainty are very closely linked,” says first author Amrita Lamba, a PhD student in the cognitive, linguistic, and psychological sciences department at Brown University.

“This study demonstrates that, if we do not have anxiety, we’re actually able to learn more once we detect uncertainty in social interactions, which helps us to avoid being exploited and to learn who can be trusted. With every uncertain social situation we navigate, with every change in trustworthiness we observe in people, we are fine-tuning our opinions of them and adjusting our relationships with them accordingly.”

For the study, researchers recruited 350 clinically diverse participants—about a quarter of whom showed symptoms of generalized anxiety disorder—and to design two activities that would measure the participants’ abilities to learn and adapt in uncertain situations.

First, the researchers asked participants to gamble on three different online slot machines. Unbeknownst to the subjects, the machines were rigged: One was designed to return consistently good winnings at the beginning, but it became unprofitable after the first few spins; the second consistently lost participants’ money at first, but then it began to become more profitable; the third was inconsistent and ultimately resulted in a zero-sum, zero-loss game for participants.

The researchers note that almost all participants, including those with anxiety symptoms, noticed the machines’ patterns and adjusted their behavior accordingly: They invested less money in the first slot machine after observing a decrease in luck, and they invested more in the second machine when they noticed their returns improving.

Next, the researchers guided participants through a trust game. They told participants they would take part in a multi-person activity where they could give the other players between 10 cents and $1 per round. The money they gave away in each round would immediately quadruple in value. Their fellow players would then be given the option to return a portion of that money to the participant.

Participants were, in reality, playing the trust game with algorithms, not people. Some algorithms initially gave back a large percentage of the quadrupled money but gradually became less generous; others were initially stingy with the amounts they gave back but increased their generosity over time.

Researchers noticed that the healthy people’s behaviors shifted even more quickly in the trust game than in the slot machine game—that is, they reacted to changing behaviors in the other “players” by quickly beginning to give more money to those who gradually began to share generously and quickly learning to give less to those who began to share sparingly. Their fast reflexes suggested to the researchers that most healthy people have an easier time adapting to uncertain social situations than to non-social uncertainty.

“For a long time, there has been debate about whether the way we process and learn from information in the non-social domain is different than how we learn about people,” Lamba says.

“This research demonstrates that we’re uniquely adept at what’s called ‘reward learning’ in social situations—that even though the underlying neural circuitry is largely the same for social and non-social learning, social learning in particular seems to recruit a set of mechanisms that makes us very flexible and quick to adapt when we detect uncertainty or threat in the environment.”

But not all of the participants were as quick to adjust their behavior. Participants who had reported symptoms of general anxiety disorder did give more money to their consistently generous fellow “players” and less money to consistently stingy ones—but when they encountered “players” who started off generous and became increasingly stingy, they continued to give away large portions of their money. In other words, anxious patients seemed to continue to invest in relationships that were no longer paying off for them.

“People who don’t have anxiety are good at updating their belief that someone is trustworthy when the data begin to say, actually, this person isn’t as trustworthy as they were before,” says
Oriel FeldmanHall, an assistant professor of cognitive, linguistic and psychological sciences. “But people with anxiety struggle with this. They try to give people more and more money even though they’re getting signals that these people are less trustworthy than they originally thought.”

The results did ‘t surprise the researchers. They knew from previous research that people with symptoms of anxiety tend to be less tolerant of uncertainty. Even in some non-social situations, FeldmanHall says, anxious people are less able to adapt to evolving situations.

Why did the high-anxiety participants adapt better on the slot machines than in the trust game? Were the anxious participants aware of the changing trustworthiness of the other “players” in the trust game—and if so, why did they choose to ignore the behavioral changes? Researchers say it wasn’t clear what exactly was happening in anxious people’s brains as they played the trust game, and they hope future research will shed light on their thought processes.

“People with anxiety could be unable to detect changes in the social environment, or they could be making a conscious decision to invest in a good relationship rather than protect themselves economically,” Lamba says. “That’s something we would want to tease out in follow-up work. There could be a lot of variation in where the decision-making mechanism diverges—it could be the case that not all anxious people think alike.”

FeldmanHall says this study is an important first step in measuring people’s adaptability to the changing behavior of others. Most previous research has monitored people as they react to others whose behavior is static—hardly a realistic reflection of our ever-changing personalities and social dynamics.

“What we have modeled here is much more akin to the real world of how people behave,” FeldmanHall says. “Our social worlds are incredibly important to learn about, because they have the most impact on our daily well-being. We need more studies that model our social worlds more accurately.”