In an effort to better understand how anxiety traits affect one’s ability to learn under uncertainty, and specifically to what extent social factors influence this relation, researchers from Brown University tested the ability of 257 individuals to adapt to neutral, positive and exploitative patterns in social and non-social contexts.
The authors start with the proven premise that healthy humans are generally well-adapted to learning from rewards in social contexts. However, exactly how social learning occurs under uncertainty—like knowing when to trust somebody and how to deal with changes in behavior—are not well understood.
Each subject completed a non-social slot-machine game and a social trust game. In the trust game, individuals were paired with other “online players”, who were in fact machine algorithms programmed to model various strategies.
In both the social and slot-machine games, the probability of winning a round varied dynamically according to certain predetermined pathways. In a “bad first impression” pathway, players made monetary gains in initial rounds and then losses in later ones; a “good first impression” pathway ran inversely; and a “neutral first impression” pathway went from neutral to loss to gains (Fig 3.)
Other than a 4% randomization for noise, the trajectories between social and non-social games were identical, and overall possible payouts between all three scenarios and both games were identical. That is, consistently betting the same amount would result in the same final return, regardless of game or scenario.
The researchers were interested in knowing how players would adapt their betting strategies to changes in the behavior of their “online partners” and slot-machine payout, and their results have interesting implications for the relation between anxiety, social cues, and learning.
Both groups over-invested in the non-social slot-machine game, and made less appropriate decisions as the rewards evolved, lending support to the hypothesis that social contexts augment learning under uncertainty.
Importantly, non-anxious subjects were better at adjusting their strategies in social contexts. They gave more weight to losses than wins when determining strategy and adapted better to exploitative social partners. Subjects with anxiety, on the other hand, continued investing in exploitative social partners, despite clear indications that their partner’s strategy had changed.
Not only does this study have implications for understanding (mal)adaptive learning in social contexts, it also demonstrates how individuals with anxiety traits may fall prey to Machiavellian individuals, as negative social interactions are known to increase anxiety traits, resulting in a vicious cycle of exploitation and inappropriate trust.
Future studies will benefit from further exploring the dynamics of anxiety and maladaptive learning, including proportionality, anxiety type, and possible corrective therapies.
The study, “Anxiety Impedes Adaptive Social Learning Under Uncertainty” appeared in Psychological Science in May, 2020.