Latent motives guide structure learning during adaptive social choice

Abstract

Predicting the behavior of others is an essential part of human cognition that enables strategic social behavior (e.g., cooperation), and is impaired in multiple clinical populations. Despite its ubiquity, social prediction poses a generalization problem that remains poorly understood: We can neither assume that others will simply repeat their past behavior in new settings, nor that their future actions are entirely unrelated to the past. Here we demonstrate that humans solve this challenge using a structure learning mechanism that uncovers other people's latent, unobservable motives, such as greed and risk aversion. In three studies, participants were tasked with predicting the decisions of another player in multiple unique economic games such as the Prisoner's Dilemma. Participants achieved accurate social prediction by learning the hidden motivational structure underlying the player's actions to cooperate or defect (e.g., that greed led to defecting in some cases but cooperation in others). This motive-based abstraction enabled participants to attend to information diagnostic of the player's next move and disregard irrelevant contextual cues. Moreover, participants who successfully learned another's motives were more strategic in a subsequent competitive interaction with that player, reflecting that accurate social structure learning can lead to more optimal social behaviors. These findings demonstrate that advantageous social behavior hinges on parsimonious and generalizable mental models that leverage others' latent intentions.

Optimal resource allocation and its consequences on behavioral strategies, personality traits and preferences

“…For example, seeing someone allocate all their time to finding food would work as a reliable sign that food yields the highest marginal benefit, while other investments such as education may not be prioritized. Additionally, people may infer the risk aversion and patience of individuals based on how they allocate their resources (van Baar et al, 2022). For example, they may infer that someone spending all their resources on food consumption is likely to be more risk averse and less patient than an individual spending only a small portion of their resources on food.…”

A hierarchical Bayesian model of adaptive teaching

“…For example, knowing that a learner can solve differential equations means that they already know how to do algebra and arithmetic; an effective teacher doesn't needs to probe all of these possible sets of knowledge to have an idea of what the learner knows and doesn't know. People's background knowledge can also come in a variety of forms; for example, visual access (Hawkins, Gweon, & Goodman, 2021), latent motives such as greediness or risk aversion (van Baar, Nassar, Deng, & FeldmanHall, 2022),…”

A Bayesian decision-theoretic framework for studying motivated reasoning

“…The hunters must coordinate to capture the meaty stag, but coordinating is uncertain and depends on an infinite chain of expectations about how the other hunter is expected to act. Mathematical models of this game can describe how social structures (Skyrms, 2004) and norms (Bicchieri, 2005) emerge, as well as how people infer another person's motivations and belief-states during strategic interaction (van Baar, Nassar, Deng, & FeldmanHall, 2022). Rational decision-making in these games can have negative consequences…”