Prior preferences beneficially influence social and non-social learning

Tor Tarantola, Dharshan Kumaran, Peter Dayan, Benedetto De MartinoNature Communications 2017 · DOI

Abstract

Our personal preferences affect a broad array of social behaviors. This includes the way we learn the preferences of others, an ability that often relies on limited or ambiguous information. Here we report an egocentric influence on this type of social learning that is reflected in both performance and response times. Using computational models that combine inter-trial learning and intra-trial choice, we find transient effects of participants' preferences on the learning process, through the influence of priors, and persistent effects on the choice process. A second experiment shows that these effects generalize to non-social learning, though participants in the social learning experiment appeared to additionally benefit by using their knowledge about the popularity of certain preferences. We further find that the domain-general egocentric influences we identify can yield performance advantages in uncertain environments. People often assume that other people share their preferences, but how exactly this bias manifests itself in learning and decision-making is unclear. Here, authors show that a person's own preferences influence learning in both social and non-social situations, and that this bias improves performance.

Introduction

We often assume that others share our tastes—a phenomenon known as "social projection." This study reveals that this egocentric bias isn't just a social quirk, but a domain-general learning strategy that influences both our initial beliefs and our persistent decision-making process. Crucially, the researchers demonstrate through computational modeling and simulations that this bias is actually a "rational" heuristic that improves learning performance in uncertain environments.

Methods

The researchers conducted two main behavioral experiments using food snack preferences. In the Social Learning task, participants learned the preferences of a "partner" (a pilot participant) through 600 trials of two-alternative forced choices with probabilistic feedback (80% accurate). In the Non-social Learning task, a different group performed the exact same task but was told they were learning "correct" items in a set, with no mention of another person’s preferences.

Experimental task flow for social learning, illustrating the sequence from partner's hidden choice to participant response and feedback.

To analyze the data, they employed Hierarchical Bayesian Modeling and Drift Diffusion Models (DDM). The DDM decomposes choice and response time (RT) data into underlying processes like evidence accumulation rate and starting point bias. By combining inter-trial Bayesian learning with intra-trial DDM, the authors could pinpoint exactly where the participants' own preferences were influencing their behavior.

Key Insights and Ideas

  • Domain Generality: The egocentric bias appeared in both social and non-social groups. This suggests that using our own values as a template is a fundamental learning mechanism, not just a social strategy.
  • The Dual Influence: The best-fitting computational model showed that preferences influence behavior at two levels: as a prior (affecting the starting belief) and as a choice bias (persistently affecting the decision process even after evidence is accumulated).
  • Social Intelligence: While both groups projected their own tastes, the social group was uniquely sensitive to item popularity. They performed better on the first trial because they could distinguish between their own "weird" tastes and what is generally liked by the public.
  • Efficiency of Bias: In simulations, "biased" agents outperformed "neutral" agents. Because individual preferences are often correlated with the majority, projecting one's own taste provides a statistically significant "head start" that outweighs the cost of being wrong occasionally.

Key Results

Participants were significantly more accurate and faster when the "correct" answer matched their own preference (congruence). Even after learning plateaued (the last 10 trials), the influence of the participant's own preference remained a significant predictor of performance.

Comparison of behavioral data and model predictions for non-social learning, showing higher accuracy when preferences are shared.

Simulated agents using the "Dual Influence" strategy outperformed neutral learners by an average of 0.71 percentage points. On the first trial, the advantage was as high as 7.2 percentage points. The model predictions closely matched the observed behavioral data, confirming that the egocentric bias facilitates faster learning in these environments.

Discussion and Future Work

The authors acknowledge that their experiment used a stable environment (the 80/20 feedback ratio never changed). In the real world, environments are often volatile, meaning the "correct" choice might change over time. Future research could explore whether egocentric biases remain helpful or become a hindrance when the underlying rules of the environment are shifting.

Additionally, while the study provides a clear computational account, the neural architecture supporting this integration—specifically how the brain distinguishes between "my value" and "learned value"—remains a promising area for neuroimaging studies. Finally, the researchers suggest that more complex social inferences might involve hierarchical versions of the models presented here.

Summary updated

Abstract

Our personal preferences affect a broad array of social behaviors. This includes the way we learn the preferences of others, an ability that often relies on limited or ambiguous information. Here we report an egocentric influence on this type of social learning that is reflected in both performance and response times. Using computational models that combine inter-trial learning and intra-trial choice, we find transient effects of participants' preferences on the learning process, through the influence of priors, and persistent effects on the choice process. A second experiment shows that these effects generalize to non-social learning, though participants in the social learning experiment appeared to additionally benefit by using their knowledge about the popularity of certain preferences. We further find that the domain-general egocentric influences we identify can yield performance advantages in uncertain environments.People often assume that other people share their preferences, but how exactly this bias manifests itself in learning and decision-making is unclear. Here, authors show that a person's own preferences influence learning in both social and non-social situations, and that this bias improves performance.
BibTeX
@inproceedings{tarantola2017prior,
  author = {Tor Tarantola and Dharshan Kumaran and Peter Dayan and Benedetto De Martino},
  title = {{Prior preferences beneficially influence social and non-social learning}},
  booktitle = {Nature Communications},
  year = {2017},
  doi = {10.1038/s41467-017-00826-8},
}
Figure