A simple threshold captures the social learning of conventions.

Area of Science:

• Cognitive Science
• Social Science
• Behavioral Economics

Background:

• Learning social conventions from limited, noisy data without explicit instruction is a persistent challenge.
• Existing theories of social learning, including imitation and optimization, inadequately explain observed human behavior in coordination tasks.
• Participants' choices in social network coordination experiments deviate from standard imitation and optimization models.

Purpose of the Study:

• To investigate how individuals learn social conventions in coordination experiments.
• To propose and validate a new model of social learning that better captures human behavior.
• To demonstrate the efficacy of the tolerance principle (TP) in modeling social learning.

Main Methods:

• Conducted coordination experiments with participants interacting in social networks.
• Analyzed behavioral data to identify deviations from imitation and optimization theories.
• Applied the tolerance principle (TP), a parameter-free equation, to model a categorical, two-stage learning process.
• Compared the performance of TP-based agents against imitation, optimization, and Bayesian inference models.

Main Results:

• Human social learning follows a two-stage process: probabilistic behavior followed by stabilization upon reaching a mental threshold.
• TP-based agents demonstrated more accurate social learning than imitation and optimization agents.
• The TP model effectively explained how a critical mass of dissenters can overturn conventions.
• TP outperformed other models in reproducing human learning rates in a dyadic experiment with controlled noise.

Conclusions:

• A simple, mathematical threshold underlies individual and social learning of behavioral conventions.
• The tolerance principle provides a robust framework for understanding social learning across various domains, from language to behavior.
• This threshold-based model offers a more accurate account of human social learning than dominant existing theories.