Theoretical and Computational Neuroscience
Author: Alejandro Ramos Usaj | email: aler.usaj@gmail.com
Alejandro Ramos Usaj 1°, Guillermo Solovey 1°
1° Instituto de Calculo, FCEyN, UBA.
Recent evidence has pointed out that confidences drives a confirmation bias of new information at the brain level and a recent agent-based model showed that having a confirmation bias is adaptive when coupled with efficient individual metacognition. To extend these lines of resarch we consider social learning as type of information integration process in which individuals integrate personal and social information to improve their own accuracy. We develop a model were an individual samples information from a noisy stimulus, formalized as sampling from a known distribution with a mean fixed at the true signal for the stimulus and a variance reflecting individual reliability (or hability). In light of this information the individual makes a binary decision following an indicator function with fixed thresholds and we estimate confidence in this initial decision by calculating the log-odds in favour of the chosen option. For each individual we calculate metacognitive efficiency using stablished signal decision theory measures modulated by the correlation between individual reliability and the confidence calculation. We show that optimal social information use does not depend solely on individual metacognition but also on the number of agents providing the individual with social information and the reliability of each agent. Finally we explore the theorethical and empirical implications of our findings for future works.