A constructivist examination of acceptance bias based on the Recursive Metropolis-Hastings method.

Focusing on the fact that social organizations are finite size queue structures

Abstract

The term symbol emergence refers to the phenomenon where humans share symbols and their referents with others without referencing the thoughts of others. To examine the dynamics of symbol emergence constructively, Inukai and others proposed the Recursive Metropolis-Hastings Naming Game (RMHNG), which represents a joint attention naming game among multiple agents through decentralized Bayesian inference. This study attempts to explain human bias from the perspective of decentralized Bayesian inference. The bias term refers to the suggestion by Okumura and others that humans are more likely to accept the utterances of others than the acceptance probability indicated by the Metropolis-Hastings method. Focusing on the fact that human societies have queue structures, this study hypothesizes that these queue structures cause bias and verifies this hypothesis through experiments. In the experiments, a system comprising queue structures and RMHNG was used. The results suggested that queue structures do not necessarily lead to bias in humans.