A Proposal for Agent Simulation of Peer Review

Abstract

Peer review lies at the core of current scientific research. It is composed of a set of social norms, practices and processes that connect the abstract scientific method with the society of people that apply the method. As a social construct, peer review should be understood by building theory-informed models and comparing them with data collection. Both these activities are evolving in the era of automated computation and communication: new modeling tools and large bodies of data become available to the interested researcher. In this paper, starting from abstract principles, we develop and present a model of the peer review process. We also propose a working implementation of a subset of the general model, developed with Jason, a language that implements the BDI architecture for multi agent systems. After running a set of simulations, varying the initial distribution of reviewer skill, we compare the aggregates that our simplified model produces with recent findings, showing how the model can generate data with comparable statistical signatures. We conclude with a list of research questions that could be addressed in future work.