Replication in the Deception and Convergence of Opinions Problem

Reported results of experiments are usually trustworthy, but some of them might be obtained from errors or deceptive behavior. When an agent only read articles about experimental results and use the articles to update his subjective opinions about different theories, the existence of deception can have severe consequences. An earlier attempt to solve that problem suggested that reading replicated results would solve the problems associated with the existence of deception. In this paper, we show that result is not a general case and, for experiments subject to statistical uncertainty, the solution is simply wrong. The analysis of the effect of replicated experiments is corrected here by introducing a differentiation between honest and dishonest mistakes. We observe that, although replication does solve the problem of no convergence, under some circumstances, it is not enough for achieving a reasonable amount of certainty for a realistic number of read reports of experiments.