Cognitive Science, Bayesian Norms and Rules of Evidence
本文分析贝叶斯个人主义模型是否应作为诉讼证据规则的规范，基于认知科学研究指出其关于人类认知的经验假设不准确，因此不适合作为规范来源。
SUMMARY This paper analyses arguments that Bayesian personalist models should be normative in the formulation of evidence rules for litigation (Fienberg and Schervish, 1986; Lempert, 1986). Relying on research from cognitive science, it concludes that the models are poor sources for such norms because their empirical assumptions about human cognition are inaccurate. A rule, by its nature, implies a decription of violations of it (Wright, 1989). Argu- ments that Bayesian analysis, based on subjective probabilities (personalist analysis), should be axiomatic for the formulation of rules of evidence for litigation fail because they assume that decision-making in limited time may be conformed to Bayesian personalist theory (Fienberg and Schervish, 1986). Humans are not natural Bayesian personalists; nor is it likely that they can be Bayesian agents, at least in limited time (Cherniak, 1986). If empirical behaviour in realtime must deviate from Bayesian personalist models, then any truly Bayesian personalist norm for fact finding cannot distinguish between correct and incorrect fact finding. Courts do often rely on probabilistic evidence, e.g. in cases of alleged discrimina- tion or cases involving testimony based on empirical research about recurrent phenomena, such as blood haplotypes. The evidence in those cases, unlike personalist probabilities, requires enumeration of a group based on easily determinable charac- teristics. Because use of that evidence requires neither conjectural prior probabilities nor assumptions that personalist probabilities can be consistent in realtime, it does not rely on the same empirically inadequate model (Thagard, 1988). So the limited use of statistics in those cases has no normative implications for litigation in general. Bayesian personalists argue that, even if the human decision procedure will never be Bayesian personalist, humans can gain pragmatic benefits from following Bayesian models more closely. They point to research, typically that collected by Kahneman et al. (1982), which purports to show that laypersons systematically ignore the laws of probability. A problem used to determine whether subjects will ignore an obvious logical principle must be amenable to an interpretation which would not entail the use of the principle. Without that ambiguity, the problem would only test subjects' rote know- ledge of the obvious logical principle or of the words used in the problem. That ambi- guity conveys additional information. Under Grice's (1989) maxim of co-operation,