統計学はなぜ哲学の問題になるのか

Abstract

This article explores the intersection of philosophy and statistics by examining philosophical assumptions underlying modern mathematical statistics from onto- logical and epistemological perspectives. Statistics is of interest to philosophers because its mathematical apparatus serves as “models” of philosophical ideas. For instance, much-discussed concepts of the uniformity of nature and natural kinds each correspond to the probability models and statistical models, which are the bread and butter of most statistical methods. Dennett’s real pattern is similar in spirit to information criteria (such as AIC) used to determine the level of complexity that maximizes a model’s predictive ability. In addition, the recent development in machine learning (e.g. deep learning) models implies that these machines have their own “ontology, ” which is far more complicated and possibly efficient in understanding the world than ours. This gives rise to a Quinian problem of radical translation between our and machines’ ontologies, which, I suggest, is a key to the application of AI technologies to our society. From the epistemological perspective, the article compares the Bayesian and classical statistics with the internalist and externalist epistemology, respectively. The comparison elucidates how and in what sense the statistical methods adopted in each camp are considered to justify scientific hypotheses, and also sheds light on their epistemic problems. I conclude with a plea for more studies and interdisciplinary dialogues between statistics and the philosophy of various traditions.