近刊『科学とモデル』（名古屋大学出版会）の著者である、ペンシルバニア大学教授のMichael Weisberg博士をお迎えして、以下のような講演会を行います。Weisberg博士はBiology & Philosophy誌の編集長でもあり、当日は午前中より、同誌への投稿を含めた国際論文投稿How-toに関するワークショップを行います。奮ってご参加下さい。
Speaker: Prof. Michael Weisberg (University of Pennsylvania)
Date: October 21st 2017
Venue: Large conference room in the basement, Faculty of Letters MainBuilding, Yoshida Campus, Kyoto University.
Title: Confirmation Theory for Idealized Models
When a flu pandemic strikes, who should get vaccinated first? What’s our best strategy for minimizing the damage of global climate change? Why is Philadelphia racially segregated? Why do most sexually reproducing species have only two sexes, in roughly even proportions? These and many other scientific and practical problems are studied with highly idealized mathematical and computational models. When should we believe these models and follow the advice they suggest? Philosophy of science tells us that we should believe models when they are well-confirmed, but this simple answer isn’t very helpful here. Traditional confirmation theory explains how empirical evidence bears on the truth of hypotheses and theories, but the highly idealized models at the heart of the life and social sciences are known to be false from the outset. Moreover, classical ideas about confirmation have been developed for relatively simple hypotheses, while many contemporary models have thousands of variables.
Despite these challenges, it is possible to develop an account of model confirmation that can speak to the reliability of models and their results. I will sketch a theory that has two parts: First, theorists validate models, confirming hypotheses about model/target system relations. Second, they employ robustness analysis to investigate the stability of model results. Taken together, validation and robustness tell us when models are reliable and help us understand the appropriate domain of their application. Not only does this theory better align our accounts of scientific method with modern theoretical practice, it also helps us understand when to believe the results of models.