Effects of Fact Mutability in the Interpretation of Counterfactuals

摘要: Effects of Fact Mutability in the Interpretation Counterfactuals Morteza Dehghani (morteza@northwestern.edu) Department EECS, 2145 Sheridan Rd Evanston, IL 60208-0834 USA Rumen Iliev (r-iliev@northwestern.edu) Psychology, 2029 Road 60208-2710 Stefan Kaufmann (kaufmann@northwestern.edu) Linguistics, 2016 60208-4090 2000) and psychological evidence supporting this model. Also same section, we review Hiddelston’s (2005) extension to Causal Bayesian Networks. We then discuss Kahneman Miller’s Norm Theory (1986) two experiments designed test correctness predictions AI models. Abstract This paper explores relationship between fact mutability, intervention human evaluation counterfactual conditionals. Two are reported that show effects causal strength distance on mutability intervention. Subjects’ answers compared three models reasoning Artificial Intelligence. comparison demonstrates logical inferences graph topologies not sufficient for modeling all aspects reasoning. The Stalnaker/Lewis Many inspired by model-theoretic accounts Stalnaker (1968) Lewis (1973). Minor differences aside, both crucially rely a notion comparative similarity possible worlds relative “actual” world i evaluation. Thus Lewis’s truth conditions state ‘If it were Antecedent, would be Consequent’ (A € C) is true at “if only if, if there an antecedent- accessible from i, consequent holds every antecedent-world least as close certain antecedent-world” (p. 49). Assuming simplicity set A-worlds maximally similar means A C those A-worlds. account various properties counterfactuals imposing underlying relation, but neither attempts detailed analysis notion. However, (1979), noting his theory “must fleshed out with appropriate will differ context context,” gives informal ranked list general “weights or priorities” determining similarity: first, avoid big, widespread, diverse violations law; second, maximize spatio-temporal region perfect match particular fact; third, small, localized fourth, secure approximate facts. Despite informality these guidelines, one can discern priority laws over fact, “big” discrepancies “small” ones. Much subsequent work based intuitions viewed make more precise. view Keywords: Counterfactual Reasoning; Networks; Theory. Introduction has long been subject interest philosophers (e.g. Leibniz, 1686; Hume, 1748; Goodman, 1947; Lewis, 1973; Stalnaker, 1968). More recently linguists, psychologist, later cognitive scientists, have become interested study concept “what been” how about events almost happened provides us knowledge cannot deduced simple facts indicative conditionals Miller, 1986; Sternberg Gastel, 1989). In last decades several model (Ginsberg, Costello McCarthy, 1999; Pearl, 2000; Hiddleston, 2005, among others). advantage such formal they precise cases. briefly demonstrate using do capture full spectrum Specifically, illustrate affect interpretation statements. conclusion, argue utilizing findings crucial building cognitively plausible computational First, counterfactuals. Next, Networks (Spirtes, Glymour, Scheines, 1993;