Causality and maximum entropy updating

Daniel Hunter
1989 International Journal of Approximate Reasoning  
This paper examines an objection to maximum entropy updating and argues that the problem arises from an inadequate representation of causal information. The objection is that maximum entropy updating renders probabilistically dependent previously independent events when probabilistic information about an effect of the two events is presented. It is believed by many that such information should not render the events dependent. This paper accepts the view that independence should be preserved by
more » ... aximum entropy updating, but argues that it indeed will be when the causal information is presented in an appropriate form. It is argued that presenting the causal information in the form of conditional probabilities is inappropriate. An alternative way of presenting such information, in terms of probabilities of statements known as "'counterfactual conditionals, "' is described. It is shown that when the causal information is expressed by counterfactual conditionals, maximum entropy updating produces results that agree with intuitions shared by its critics and defenders alike about how such information should affect probabilities. An efficient algorithm is given for updating causal information in the form of probabilities of counterfactuals. Finally, the theory of probabilistic counterfactuals developed in this paper is applied to the interpretation of empirical results concerning the way in which people reason under uncertainty.
doi:10.1016/0888-613x(89)90015-7 fatcat:tt6axjwbrvgdpdb6eeisz4bwhq