Doxastic Attitudes as Belief-Revision Policies Alexandru Baltag Ben Rodenh¨ auser Sonja Smets ILLC, University of Amsterdam Abstract. While propositional doxastic atti- tudes, like knowledge and belief, capture an agent’s opinion about certain propositions, her attitudes towards sources of information express her opinion about the reliability (or trustwor- thiness) of those sources. If an agent trusts a witness, then she will, within certain limits, tend to accept his testimony as veridical. But if she considers the witness to be a notorious liar, she may come to believe the opposite of what he tells her. In this paper, we put such at- titudes towards sources (or dynamic (doxastic) attitudes ) center stage, and formalize them as belief-revision strategies : policies governing how an agent changess her beliefs whenever new in- formation from a certain (type of) source is re- ceived. We present a semantic, qualitative mod- elling of this notion and investigate its proper- ties. Introduction This paper explores the idea that an agent’s “information uptake” (i.e. what she does with some new informational input) depends substantially on her attitude towards the source of information: her assessment of the reliability of the source. Evidence obtained by direct observation, e.g., is normally con- sidered to be more reliable than testimonial evidence, and testimonies from different wit- nesses may be differently assessed, depend- ing on the reliability of each witness. For- mally, we encode attitudes towards sources as strategies for belief change, applicable to any information received from a particu- lar (type of) source. Such a strategy pre- encodes what the recipient will do if an input from that source is received. Traces of this idea are found scattered across the literature, sometimes formulated in terms of notions like “evidential reliabil- ity” or “epistemic trust”. 1 In Bayesian Epis- temology, the reliability of a source is cap- tured by weights or probabilities attached to the new information, which determine differ- ent ways of processing it (e.g., Bayesian con- ditioning versus Jeffrey conditioning). 2 In Belief Revision theory, various methods for (iterated) belief revision have been proposed that can be understood as corresponding to different attitudes to the incoming informa- tion. 3 While most previous authors have focused on quantitative approaches formalizing de- grees of acceptance or degrees of trust, we propose a qualitative-relational setting that allows us to model a much more general class of doxastic attitudes, including various forms of trust, distrust and “semi-trust”. We give a semantic formalization of these concepts, study the strength order between different dynamic attitudes and the natural opera- tions with them, and show how the standard propositional attitudes can be recovered as fixed points of dynamic attitudes. In the 1 E.g., Spohn (2009) studies a variety of revision operations, parametrized by their “evidential force”, meant to capture the idea that information one ac- cepts comes in various degrees of “firmness”. And Lehrer and Wagner (1981) suggest to model the trust an agent places in another agent’s claims using a no- tion of “epistemic weight”. 2 Jeffrey (2004), Halpern (2003). 3 Boutilier (1996), Spohn (1985, 2009), Nayak (1994), Rott (2004, 2006), among others. 1