Bernhard Salow

I argue that a little learning is often dangerous even for ideal reasoners who are operating in extremely simple scenarios and know all the relevant facts about how the evidence is generated. More precisely, I show that, on many plausible ways of assigning value to a credence in a hypothesis H, ideal Bayesians should sometimes expect other ideal Bayesians to end up with a worse credence if they gather additional evidence, even when they agree completely about the likelihoods of the evidence given both H and not-H. This provides a new reason for pessimism about the prospect of disagreeing individuals resolving their disagreement by consulting additional evidence.

Recent work in epistemology implicitly assumes that we can measure the quality or strength of someone’s knowledge (whether understood intuitively, or by its normative connections to action, inquiry, belief, or assertion) by the number of iterations it permits. I show that this idea is hopeless, because, even in set-ups that look maximally friendly, one can construct cases where someone goes from having available only a single iteration of knowledge that p to having arbitrarily many such iterations, without their knowledge that p becoming better in any way.

We use the normality framework of Goodman and Salow (2018, 2021, 2023b) to investigate of dynamics of rational belief. The guiding idea is that people are entitled to believe that their circumstances aren’t especially abnormal. More precisely, a rational agent’s beliefs rule out all and only those possibilities that are either (i) ruled out by their evidence or (ii) sufficiently less normal than some other possibility not ruled out by their evidence. Working within this framework, we argue that the logic of rational belief revision is much weaker than is usually supposed. We do so by isolating a natural family of orthodox principles about belief revision, describing realistic cases in which these principles seem to fail, and showing how these counterexamples are predicted by independently motivated models of the cases in question. In these models, whether one evidential possibility counts as sufficiently less normal than another is determined by underlying probabilities (together with a contextually determined question). We argue that the resulting probabilistic account of belief compares favorably with other such accounts, including Lockeanism (Foley, 1993), a ‘stability’ account inspired by Leitgeb (2017), the ‘tracking theory’ of Lin and Kelly (2012), and the influential precursor of Levi (1967). We show that all of these accounts yield subtly different but similarly heterodox logics of belief revision.

The strongest version of the dogmatism puzzle argues that, when we know something, we should resolve to ignore or avoid evidence against it. The best existing responses are fallibilist, and hold that decisions should be governed by underlying probabilities rather than our knowledge. I argue that this is an overreaction: by paying close attention to the principles governing belief-revision, and to subtly different ways in which knowledge can govern decision-making, we can dissolve the puzzle without the need for controversial theoretical commitments. The resulting solution demonstrates fruitful and underexplored points of interaction between ‘traditional’ epistemology and ‘formal’ theories of belief-revision, and clears the ground for more systematic theorizing about how and when we should be open to changing our minds.

Philosophers have recently attempted to justify particular belief revision procedures by arguing that they are the optimal means towards the epistemic end of accurate credences. These attempts, however, presuppose that means should be evaluated according to classical expected utility theory; and there is a long tradition maintaining that expected utility theory is too restrictive as a theory of means–end rationality, ruling out too many natural ways of taking risk into account. In this paper, we investigate what belief-revision procedures are supported by accuracy-theoretic considerations once we depart from expected utility theory to allow agents to be risk-sensitive. We argue that if accuracy-theoretic considerations tell risk-sensitive agents anything about belief-revision, they tell them the same thing they tell risk-neutral agents: they should conditionalize.

We offer a general framework for theorizing about the structure of knowledge and belief in terms of the comparative normality of situations compatible with one’s evidence. The guiding idea is that, if a possibility is sufficiently less normal than one’s actual situation, then one can know that that possibility does not obtain. This explains how people can have inductive knowledge that goes beyond what is strictly entailed by their evidence. We motivate the framework by showing how it illuminates knowledge about the future, knowledge of lawful regularities, knowledge about parameters measured using imperfect instruments, the connection between knowledge, belief, and probability, and the dynamics of knowledge and belief in response to new evidence.

There are many things—call them ‘experts’—that you should defer to in forming your opinions. The trouble is, many experts are modest: they’re less than certain that they are worthy of deference. When this happens, the standard theories of deference break down: the most popular (“Reflection”-style) principles collapse to inconsistency, while their most popular (“New-Reflection”-style) variants allow you to defer to someone while regarding them as an anti-expert. We propose a middle way: deferring to someone involves preferring to make any decision using their opinions instead of your own. In a slogan, deferring opinions is deferring decisions. Generalizing the proposal of Dorst (2020a), we first formulate a new principle that shows exactly how your opinions must relate to an expert’s for this to be so. We then build off the results of Levinstein (2019) and Campbell-Moore (2020) to show that this principle is also equivalent to the constraint that you must always expect the expert’s estimates to be more accurate than your own. Finally, we characterize the conditions an expert’s opinions must meet to be worthy of deference in this sense, showing how they sit naturally between the too-strong constraints of Reflection and the too-weak constraints of New Reflection.

An important question in epistemology is whether the KK principle is true, i.e., whether an agent who knows that p is also thereby in a position to know that she knows that p. We explain how a “transparency” account of self-knowledge, which maintains that we learn about our attitudes towards a proposition by reflecting not on ourselves but rather on that very proposition, supports an affirmative answer. In particular, we show that such an account allows us to reconcile a version of the KK principle with an “externalist” or “reliabilist” conception of knowledge commonly thought to make that principle particularly problematic.

Epistemologists have recently noted a tension between (i) denying access internalism, and (ii) maintaining that rational agents cannot be epistemically akratic, believing claims akin to ‘p, but I shouldn’t believe p’. I bring out the tension, and develop a new way to resolve it. The basic strategy is to say that access internalism is false, but that counterexamples to it are ‘elusive’ in a way that prevents rational agents from suspecting that they themselves are counterexamples to the internalist principles. I argue that this allows us to do justice to the motivations behind both (i) and (ii). And I explain in some detail what a view of evidence that implements this strategy, and makes it independently plausible, might look like.

The status of the knowledge iteration principles in the account provided by Lewis in “Elusive Knowledge” is disputed. By distinguishing carefully between what in the account describes the contribution of the attributor’s context and what describes the contribution of the subject’s situation, we can resolve this dispute in favour of Holliday’s claim that the iteration principles are rendered invalid. However, that is not the end of the story. For Lewis’s account still predicts that counterexamples to the negative iteration principle ) come out as elusive: such counterexamples can occur only in possibilities which the attributors of knowledge are ignoring. This consequence is more defensible than it might look at first sight.

Suppose you’d like to believe that p, whether or not it’s true. What can you do to help? A natural initial thought is that you could engage in Intentionally Biased Inquiry : you could look into whether p, but do so in a way that you expect to predominantly yield evidence in favour of p. This paper hopes to do two things. The first is to argue that this initial thought is mistaken: intentionally biased inquiry is impossible. The second is to show that reflections on intentionally biased inquiry strongly support a controversial ‘access’ principle which states that, for all p, if p is part of our evidence, then that p is part of our evidence is itself part of our evidence.