The standard conception of misleading evidence has it that e is misleading evidence that p iff e is evidence that p and p is false. I argue that this conception yields incorrect verdicts when we consider what it is for evidence to be misleading with respect to questions like whether p. Instead, we should adopt a conception of misleading evidence according to which e is misleading with respect to a question only if e is in-fact irrelevant to that question – (...) a relation that requires that e fail to be both explanatorily connected to and evidence for any truth that answers the question. This modified conception allows e to be non-misleading evidence with respect to a question even if e is evidence for the false answer to that question. It also points toward a new argument for and explanation of the truth of a central premise – what Maria Lasonen-Aarnio dubs “Entitlement” – that generates the infamous dogmatism puzzle. (shrink)

As inductive decision-making procedures, the inferences made by machine learning programs are subject to underdetermination by evidence and bear inductive risk. One strategy for overcoming these challenges is guided by a presumption in philosophy of science that inductive inferences can and should be value-free. Applied to machine learning programs, the strategy assumes that the influence of values is restricted to data and decision outcomes, thereby omitting internal value-laden design choice points. In this paper, I apply arguments from feminist philosophy of (...) science to machine learning programs to make the case that the resources required to respond to these inductive challenges render critical aspects of their design constitutively value-laden. I demonstrate these points specifically in the case of recidivism algorithms, arguing that contemporary debates concerning fairness in criminal justice risk-assessment programs are best understood as iterations of traditional arguments from inductive risk and demarcation, and thereby establish the value-laden nature of automated decision-making programs. Finally, in light of these points, I address opportunities for relocating the value-free ideal in machine learning and the limitations that accompany them. (shrink)

The Pessimistic Induction over the history of science argues that because most past theories considered empirically successful in their time turn out to be not even approximately true, most present ones probably aren’t approximately true either. But why did past scientists accept those incorrect theories? Kyle Stanford’s ‘Problem of Unconceived Alternatives’ is one answer to that question: scientists are bad at exhausting the space of plausible hypotheses to explain the evidence available to them. Here, I offer another answer, which I (...) call the ‘Problem of Misleading Evidence.’ I argue that this proposal is, in important respects, superior to Stanford’s. (shrink)

In this article, we address a major outstanding question of probabilistic Bayesian epistemology: how should a rational Bayesian agent update their beliefs upon learning an indicative conditional? A number of authors have recently contended that this question is fundamentally underdetermined by Bayesian norms, and hence that there is no single update procedure that rational agents are obliged to follow upon learning an indicative conditional. Here we resist this trend and argue that a core set of widely accepted Bayesian norms is (...) sufficient to identify a normatively privileged updating procedure for this kind of learning. Along the way, we justify a privileged formalization of the notion of ‘epistemic conservativity’, offer a new analysis of the Judy Benjamin problem, and emphasize the distinction between interpreting the content of new evidence and updating one’s beliefs on the basis of that content. (shrink)

Shuford, Albert and Massengill proved, a half century ago, that the logarithmic scoring rule is the only proper measure of inaccuracy determined by a differentiable function of probability assigned the actual cell of a scored partition. In spite of this, the log rule has gained less traction in applied disciplines and among formal epistemologists that one might expect. In this paper we show that the differentiability criterion in the Shuford et. al. result is unnecessary and use the resulting simplified characterization (...) of the logarithmic rule to give novel arguments in favor of it. (shrink)

Wason’s selection task is a paramount experimental problem in the study of human reasoning, often connected with the celebrated ravens paradox in the philosophical literature. Various normative accounts of the selection task rely on a Bayesian approach. Some claim vindication of participants’ rationality. Others don’t, thus following Wason’s original intuition that observed responses are mistaken. In this article we argue that despite claims to the contrary, all these accounts actually speak to the same effect: Wason was right. First, we provide (...) a new accuracy-based analysis of the selection task that includes the existing proposals as special cases. We then show on this basis that none can actually vindicate participants’ rationality. We conclude that all normative renditions considered eventually concur: all in all, Bayesians should follow Wason in the selection task. (shrink)

Two different programmes are in the business of explicating accuracy—the truthlikeness programme and the epistemic utility programme. Both assume that truth is the goal of inquiry, and that among inquiries that fall short of realizing the goal some get closer to it than others. Truthlikeness theorists have been searching for an account of the accuracy of propositions. Epistemic utility theorists have been searching for an account of the accuracy of credal states. Both assume we can make cognitive progress in an (...) inquiry even while falling short of the target. I show that the prospects for combining these two programmes are bleak. A core accuracy principle, proximity, that is universally embraced within the truthlikeness programme turns out to be incompatible with a central principle within the epistemic utility programme, namely propriety. 1Truthlikeness and Epistemic Utility 2Inquiries 3Accuracy for Propositions 4Proximity 5Accuracy for Credal States 6Propriety 7Proximity for Credal States 8Extensionality 9Admissible Weightings 10Propriety Violates Proximity 11Possible Responses 11.1Retreat to convexity 11.2Reject boundedness 11.3Reject additivity 12The Upshot. (shrink)

Deception has long been an important topic in philosophy. However, the traditional analysis of the concept, which requires that a deceiver intentionally cause her victim to have a false belief, rules out the possibility of much deception in the animal kingdom. Cognitively unsophisticated species, such as fireflies and butterflies, have simply evolved to mislead potential predators and/or prey. To capture such cases of “functional deception,” several researchers Machiavellian intelligence II, Cambridge University Press, Cambridge, pp 112–143, 1997; Searcy and Nowicki, The (...) evolution of animal communication, Princeton University Press, Princeton, 2005; Skyrms, Signals, Oxford University Press, Oxford 2010) have endorsed the broader view that deception only requires that a deceiver benefit from sending a misleading signal. Moreover, in order to facilitate game-theoretic study of deception in the context of Lewisian sender-receiver games, Brian Skyrms has proposed an influential formal analysis of this view. Such formal analyses have the potential to enhance our philosophical understanding of deception in humans as well as animals. However, as we argue in this paper, Skyrms’s analysis, as well as two recently proposed alternative analyses, are seriously flawed and can lead us to draw unwarranted conclusions about deception. (shrink)

Epistemic decision theory aims to defend a variety of epistemic norms in terms of their facilitation of epistemic ends. One of the most important components of EpDT is known as a scoring rule. This thesis addresses some problems about scoring rules in EpDT. I consider scoring rules both for precise credences and for imprecise credences. For scoring rules in the context of precise credences, I examine the rationale for requiring a scoring rule to be strictly proper, and argue that no (...) satisfactory justification has been given. I then investigate one possible response to my argument and show the problems with this response. The conclusion is that there is a further need for justifying the requirement that a scoring rule should be strictly proper for precise credences. For scoring rules in the context of imprecise credences, an impossibility result has been established in the literature purporting to show that no strictly proper, continuous and real-valued scoring rule exists. However, a precise statement of the impossibility result requires precise definitions both of strict propriety and of continuity in the context of imprecise credences. Moreover, the result implies that we need to drop one of the three properties - strict propriety, continuity or being real-valued - to have a scoring rule for applying EpDT to imprecise credences. So, firstly, I offer definitions of strict propriety and of continuity and clarify the impossibility result. Then, I investigate what will happen if we drop one of the three properties. I argue that we should drop the property of being real-valued and I offer the general forms of two kinds of strictly proper, continuous and lexicographic scoring rules for imprecise credences. (shrink)

Accuracy-based arguments for conditionalization and probabilism appear to have a significant advantage over their Dutch Book rivals. They rely only on the plausible epistemic norm that one should try to decrease the inaccuracy of one’s beliefs. Furthermore, conditionalization and probabilism apparently follow from a wide range of measures of inaccuracy. However, we argue that there is an under-appreciated diachronic constraint on measures of inaccuracy which limits the measures from which one can prove conditionalization, and none of the remaining measures allow (...) one to prove probabilism. That is, among the measures in the literature, there are some from which one can prove conditionalization, others from which one can prove probabilism, but none from which one can prove both. Hence at present, the accuracy-based approach cannot underwrite both conditionalization and probabilism. (shrink)

There is an ongoing debate about which rule we ought to use for scoring probability estimates. Much of this debate has been premised on scoring-rule monism, according to which there is exactly one best scoring rule. In previous work, I have argued against this position. The argument given there was based on purely a priori considerations, notably the intuition that scoring rules should be sensitive to truthlikeness relations if, and only if, such relations are present among whichever hypotheses are at (...) issue. The present paper offers a new, quasi-empirical argument against scoring-rule monism. This argument uses computational simulations to show that different scoring rules can have different economical consequences, depending on the context of use. (shrink)

Traditionally, epistemologists have distinguished between epistemic and pragmatic goals. In so doing, they presume that much of game theory is irrelevant to epistemic enterprises. I will show that this is a mistake. Even if we restrict attention to purely epistemic motivations, members of epistemic groups will face a multitude of strategic choices. I illustrate several contexts where individuals who are concerned solely with the discovery of truth will nonetheless face difficult game theoretic problems. Examples of purely epistemic coordination problems and (...) social dilemmas will be presented. These show that there is a far deeper connection between economics and epistemology than previous appreciated. (shrink)

In this paper, I challenge the distinction between “epistemic” and “ontic” states propounded by Harrigan and Spekkens (Found Phys 40:125–157, 2010) by pointing out that because knowledge is factive, any state that represents someone’s knowledge about a physical system thereby also represents something about the physical system itself, so there is no such thing as “mere knowledge”. This criticism leads to the reformulation of the main question of the debate: instead of asking whether a given state is ontic or epistemic, (...) we should instead ask whether a given change of a state is ontic or epistemic. In particular, in the context of quantum mechanics, one can ask whether the collapse of the quantum state could be understood as an epistemically successful change of the observer’s beliefs about the complete state of the system that is not associated with any change in the physical reality. I argue that the answer to this question should be in the negative because it is possible that, in a series of measurements, the collapse rule tells us to update a certain state to a different one and then back to the same state; if both of these updates are merely changes of our beliefs, then they could not both be epistemically successful. (shrink)

Sometimes we are interested in how groups are doing epistemically in aggregate. For instance, we may want to know the epistemic impact of a change in school curriculum or the epistemic impact of abolishing peer review in the sciences. Being able to say something about how groups are doing epistemically is especially important if one is interested in pursuing a consequentialist approach to social epistemology of the sort championed by Goldman. According to this approach we evaluate social practices and institutions (...) from an epistemic perspective based on how well they promote the aggregate level of epistemic value across a community. The aim of this paper is to investigate this concept of group epistemic value and defend a particular way of measuring it. (shrink)

Suppose that beliefs come in degrees. How should we then measure the accuracy of these degrees of belief? Scoring rules are usually thought to be the mathematical tool appropriate for this job. But there are many scoring rules, which lead to different ordinal accuracy rankings. Recently, Fallis and Lewis [2016] have given an argument that, if sound, rules out many popular scoring rules, including the Brier score, as genuine measures of accuracy. I respond to this argument, in part by noting (...) that the argument fails to account for verisimilitude—that certain false hypotheses might be closer to the truth than other false hypotheses are. Oddie [forthcoming], however, has argued that no member of a very wide class of scoring rules (the so-called proper scores) can appropriately handle verisimilitude. I explain how to respond to Oddie's argument, and I recommend a class of weighted scoring rules that, I argue, genuinely measure accuracy while escaping the arguments of Fallis and Lewis as well as Oddie. (shrink)