Event segmentation theory posits that people segment continuous experience into discrete events and that event boundaries occur when there are large transient increases in prediction error. Here, we set out to test this theory in the context of story listening, by using a deep learning language model (GPT‐2) to compute the predicted probability distribution of the next word, at each point in the story. For three stories, we used the probability distributions generated by GPT‐2 to compute the time series of (...) prediction error. We also asked participants to listen to these stories while marking event boundaries. We used regression models to relate the GPT‐2 measures to the human segmentation data. We found that event boundaries are associated with transient increases in Bayesian surprise but not with a simpler measure of prediction error (surprisal) that tracks, for each word in the story, how strongly that word was predicted at the previous time point. These results support the hypothesis that prediction error serves as a control mechanism governing event segmentation and point to important differences between operational definitions of prediction error. (shrink)

The inconsistent relation between word frequency and free recall performance and the non-monotonic relation found between the two cannot all be explained by current theories. We propose a theoretical framework that can explain all extant results. Based on an ecological psychology analysis of the free recall situation in terms of environmental and informational resources available to the participants, we propose that because participants’ cognitive system has been shaped by their native language, free recall performance is best understood as the end (...) result of relational properties that preexist the experimental situation and of the way the words from the experimental list interact with those. In addition to this, we borrow from predictive coding theory the idea that the brain constantly predicts “what is coming next” so that it is mainly prediction errors that will propagate information forward. Our ecological psychology analysis indicates there will be “prediction errors” because the word frequency distribution in an experimental word list is inevitably different from the particular Zipf’s law distribution of the words in the language that shaped participants’ brains. We further propose the particular distributional discrepancies inherent to a given word list will trigger, as a function of the words that are included in the list, their order, and of the words that are absent from the list, a surprisal signal in the brain, something that is isomorphic to the concept of Bayesian surprise. The precise moment when Bayesian surprise is triggered will determine to what word of the list that Bayesian surprise will be associated with, and the word the Bayesian surprise will be associated with will benefit from it and become more memorable as a direct function of the magnitude of the surprisal. Two experiments are presented that show a proxy of Bayesian surprise explains the free recall performance and that no effect of word frequency is found above and beyond the effect of that proxy variable. We then discuss how our view can account for all data extant in the literature on the effect of word frequency on free recall. (shrink)

Diversity of evidence is widely claimed to be crucial for evidence amalgamation to have distinctive epistemic merits. Bayesian epistemologists capture this idea in the variety-of-evidence thesis: ceteris paribus, the strength of confirmation of a hypothesis by an evidential set increases with the diversity of the evidential elements in that set. Yet, formal exploration of this thesis has shown that it fails to be generally true. This article demonstrates that the thesis fails in even more circumstances than recent results would (...) lead us to expect. Most importantly, it can fail whatever the chance that the evidential sources are unreliable. Our results hold for two types of degrees of variety: reliability independence and testable aspect independence. We conclude that the variety-of-evidence thesis can, at best, be interpreted as an exception-prone rule of thumb. (shrink)

Diversity of evidence is widely claimed to be crucial for evidence amalgamation to have distinctive epistemic merits. Bayesian epistemologists capture this idea in the variety-of-evidence thesis: ceteris paribus, the strength of confirmation of a hypothesis by an evidential set increases with the diversity of the evidential elements in that set. Yet, formal exploration of this thesis has shown that it fails to be generally true. This article demonstrates that the thesis fails in even more circumstances than recent results would (...) lead us to expect. Most importantly, it can fail whatever the chance that the evidential sources are unreliable. Our results hold for two types of degrees of variety: reliability independence and testable aspect independence. We conclude that the variety-of-evidence thesis can, at best, be interpreted as an exception-prone rule of thumb. (shrink)

Bayesian cognitive science is a research programme that relies on modelling resources from Bayesian statistics for studying and understanding mind, brain, and behaviour. Conceiving of mental capacities as computing solutions to inductive problems, Bayesian cognitive scientists develop probabilistic models of mental capacities and evaluate their adequacy based on behavioural and neural data generated by humans (or other cognitive agents) performing a pertinent task. The overarching goal is to identify the mathematical principles, algorithmic procedures, and causal mechanisms that (...) enable cognitive agents to take uncertainty into account and weigh it appropriately in producing adaptive behaviour. -/- The appeal of Bayesian cognitive science derives from its transparency, normativity, and unifying power. Resources from Bayesian statistics allow cognitive scientists to characterise transparently, in terms of random variables and probabilistic dependencies between them, the inductive problems that many mental capacities are presumed to solve. Given this characterisation, researchers can define an upper bound on one’s performance on that problem and use it as a benchmark for interpreting experimental results, formulating hypotheses about why and when deviations from optimal performance should be expected, and seeking explanations of mental capacities within one encompassing theoretical framework. -/- Although critics question the testability, explanatory power, and plausibility of Bayesian models of mental capacities, Bayesian cognitive science has proved itself to be an incredibly fruitful research programme. Fuelled by empirical and theoretical results from psychology, neuroscience, philosophy, computer science, artificial intelligence, and evolution, Bayesian cognitive science has: advanced our understanding of why cognitive systems should keep track of uncertainty and use it to facilitate adaptive behaviour; clarified how the brain might encode information about the uncertainty of its stimuli and perform computations with probability distributions; and crystallised the insight that cognitive agents’ management of uncertainty might be grounded in predictive processes aimed at avoiding surprising exchanges with the environment. (shrink)

The full Bayesian signi/cance test (FBST) for precise hypotheses is presented, with some illustrative applications. In the FBST we compute the evidence against the precise hypothesis. We discuss some of the theoretical properties of the FBST, and provide an invariant formulation for coordinate transformations, provided a reference density has been established. This evidence is the probability of the highest relative surprise set, “tangential” to the sub-manifold (of the parameter space) that defines the null hypothesis.

We appeal to the theory of Bayesian Networks to model different strategies for obtaining confirmation for a hypothesis from experimental test results provided by less than fully reliable instruments. In particular, we consider (i) repeated measurements of a single test consequence of the hypothesis, (ii) measurements of multiple test consequences of the hypothesis, (iii) theoretical support for the reliability of the instrument, and (iv) calibration procedures. We evaluate these strategies on their relative merits under idealized conditions and show some (...) surprising repercussions on the variety-of-evidence thesis and the Duhem-Quine thesis. (shrink)

Abstract: The Pull Bayesian Significance Test (FBST) for precise hy- potheses is applied to a Multivariate Normal Structure (MNS) model. In the FBST we compute the evidence against the precise hypothesis. This evi- dence is the probability of the Highest Relative Surprise Set (HRSS) tangent to the sub-manifold (of the parameter space) that defines the null hypothesis. The MNS model we present appears when testing equivalence conditions for genetic expression measurements, using micro-array technology.

Viewing the brain as an organ of approximate Bayesian inference can help us understand how it represents the self. We suggest that inferred representations of the self have a normative function: to predict and optimise the likely outcomes of social interactions. Technically, we cast this predict-and-optimise as maximising the chance of favourable outcomes through active inference. Here the utility of outcomes can be conceptualised as prior beliefs about final states. Actions based on interpersonal representations can therefore be understood as (...) minimising surprise – under the prior belief that one will end up in states with high utility. Interpersonal representations thus serve to render interactions more predictable, while the affective valence of interpersonal inference renders self-perception evaluative. Distortions of self-representation contribute to major psychiatric disorders such as depression, personality disorder and paranoia. The approach we review may therefore operationalise the study of interpersonal representations in pathological states. (shrink)

The plausibility of uncommon events and miracles based on testimony of such an event has been much discussed. When analyzing the probabilities involved, it has mostly been assumed that the common events can be taken as data in the calculations. However, we usually have only testimonies for the common events. While this difference does not have a significant effect on the inductive part of the inference, it has a large influence on how one should view the reliability of testimonies. In (...) this work, a full Bayesian solution is given for the more realistic case, where one has a large number of testimonies for a common event and one testimony for an uncommon event. A free-running parameter is given for the unreliability of testimony, to be determined from data. It is seen that, in order for there to be a large amount of testimonies for a common event, the testimonies will probably be quite reliable. For this reason, because the testimonies are quite reliable based on the testimonies for the common events, the probability for the uncommon event, given a testimony for it, is also higher. Perhaps surprisingly, in the simple case, the increase in plausibility from testimony for the uncommon events is of the same magnitude as the decrease in plausibility from induction. In summary, one should be more open-minded when considering the plausibility of uncommon events. (shrink)

According to Bayesian confirmation theory, evidence E (incrementally) confirms (or supports) a hypothesis H (roughly) just in case E and H are positively probabilistically correlated (under an appropriate probability function Pr). There are many logically equivalent ways of saying that E and H are correlated under Pr. Surprisingly, this leads to a plethora of non-equivalent quantitative measures of the degree to which E confirms H (under Pr). In fact, many non-equivalent Bayesian measures of the degree to which E (...) confirms (or supports) H have been proposed and defended in the literature on inductive logic. I provide a thorough historical survey of the various proposals, and a detailed discussion of the philosophical ramifications of the differences between them. I argue that the set of candidate measures can be narrowed drastically by just a few intuitive and simple desiderata. In the end, I provide some novel and compelling reasons to think that the correct measure of degree of evidential support (within a Bayesian framework) is the (log) likelihood ratio. The central analyses of this research have had some useful and interesting byproducts, including: (i ) a new Bayesian account of (confirmationally) independent evidence, which has applications to several important problems in con- firmation theory, including the problem of the (confirmational) value of evidential diversity, and (ii ) novel resolutions of several problems in Bayesian confirmation theory, motivated by the use of the (log) likelihood ratio measure, including a reply to the Popper-Miller critique of probabilistic induction, and a new analysis and resolution of the problem of irrelevant conjunction (a.k.a., the tacking problem). (shrink)

According to [Bayesian] models” in cognitive neuroscience, says a recent textbook, “the human mind behaves like a capable data scientist”. Do they? That is to say, do such model show we are rational? I argue that Bayesian models of cognition, perhaps surprisingly, do not and indeed cannot, show that we are Bayesian-rational. The key reason is that such models appeal to approximations, a fact that carries significant implications. After outlining the argument, I critique two responses, seen in (...) recent cognitive neuroscience. One says that the mind can be seen as approximately Bayes-rational, while the other reconceives norms of rationality. (shrink)

It has been widely asserted that humans have a “Bayesian brain.” Surprisingly, however, this term has never been defined and appears to be used differently by different authors. I argue that Bayesian brain should be used to denote the realist view that brains are actual Bayesian machines and point out that there is currently no evidence for such a claim.

In applying Bayes’s theorem to the history of science, Bayesians sometimes assume – often without argument – that they can safely ignore very implausible theories. This assumption is false, both in that it can seriously distort the history of science as well as the mathematics and the applicability of Bayes’s theorem. There are intuitively very plausible counter-examples. In fact, one can ignore very implausible or unknown theories only if at least one of two conditions is satisfied: (i) one is certain (...) that there are no unknown theories which explain the phenomenon in question, or (ii) the likelihood of at least one of the known theories used in the calculation of the posterior is reasonably large. Often in the history of science, a very surprising phenomenon is observed, and neither of these criteria is satisfied. (shrink)

The paper provides a formal proof that efficient estimates of parameters, which vary as as little as possible when measurements are repeated, may be expected to provide more accurate predictions. The definition of predictive accuracy is motivated by the work of Akaike (1973). Surprisingly, the same explanation provides a novel solution for a well known problem for standard theories of scientific confirmation — the Ravens Paradox. This is significant in light of the fact that standard Bayesian analyses of the (...) paradox fail to account for the predictive utility of universal laws like All ravens are black. (shrink)

ABSTRACTRational agents have consistent beliefs. Bayesianism is a theory of consistency for partial belief states. Rational agents also respond appropriately to experience. Dogmatism is a theory of how to respond appropriately to experience. Hence, Dogmatism and Bayesianism are theories of two very different aspects of rationality. It's surprising, then, that in recent years it has become common to claim that Dogmatism and Bayesianism are jointly inconsistent: how can two independently consistent theories with distinct subject matter be jointly inconsistent? In this (...) essay I argue that Bayesianism and Dogmatism are inconsistent only with the addition of a specific hypothesis about how the appropriate responses to perceptual experience are to be incorporated into the formal models of the Bayesian. That hypothesis isn't essential either to Bayesianism or to Dogmatism, and so Bayesianism and Dogmatism are jointly consistent. That leaves the matter of how experiences and credences are related, a... (shrink)

In this paper we present the results of a simulation study of credence developments in groups of communicating Bayesian agents, as they update their beliefs about a given proposition p. Based on the empirical literature, one would assume that these groups of rational agents would converge on a view over time, or at least that they would not polarize. This paper presents and discusses surprising evidence that this is not true. Our simulation study shows that these groups of (...) class='Hi'>Bayesian agents show group polarization behavior under a broad range of circumstances. This is, we think, an unexpected result, that raises deeper questions about whether the kind of polarization in question is irrational. If one accepts Bayesian agency as the hallmark of epistemic rationality, then one should infer that the polarization we find is also rational. On the other hand, if we are inclined to think that there is something epistemically irrational about group polarization, then something must be off in the model employed in our simulation study. We discuss several possible interfering factors, including how epistemic trust is defined in the model. Ultimately, we propose that the notion of Bayesian agency is missing something in general, namely the ability to respond to higher-order evidence. (shrink)

In this paper, we provide a Bayesian analysis of the well-known surprise exam paradox. Central to our analysis is a probabilistic account of what it means for the student to accept the teacher's announcement that he will receive a surprise exam. According to this account, the student can be said to have accepted the teacher's announcement provided he adopts a subjective probability distribution relative to which he expects to receive the exam on a day on which he (...) expects not to receive it. We show that as long as expectation is not equated with subjective certainty there will be contexts in which it is possible for the student to accept the teacher's announcement, in this sense. In addition, we show how a Bayesian modeling of the scenario can yield plausible explanations of the following three intuitive claims: (1) the teacher's announcement becomes easier to accept the more days there are in class; (2) a strict interpretation of the teacher's announcement does not provide the student with any categorical information as to the date of the exam; and (3) the teacher's announcement contains less information about the date of the exam the more days there are in class. To conclude, we show how the surprise exam paradox can be seen as one among the larger class of paradoxes of doxastic fallibilism, foremost among which is the paradox of the preface. (shrink)

Coherence is a key concept in many accounts of epistemic justification within ‘traditional’ analytic epistemology. Within formal epistemology, too, there is a substantial body of research on coherence measures. However, there has been surprisingly little interaction between the two bodies of literature. The reason is that the existing formal literature on coherence measure operates with a notion of belief system that is very different from—what we argue is—a natural Bayesian formalisation of the concept of belief system from traditional epistemology. (...) Therefore, formal epistemology has so far only been concerned with one particular—arguably not even very natural—way of formalising coherence of belief systems; it has by no means refuted the viability of coherentism. In contrast to the existing literature, we formalise belief systems as families of assignments of (conditional) degrees of belief (which may be compatible with several subjective probability measures). Within this framework, we propose a Bayesian formalisation of the thrust of BonJour’s coherence concept in The structure of empirical knowledge (Harvard University Press, Cambridge, 1985), using a combination of Bayesian confirmation theory and basic graph theory. In excursions, we introduce graded notions for both logical and probabilistic consistency of belief systems—the latter being based on certain geometrical structures induced by probabilistic belief systems. For illustration, we reconsider BonJour’s “ravens” challenge (op. cit., p. 95f.). Finally, potential objections to our proposed formal coherence notion are explored. (shrink)

The debates between Bayesian, frequentist, and other methodologies of statistics have tended to focus on conceptual justifications, sociological arguments, or mathematical proofs of their long run properties. Both Bayesian statistics and frequentist (“classical”) statistics have strong cases on these grounds. In this article, we instead approach the debates in the “Statistics Wars” from a largely unexplored angle: simulations of different methodologies’ performance in the short to medium run. We conducted a large number of simulations using a straightforward decision (...) problem based around tossing a coin with unknown bias and then placing bets. In this simulation, we programmed four players, inspired by Bayesian statistics, frequentist statistics, Jon Williamson’s version of Objective Bayesianism, and a player who simply extrapolates from observed frequencies to general frequencies. The last player functions as a benchmark: a statistical methodology should at least outperform a crude form of induction. We focused on the performance of these methodologies in guiding the players towards good decisions. Unlike an earlier simulation study of this type, we found no systematic difference in performance between the Bayesian and frequentist players, provided the Bayesian used a flat prior and the frequentist used a low confidence level. Unlike that study, we were able to use Big Data methods to mitigate problems of random error in the simulation results. The Williamsonian player, who is a novel element of our study, also had no systematic differences in their performance, provided that they use a low confidence level. These players performed similarly even in the very short run, when players were making different decisions. Our study indicates that all three methodologies should be taken seriously by philosophers and practitioners of statistics. However, the frequentist and Williamsonian performed poorly when their confidence levels were high, and the Bayesian was surprisingly harmed by biased priors, providing some unexpected lessons for these methodologies when facing this type of decision problem. (shrink)

In this paper we discuss the ``admissibility troubles'' for Bayesian accounts of direct inference proposed in, which concern the existence of surprising, unintuitive defeaters even for mundane cases of direct inference. We first show that one could reasonably suspect that the source of these troubles was informal talk about higher-order probabilities: for cardinality-related reasons, classical probability spaces abound in defeaters for direct inference. We proceed to discuss the issues in the context of the rigorous framework of Higher Probability Spaces. (...) However, we show that the issues persist; we prove a few facts which pertain both to classical probability spaces and to HOPs, in our opinion capturing the essence of the problem. In effect we strengthen the message from the admissibility troubles: they arise not only for approaches using classical probability spaces---which are thus necessarily informal about metaprobabilistic phenomena like agents having credences in propositions about chances---but also for at least one respectable framework specifically tailored for rigorous discussion of higher-order probabilities. (shrink)

Calibration inductive logics are based on accepting estimates of relative frequencies, which are used to generate imprecise probabilities. In turn, these imprecise probabilities are intended to guide beliefs and decisions — a process called “calibration”. Two prominent examples are Henry E. Kyburg's system of Evidential Probability and Jon Williamson's version of Objective Bayesianism. There are many unexplored questions about these logics. How well do they perform in the short-run? Under what circumstances do they do better or worse? What is their (...) performance relative to traditional Bayesianism? -/- In this article, we develop an agent-based model of a classic binomial decision problem, including players based on variations of Evidential Probability and Objective Bayesianism. We compare the performances of these players, including against a benchmark player who uses standard Bayesian inductive logic. We find that the calibrated players can match the performance of the Bayesian player, but only with particular acceptance thresholds and decision rules. Among other points, our discussion raises some challenges for characterising “cautious” reasoning using imprecise probabilities. Thus, we demonstrate a new way of systematically comparing imprecise probability systems, and we conclude that calibration inductive logics are surprisingly promising for making decisions. (shrink)

A research program is announced, and initial, exciting progress described. Many inference problems, poorly modeled by some traditional approaches, are surprisingly well handled by kinds of simple-minded Bayesian approximations. Fuller Bayesian approaches are typically more accurate but rarely are they either fast or frugal. Open issues include codifying when to use which heuristic and to give detailed evolutionary explanations.

A key challenge in experimental social science research is the incentivisation of subjects such that they take the tasks presented to them seriously and answer honestly. If subject responses can be evaluated against an objective baseline, a standard way of incentivising participants is by rewarding them monetarily as a function of their performance. However, the subject area of experimental philosophy is such that this mode of incentivisation is not applicable as participant responses cannot easily be scored along a true-false spectrum (...) by the experimenters. We claim that experimental philosophers’ neglect of and claims of unimportance about incentivisation mechanisms in their surveys and experiments has plausibly led to poorer data quality and worse conclusions drawn overall, potentially threatening the research programme of experimental philosophy in the long run. As a solution to this, we propose the adoption of the Bayesian Truth Serum, an incentive-compatible mechanism used in economics and marketing, designed for eliciting honest responding in subjective data designs by rewarding participant answers that are surprisingly common. We argue that the Bayesian Truth Serum (i) adequately addresses the issue of incentive compatibility in subjective data research designs and (ii) that it should be applied to the vast majority of research in experimental philosophy. Further, we (iii) provide an empirical application of the method, demonstrating its qualified impact on the distribution of answers on a number of standard experimental philosophy items and outline guidance for researchers aiming to apply this mechanism in future research by specifying the additional costs and design steps involved. (shrink)

Much research has linked surprise to violation of expectations, but it has been less clear how one can be surprised when one has no particular expectation. This paper discusses a computational theory based on Algorithmic Information Theory, which can account for surprises in which one initially expects randomness but then notices a pattern in stimuli. The authors present evidence that a “randomness deficiency” heuristic leads to surprise in such cases.

A key challenge in experimental social science research is the incentivisation of subjects such that they take the tasks presented to them seriously and answer honestly. If subject responses can be evaluated against an objective baseline, a standard way of incentivising participants is by rewarding them monetarily as a function of their performance. However, the subject area of experimental philosophy is such that this mode of incentivisation is not applicable as participant responses cannot easily be scored along a true-false spectrum (...) by the experimenters. We claim that experimental philosophers’ neglect of and claims of unimportance about incentivisation mechanisms in their surveys and experiments has plausibly led to poorer data quality and worse conclusions drawn overall, potentially threatening the research programme of experimental philosophy in the long run. As a solution to this, we propose the adoption of the Bayesian Truth Serum, an incentive-compatible mechanism used in economics and marketing, designed for eliciting honest responding in subjective data designs by rewarding participant answers that are surprisingly common. We argue that the Bayesian Truth Serum (i) adequately addresses the issue of incentive compatibility in subjective data research designs and (ii) that it should be applied to the vast majority of research in experimental philosophy. Further, we (iii) provide an empirical application of the method, demonstrating its qualified impact on the distribution of answers on a number of standard experimental philosophy items and outline guidance for researchers aiming to apply this mechanism in future research by specifying the additional costs and design steps involved. (shrink)

In this paper we dispel the supposed ``admissibility troubles'' for Bayesian accounts of direct inference proposed by Wallmann and Hawthorne, which concern the existence of surprising, unintuitive defeaters even for mundane cases of direct inference. We show that if one follows the majority of authors in the field in using classical probability spaces unimbued with any additional structure, one should expect similar phenomena to arise and should consider them unproblematic in themselves: defeaters abound! We then show that the framework (...) of Higher Probability Spaces allows the natural modelling of the discussed cases which produces no troubles of this kind. (shrink)

Many philosophers regard the imprecise credence framework as a more realistic model of probabilistic inferences with imperfect empirical information than the traditional precise credence framework. Hence, it is surprising that the literature lacks any discussion on how to update one’s imprecise credences when the given evidence itself is imprecise. To fill this gap, I consider two updating principles. Unfortunately, each of them faces a serious problem. The first updating principle, which I call “generalized conditionalization,” sometimes forces an agent to change (...) her imprecise degrees of belief even though she does not have new evidence. The second updating principle, which I call “the generalized dynamic Keynesian model,” may result in a very precise credal state although the agent does not have sufficiently strong evidence to justify such an informative doxastic state. This means that it is much more difficult to come up with an acceptable updating principle for the imprecise credence framework than one might have thought it would be. (shrink)

There is a trade-off between specificity and accuracy in existing models of belief. Descriptions of agents in the tripartite model, which recognizes only three doxastic attitudes—belief, disbelief, and suspension of judgment—are typically accurate, but not sufficiently specific. The orthodox Bayesian model, which requires real-valued credences, is perfectly specific, but often inaccurate: we often lack precise credences. I argue, first, that a popular attempt to fix the Bayesian model by using sets of functions is also inaccurate, since it requires (...) us to have interval-valued credences with perfectly precise endpoints. We can see this problem as analogous to the problem of higher order vagueness. Ultimately, I argue, the only way to avoid these problems is to endorse Insurmountable Unclassifiability. This principle has some surprising and radical consequences. For example, it entails that the trade-off between accuracy and specificity is in-principle unavoidable: sometimes it is simply impossible to characterize an agent’s doxastic state in a way that is both fully accurate and maximally specific. What we can do, however, is improve on both the tripartite and existing Bayesian models. I construct a new model of belief—the minimal model—that allows us to characterize agents with much greater specificity than the tripartite model, and yet which remains, unlike existing Bayesian models, perfectly accurate. (shrink)

We introduce a family of rules for adjusting one's credences in response to learning the credences of others. These rules have a number of desirable features. 1. They yield the posterior credences that would result from updating by standard Bayesian conditionalization on one's peers' reported credences if one's likelihood function takes a particular simple form. 2. In the simplest form, they are symmetric among the agents in the group. 3. They map neatly onto the familiar Condorcet voting results. 4. (...) They preserve shared agreement about independence in a wide range of cases. 5. They commute with conditionalization and with multiple peer updates. Importantly, these rules have a surprising property that we call synergy - peer testimony of credences can provide mutually supporting evidence raising an individual's credence higher than any peer's initial prior report. At first, this may seem to be a strike against them. We argue, however, that synergy is actually a desirable feature and the failure of other updating rules to yield synergy is a strike against them. (shrink)

We argue that social deliberation may increase an agent’s confidence and credence under certain circumstances. An agent considers a proposition H and assigns a probability to it. However, she is not fully confident that she herself is reliable in this assignment. She then endorses H during deliberation with another person, expecting him to raise serious objections. To her surprise, however, the other person does not raise any objections to H. How should her attitudes toward H change? It seems plausible (...) that she should increase the credence she assigns to H and, at the same time, increase the reliability she assigns to herself concerning H. A Bayesian model helps us to investigate under what conditions, if any, this is rational. (shrink)

There is surprising evidence that introspection of our phenomenal states varies greatly between individuals and within the same individual over time. This puts pressure on the notion that introspection gives reliable access to our own phenomenology: introspective unreliability would explain the variability, while assuming that the underlying phenomenology is stable. I appeal to a body of neurocomputational, Bayesian theory and neuroimaging findings to provide an alternative explanation of the evidence: though some limited testing conditions can cause introspection to be (...) unreliable, mostly it is our phenomenology itself that is variable. With this account of phenomenal variability, the occurrence of the surprising evidence can be explained while generally retaining introspective reliability. (shrink)

What is it to believe something might be the case? We develop a puzzle that creates difficulties for standard answers to this question. We go on to propose our own solution, which integrates a Bayesian approach to belief with a dynamic semantics for epistemic modals. After showing how our account solves the puzzle, we explore a surprising consequence: virtually all of our beliefs about what might be the case provide counterexamples to the view that rational belief is closed under (...) logical implication. (shrink)

This paper presents and examines the argument from divine hiddenness as an evidential argument. It argues that a key thought that motivates the argument, namely, that it's surprising that God's existence is not more obvious, does not alone secure the conclusion that divine hiddenness is evidence against God. The evidential problem of divine hiddenness is illustrated using Bayesian models.

It is now well known that, on pain of triviality, the probability of a conditional cannot be identified with the corresponding conditional probability [25]. This surprising impossibility result has a qualitative counterpart. In fact, Peter Gärdenfors showed in [13] that believing ‘If A then B’ cannot be equated with the act of believing B on the supposition that A — as long as supposing obeys minimal Bayesian constraints. Recent work has shown that in spite of these negative results, the (...) question ‘how to accept a conditional?’ has a clear answer. Even if conditionals are not truth-carriers, they do have precise acceptability conditions. Nevertheless most epistemic models of conditionals do not provide acceptance conditions for iterated conditionals. One of the main goals of this essay is to provide a comprehensive account of the notion of epistemic conditionality covering all forms of iteration. First we propose an account of the basic idea of epistemic conditionality, by studying the conditionals validated by epistemic models where iteration is permitted but not constrained by special axioms. Our modeling does not presuppose that epistemic states should be represented by belief sets (we only assume that to each epistemic state corresponds an associated belief state). A full encoding of the basic epistemic conditionals (encompassing all forms of iteration) is presented and a representation result is proved. In the second part of the essay we argue that the notion of change involved in the evaluation of conditionals is suppositional, and that such notion should be distinguished from the notion of updating (modelled by AGM and other methods). We conclude by considering how some of the recent modellings of iterated change fare as methods for iterated supposing. (shrink)

Linguistic predictions may be generated from and evaluated against a representation of events and referents described in the discourse. Compatible with this idea, recent work shows that predictions about novel noun phrases include their definiteness. In the current follow-up study, we ask whether people engage similar prediction-related processes for definite and indefinite referents. This question is relevant for linguistic theories that imply a processing difference between definite and indefinite noun phrases, typically because definiteness is thought to require a uniquely identifiable (...) referent in the discourse. We addressed this question in an event-related potential (ERP) study (N = 48) with preregistration of data acquisition, preprocessing, and Bayesian analysis. Participants read Dutch mini-stories with a definite or indefinite novel noun phrase (e.g., “het/een huis,” the/a house), wherein (in)definiteness of the article was either expected or unexpected and the noun was always strongly expected. Unexpected articles elicited enhanced N400s, but unexpectedly indefinite articles also elicited a positive ERP effect at frontal channels compared to expectedly indefinite articles. We tentatively link this effect to an antiuniqueness violation, which may force people to introduce a new referent over and above the already anticipated one. Interestingly, expectedly definite nouns elicited larger N400s than unexpectedly definite nouns (replicating a previous surprising finding) and indefinite nouns. Although the exact nature of these noun effects remains unknown, expectedly definite nouns may have triggered the strongest semantic activation because they alone refer to specific and concrete referents. In sum, results from both the articles and nouns clearly demonstrate that definiteness marking has a rapid effect on processing, counter to recent claims regarding definiteness processing. (shrink)

This book examines the philosophical conception of abductive reasoning as developed by Charles S. Peirce, the founder of American pragmatism. It explores the historical and systematic connections of Peirce's original ideas and debates about their interpretations. Abduction is understood in a broad sense which covers the discovery and pursuit of hypotheses and inference to the best explanation. The analysis presents fresh insights into this notion of reasoning, which derives from effects to causes or from surprising observations to explanatory theories. The (...) author outlines some logical and AI approaches to abduction as well as studies various kinds of inverse problems in astronomy, physics, medicine, biology, and human sciences to provide examples of retroductions and abductions. The discussion covers also everyday examples with the implication of this notion in detective stories, one of Peirce’s own favorite themes. The author uses Bayesian probabilities to argue that explanatory abduction is a method of confirmation. He uses his own account of truth approximation to reformulate abduction as inference which leads to the truthlikeness of its conclusion. This allows a powerful abductive defense of scientific realism. This up-to-date survey and defense of the Peircean view of abduction may very well help researchers, students, and philosophers better understand the logic of truth-seeking. (shrink)

When does it make sense to act randomly? A persuasive argument from Bayesian decision theory legitimizes randomization essentially only in tie-breaking situations. Rational behaviour in humans, non-human animals, and artificial agents, however, often seems indeterminate, even random. Moreover, rationales for randomized acts have been offered in a number of disciplines, including game theory, experimental design, and machine learning. A common way of accommodating some of these observations is by appeal to a decision-maker’s bounded computational resources. Making this suggestion both (...) precise and compelling is surprisingly difficult. Toward this end, I propose two fundamental rationales for randomization, drawing upon diverse ideas and results from the wider theory of computation. The first unifies common intuitions in favour of randomization from the aforementioned disciplines. The second introduces a deep connection between randomization and memory: access to a randomizing device is provably helpful for an agent burdened with a finite memory. Aside from fit with ordinary intuitions about rational action, the two rationales also make sense of empirical observations in the biological world. Indeed, random behaviour emerges more or less where it should, according to the proposal. (shrink)

The Doomsday argument and anthropic reasoning are two puzzling examples of probabilistic confirmation. In both cases, a lack of knowledge apparently yields surprising conclusions. Since they are formulated within a Bayesian framework, they constitute a challenge to Bayesianism. Several attempts, some successful, have been made to avoid these conclusions, but some versions of these arguments cannot be dissolved within the framework of orthodox Bayesianism. I show that adopting an imprecise framework of probabilistic reasoning allows for a more adequate representation (...) of ignorance in Bayesian reasoning and explains away these puzzles. (shrink)

The paper introduces, compares and contrasts formal models of source reliability proposed in the epistemology literature, in particular the prominent models of Bovens and Hartmann and Olsson :127–143, 2011). All are Bayesian models seeking to provide normative guidance, yet they differ subtly in assumptions and resulting behavior. Models are evaluated both on conceptual grounds and through simulations, and the relationship between models is clarified. The simulations both show surprising similarities and highlight relevant differences between these models. Most importantly, however, (...) our evaluations reveal that important normative concerns arguably remain unresolved. The philosophical implications of this for testimony are discussed. (shrink)

Search in an environment with an uncertain distribution of resources involves a trade-off between exploitation of past discoveries and further exploration. This extends to information foraging, where a knowledge-seeker shifts between reading in depth and studying new domains. To study this decision-making process, we examine the reading choices made by one of the most celebrated scientists of the modern era: Charles Darwin. From the full-text of books listed in his chronologically-organized reading journals, we generate topic models to quantify his local (...) (text-to-text) and global (text-to-past) reading decisions using Kullback-Liebler Divergence, a cognitively-validated, information-theoretic measure of relative surprise. Rather than a pattern of surprise-minimization, corresponding to a pure exploitation strategy, Darwin’s behavior shifts from early exploitation to later exploration, seeking unusually high levels of cognitive surprise relative to previous eras. These shifts, detected by an unsupervised Bayesian model, correlate with major intellectual epochs of his career as identified both by qualitative scholarship and Darwin’s own self-commentary. Our methods allow us to compare his consumption of texts with their publication order. We find Darwin’s consumption more exploratory than the culture’s production, suggesting that underneath gradual societal changes are the explorations of individual synthesis and discovery. Our quantitative methods advance the study of cognitive search through a framework for testing interactions between individual and collective behavior and between short- and long-term consumption choices. This novel application of topic modeling to characterize individual reading complements widespread studies of collective scientific behavior. (shrink)

"The Personal is Political": This was an often-heard slogan of feminist groups in the late sixties and early seventies. The slogan is no doubt open to many interpretations. There is one interpretation which touches on the epistemology of social facts, viz. the slogan claims that in assessing the features of a political system, personal experiences have privileged evidentiary value. For instancte, in the face of third person reports about political corruption, I may remain unmoved in my belief that the political (...) powers are morally upstanding, and it is only when I myself am adversely affected, that I come to change my views. There are two standard patterns of explanation of this type of belief formation: (i) We know that third-person reports may be lessreliable than first-person experiences; (ii) If the third-person reports are no less reliable than first-person experiences, we may just be dealing with a standard pattern of epistemic irrationality. However, we argue that there is also a much more surprising pattern of explanation: under certain conditions, a Bayesian argument can be proffered to the effect that it is rational to change one's beliefs in the face of personal experiences and not in the face of third-person reports, even if these experiences and reports are equally reliable. Hence, the feminist slogan (at least on one particular interpretation of it) receives unexpected support from Bayesian comers. We also show that this pattern of explanation has surprising repercussions on the question of the evidentiary value of miracles in philosophy of religion. (shrink)

With his new riddle of induction, Goodman raised a problem for enumerative induction which many have taken to show that only some ‘natural’ properties can be used for making inductive inferences. Arguably, however, (i) enumerative induction is not a method that scientists use for making inductive inferences in the first place. Moreover, it seems at first sight that (ii) Goodman’s problem does not affect the method that scientists actually use for making such inferences—namely, classical statistics. Taken together, this would indicate (...) that (iii) ‘naturalness’ is not such a relevant concept for the problem of induction, after all. I have no objections against (i) and (iii). But, contrary to (ii), I argue that classical statistics does face a version of Goodman’s problem, which I call the statistical riddle of induction. Given that his problem merely is an instance of the more general problem of underdetermination, this is hardly surprising. Perhaps more importantly, I also show how my riddle can be used for exposing a common fallacy in probabilistic reasoning, without relying on Bayesian assumptions. (shrink)

Supporters of the autonomy of higher-level causation (or explanation) often appeal to proportionality, arguing that higher-level causes are more proportional than their lower-level realizers. Recently, measures based on information theory and causal modeling have been proposed that allow one to shed new light on proportionality and the related notion of specificity. In this paper we apply ideas from this literature to the issue of higher vs. lower-level causation (and explanation). Surprisingly, proportionality turns out to be irrelevant for the question of (...) whether higher-level causes (or explanations) can be autonomous; specificity is a much more informative notion for this purpose. (shrink)

Recent work in social epistemology has shown that, in certain situations, less communication leads to better outcomes for epistemic groups. In this paper, we show that, ceteris paribus, a Bayesian agent may believe less strongly that a single agent is biased than that an entire group of independent agents is biased. We explain this initially surprising result and show that it is in fact a consequence one may conceive on the basis of commonsense reasoning.