Remarkable progress in the mathematics and computer science of probability has led to a revolution in the scope of probabilistic models. In particular, ‘sophisticated’ probabilistic methods apply to structured relational systems such as graphs and grammars, of immediate relevance to the cognitive sciences. This Special Issue outlines progress in this rapidly developing field, which provides a potentially unifying perspective across a wide range of domains and levels of explanation. Here, we introduce the historical and conceptual foundations of the approach, explore (...) how the approach relates to studies of explicit probabilistic reasoning, and give a brief overview of the field as it stands today. (shrink)

Human cognition requires coping with a complex and uncertain world. This suggests that dealing with uncertainty may be the central challenge for human reasoning. In Bayesian Rationality we argue that probability theory, the calculus of uncertainty, is the right framework in which to understand everyday reasoning. We also argue that probability theory explains behavior, even on experimental tasks that have been designed to probe people's logical reasoning abilities. Most commentators agree on the centrality of uncertainty; some suggest that there is (...) a residual role for logic in understanding reasoning; and others put forward alternative formalisms for uncertain reasoning, or raise specific technical, methodological, or empirical challenges. In responding to these points, we aim to clarify the scope and limits of probability and logic in cognitive science; explore the meaning of the explanation of cognition; and re-evaluate the empirical case for Bayesian rationality. (shrink)

Why should moral philosophers, moral psychologists, and machine ethicists care about computational complexity? Debates on whether artificial intelligence (AI) can or should be used to solve problems in ethical domains have mainly been driven by what AI can or cannot do in terms of human capacities. In this paper, we tackle the problem from the other end by exploring what kind of moral machines are possible based on what computational systems can or cannot do. To do so, we analyze normative (...) ethics through the lens of computational complexity. First, we introduce computational complexity for the uninitiated reader and discuss how the complexity of ethical problems can be framed within Marr’s three levels of analysis. We then study a range of ethical problems based on consequentialism, deontology, and virtue ethics, with the aim of elucidating the complexity associated with the problems themselves (e.g., due to combinatorics, uncertainty, strategic dynamics), the computational methods employed (e.g., probability, logic, learning), and the available resources (e.g., time, knowledge, learning). The results indicate that most problems the normative frameworks pose lead to tractability issues in every category analyzed. Our investigation also provides several insights about the computational nature of normative ethics, including the differences between rule- and outcome-based moral strategies, and the implementation-variance with regard to moral resources. We then discuss the consequences complexity results have for the prospect of moral machines in virtue of the trade-off between optimality and efficiency. Finally, we elucidate how computational complexity can be used to inform both philosophical and cognitive-psychological research on human morality by advancing the moral tractability thesis. (shrink)

There are two distinct approaches to Bayesian modelling in cognitive science. Black-box approaches use Bayesian theory to model the relationship between the inputs and outputs of a cognitive system without reference to the mediating causal processes; while mechanistic approaches make claims about the neural mechanisms which generate the outputs from the inputs. This paper concerns the relationship between these two approaches. We argue that the dominant trend in the philosophical literature, which characterizes the relationship between black-box and mechanistic approaches to (...) Bayesian cognitive science in terms of the dichotomy between instrumentalism and realism, is misguided. We propose that the two distinctions are orthogonal: black-box and mechanistic approaches to Bayesian modelling can each be given either an instrumentalist or a realist interpretation. We argue that the current tendency to conflate black-box approaches with instrumentalism and mechanistic approaches with realism stems from unwarranted assumptions about the nature of scientific explanation, the ontological commitments of scientific theories, and the role of abstraction and idealization in scientific models. We challenge each of these assumptions to reframe the debates over Bayesian modelling in cognitive science. (shrink)

For most of its history, the psychology of reasoning was dominated by binary extensional logic. The so-called “new paradigm” instead puts subjective degrees of belief center stage, often represented as probabilities. We argue that the “new paradigm” is too vaguely defined and therefore does not allow a clear decision about what falls within its scope and what does not. We also show that there was not one settled theoretical “old” paradigm, before the new developments emerged, and that the alleged new (...) paradigm is less revolutionary as the term suggests. A more veridical view is that current progress is developing in continuities where rival research programs can thrive or fail in the face of new experimental findings. The article closes with some topics where more connections between competing research programs are likely to promote progress in our understanding of human reasoning. (shrink)

Tenenbaum and Griffiths (2001) have proposed that their Bayesian model of generalisation unifies Shepard’s (1987) and Tversky’s (1977) similarity-based explanations of two distinct patterns of generalisation behaviours by reconciling them under a single coherent task analysis. I argue that this proposal needs refinement: instead of unifying the heterogeneous notion of psychological similarity, the Bayesian approach unifies generalisation by rendering the distinct patterns of behaviours informationally relevant. I suggest that generalisation as a Bayesian inference should be seen as a complement to, (...) instead of a replacement of, similarity-based explanations. Furthermore, I show that the unificatory powers of the Bayesian model of generalisation can contribute to the selection of one of these models of psychological similarity. (shrink)

Several philosophers and psychologists have characterized belief in conspiracy theories as a product of irrational reasoning. Proponents of conspiracy theories apparently resist revising their beliefs given disconfirming evidence and tend to believe in more than one conspiracy, even when the relevant beliefs are mutually inconsistent. In this paper, we bring leading views on conspiracy theoretic beliefs closer together by exploring their rationality under a probabilistic framework. We question the claim that the irrationality of conspiracy theoretic beliefs stems from an inadequate (...) response to disconfirming evidence and internal incoherence. Drawing analogies to Lakatosian research programs, we argue that maintaining a core conspiracy belief can be Bayes-rational when it is embedded in a network of auxiliary beliefs, which can be revised to protect the more central belief from disconfirmation. We propose that the irrationality associated with conspiracy belief lies not in a flawed updating method, but in a failure to converge toward wellconfirmed, stable belief networks in the long run. This approach not only reconciles previously disjointed views, but also points toward more specific descriptions of why agents may be prone to adopting beliefs in conspiracy theories. (shrink)

This paper attempts to answer a central question in unsupervised learning: what does it mean to “make sense” of a sensory sequence? In our formalization, making sense involves constructing a symbolic causal theory that both explains the sensory sequence and also satisfies a set of unity conditions. The unity conditions insist that the constituents of the causal theory – objects, properties, and laws – must be integrated into a coherent whole. On our account, making sense of sensory input is a (...) type of program synthesis, but it is unsupervised program synthesis. Our second contribution is a computer implementation, the Apperception Engine, that was designed to satisfy the above requirements. Our system is able to produce interpretable human-readable causal theories from very small amounts of data, because of the strong inductive bias provided by the unity conditions. A causal theory produced by our system is able to predict future sensor readings, as well as retrodict earlier readings, and impute (fill in the blanks of) missing sensory readings, in any combination. In fact, it is able to do all three tasks simultaneously. We tested the engine in a diverse variety of domains, including cellular automata, rhythms and simple nursery tunes, multi-modal binding problems, occlusion tasks, and sequence induction intelligence tests. In each domain, we test our engine's ability to predict future sensor values, retrodict earlier sensor values, and impute missing sensory data. The Apperception Engine performs well in all these domains, significantly out-performing neural net baselines. We note in particular that in the sequence induction intelligence tests, our system achieved human-level performance. This is notable because our system is not a bespoke system designed specifically to solve intelligence tests, but a general-purpose system that was designed to make sense of any sensory sequence. (shrink)

Huttenlocher et al. (J Exp Psychol Gen 129:220–241, 2000) introduce the category adjustment model (CAM). Given that participants imperfectly remember stimuli (which we refer to as “targets”), CAM holds that participants maximize accuracy by using information about the distribution of the targets to improve their judgments. CAM predicts that judgments will be a weighted average of the imperfect memory of the target and the mean of the distribution of targets. Huttenlocher et al. (2000) report on three experiments and conclude that (...) CAM is “verified”. We attempt to replicate the conditions in Experiment 3 from Huttenlocher et al. (2000). We analyze judgment-level data rather than averaged data. We find evidence of a bias toward a set of recent targets rather than a bias toward the running mean. We do not find evidence of learning. The judgments in our dataset are not consistent with CAM. We discuss other defects in HHV—including dividing by zero. It seems that evidence for CAM is a statistical artifact that appears when researchers analyze data averaged across trials and do not consider a recency bias. (shrink)

A probabilistic Chomsky–Schützenberger hierarchy of grammars is introduced and studied, with the aim of understanding the expressive power of generative models. We offer characterizations of the distributions definable at each level of the hierarchy, including probabilistic regular, context-free, (linear) indexed, context-sensitive, and unrestricted grammars, each corresponding to familiar probabilistic machine classes. Special attention is given to distributions on (unary notations for) positive integers. Unlike in the classical case where the "semi-linear" languages all collapse into the regular languages, using analytic tools (...) adapted from the classical setting we show there is no collapse in the probabilistic hierarchy: more distributions become definable at each level. We also address related issues such as closure under probabilistic conditioning. (shrink)

According to manipulationist accounts of causal explanation, to explain an event is to show how it could be changed by intervening on its cause. The relevant change must be a ‘serious possibility’ claims Woodward 2003, distinct from mere logical or physical possibility—approximating something I call ‘scientific possibility’. This idea creates significant difficulties: background knowledge is necessary for judgments of possibility. Yet the primary vehicles of explanation in manipulationism are ‘invariant’ generalisations, and these are not well adapted to encoding such knowledge, (...) especially in the social sciences, as some of it is non-causal. Ceteris paribus (CP) laws or generalisations labour under no such difficulty. A survey of research methods such as case and comparative studies, randomised control trials, ethnography, and structural equation modeling, suggests that it would be more difficult and in some instances impossible to try to represent the output of each method in invariant generalisations; and that this is because in each method causal and non-causal background knowledge mesh in a way that cannot easily be accounted for in manipulationist terms. Ceteris paribus-generalisations being superior in this regard, a theory of explanation based on the latter is a better fit for social science. (shrink)

This article focuses on philosophical views that attempt to explain widespread belief in indeterministic choice by following a strategy that harkens back at least to Spinoza. According to this Spinozan strategy, people draw an inference from the absence of experiences of determined choice to the belief in indeterministic choice. Accounts of this kind are historically liable to overgeneralization. The pair of accounts defended in Shaun Nichols’ recent book, Bound: Essays on Free Will and Responsibility, are the most complete and empirically (...) plausible representatives of this strategy. I argue that both accounts ultimately fail, and overgeneralization worries remain for Spinozan strategists. (shrink)

We investigated people's ability to infer others’ mental states from their emotional reactions, manipulating whether agents wanted, expected, and caused an outcome. Participants recovered agents’ desires throughout. When the agent observed, but did not cause the outcome, participants’ ability to recover the agent's beliefs depended on the evidence they got. When the agent caused the event, participants’ judgments also depended on the probability of the action ; when actions were improbable given the mental states, people failed to recover the agent's (...) beliefs even when they saw her react to both the anticipated and actual outcomes. A Bayesian model captured human performance throughout, consistent with the proposal that people rationally integrate information about others’ actions and emotional reactions to infer their unobservable mental states. (shrink)

People’s beliefs about normality play an important role in many aspects of cognition and life (e.g., causal cognition, linguistic semantics, cooperative behavior). But how do people determine what sorts of things are normal in the first place? Past research has studied both people’s representations of statistical norms (e.g., the average) and their representations of prescriptive norms (e.g., the ideal). Four studies suggest that people’s notion of normality incorporates both of these types of norms. In particular, people’s representations of what is (...) normal were found to be influenced both by what they believed to be descriptively average and by what they believed to be prescriptively ideal. This is shown across three domains: people’s use of the word ‘‘normal” (Study 1), their use of gradable adjectives (Study 2), and their judgments of concept prototypicality (Study 3). A final study investigated the learning of normality for a novel category, showing that people actively combine statistical and prescriptive information they have learned into an undifferentiated notion of what is normal (Study 4). Taken together, these findings may help to explain how moral norms impact the acquisition of normality and, conversely, how normality impacts the acquisition of moral norms. (shrink)

Causal reasoning is one of our most central cognitive competencies, enabling us to adapt to our world. Causal knowledge allows us to predict future events, or diagnose the causes of observed facts. We plan actions and solve problems using knowledge about cause-effect relations. Without our ability to discover and empirically test causal theories, we would not have made progress in various empirical sciences. In the past decades, the important role of causal knowledge has been discovered in many areas of cognitive (...) psychology. Despite the ubiquity of causal reasoning, textbooks of cognitive psychology have neglected this growing field. The goal of The Oxford Handbook of Causal Reasoning is to fill this gap. The handbook brings together the leading researchers in the field of causal reasoning and offers state-of-the-art presentations of theories and research. It provides introductions of competing theories of causal reasoning, and discusses its role in various cognitive functions and domains. The final section presents research from neighboring fields. 1. Causal Reasoning: An Introduction Michael R. Waldmann Part I: Theories of Causal Cognition 2. Associative Accounts of Causal Cognition Mike E. Le Pelley, Oren Griffiths, and Tom Beesley 3. Rules of Causal Judgment: Mapping Statistical Information Onto Causal Beliefs José C. Perales, Andrés Catena, Antonio Cándido, and Antonio Maldonado 4. The Inferential Reasoning Theory of Causal Learning: Toward a Multi- Process Propositional Account Yannick Boddez, Jan De Houwer, and Tom Beckers 5. Causal Invariance as an Essential Constraint for Creating a Causal Representation of the World: Generalizing the Invariance of Causal Power Patricia W. Cheng and Hongjing Lu 6. The Acquisition and Use of Causal Structure Knowledge Benjamin Margolin Rottman 7. Formalizing Prior Knowledge in Causal Induction Thomas L. Griffiths 8. Causal Mechanisms Samuel G. B. Johnson and Woo-kyoung Ahn 9. Force Dynamics and Causation Phillip Wolff and Robert Thorstad 10. Mental Models and Causation P. N. Johnson- Laird and Sangeet S. Khemlani 11. Pseudocontingencies Klaus Fiedler and Florian Kutzner 12. Singular Causation David Danks 13. Cognitive Neuroscience of Causal Reasoning Joachim T. Operskalski and Aron K. Barbey Part II: Basic Cognitive Functions 14. Visual Impressions of Causality Peter White 15. Goal-Directed Actions Bernhard Hommel 16. Planning and Control Magda Osman 17. Reinforcement Learning and Causal Models Samuel J. Gershman 18. Causation and the Probability of Causal Conditionals David E. Over 19. Causal Models and Conditional Reasoning Mike Oaksford and Nick Chater 20. Concepts as Causal Models: Categorization Bob Rehder 21. Concepts as Causal Models: Induction Bob Rehder 22. Causal Explanation Tania Lombrozo and Nadya Vasilyeva 23. Diagnostic Reasoning Björn Meder and Ralf Mayrhofer 24. Inferring Causal Relations by Analogy Keith J. Holyoak and Hee-Seung Lee 25. Causal Argument Ulrike Hahn, Roland Bluhm, and Frank Zenker 26. Causality in Decision- Making York Hagmayer and Philip M. Fernbach Part III: Domains of Causal Reasoning 27. Intuitive Theories Tobias Gerstenberg and Joshua B. Tenenbaum 28. Space, Time, and Causality Marc J. Buehner 29. Causation in Legal and Moral Reasoning David A. Lagnado and Tobias Gerstenberg 30. The Role of Causal Knowledge in Reasoning About Mental Disorders Woo-kyoung Ahn, Nancy S. Kim, and Matthew S. Lebowitz 31. Causality and Causal Reasoning in Natural Language Torgrim Solstad and Oliver Bott 32. Social Attribution and Explanation Denis Hilton Part IV: Development, Phylogeny, and Culture 33. The Development of Causal Reasoning Paul Muentener and Elizabeth Bonawitz 34. Causal Reasoning in Non-Human Animals Christian Schloegl and Julia Fischer 35. Causal Cognition and Culture Andrea Bender, Sieghard Beller, and Douglas L. Medin. (shrink)

A great deal of research has demonstrated that learning is influenced by the learner’s prior background knowledge (e.g. Murphy, 2002; Keil, 1990), but little is known about the processes by which prior knowledge is deployed. We explore the role of explanation in deploying prior knowledge by examining the joint effects of eliciting explanations and providing prior knowledge in a task where each should aid learning. Three hypotheses are considered: that explanation and prior knowledge have independent and additive effects on learning, (...) that their joint effects on learning are subadditive, and that their effects are superadditive. A category learning experiment finds evidence for a superadditive effect: explaining drives the discovery of regularities, while prior knowledge constrains which regularities learners discover. This is consistent with an account of explanation’s effects on learning proposed in Williams & Lombrozo (in press). (shrink)

In this paper we discuss the new Tweety puzzle. The original Tweety puzzle was addressed by approaches in non-monotonic logic, which aim to adequately represent the Tweety case, namely that Tweety is a penguin and, thus, an exceptional bird, which cannot fly, although in general birds can fly. The new Tweety puzzle is intended as a challenge for probabilistic theories of epistemic states. In the first part of the paper we argue against monistic Bayesians, who assume that epistemic states can (...) at any given time be adequately described by a single subjective probability function. We show that monistic Bayesians cannot provide an adequate solution to the new Tweety puzzle, because this requires one to refer to a frequency-based probability function. We conclude that monistic Bayesianism cannot be a fully adequate theory of epistemic states. In the second part we describe an empirical study, which provides support for the thesis that monistic Bayesianism is also inadequate as a descriptive theory of cognitive states. In the final part of the paper we criticize Bayesian approaches in cognitive science, insofar as their monistic tendency cannot adequately address the new Tweety puzzle. We, further, argue against monistic Bayesianism in cognitive science by means of a case study. In this case study we show that Oaksford and Chater’s (2007, 2008) model of conditional inference—contrary to the authors’ theoretical position—has to refer also to a frequency-based probability function. (shrink)

In our target article, we argued that there is a profound functional discontinuity between the cognitive abilities of modern humans and those of all other extant species. Unsurprisingly, our hypothesis elicited a wide range of responses from commentators. After responding to the commentaries, we conclude that our hypothesis lies closer to Darwin's views on the matter than to those of many of our contemporaries.

Several philosophers and psychologists have characterized belief in conspiracy theories as a product of irrational reasoning. Proponents of conspiracy theories apparently resist revising their beliefs given disconfirming evidence and tend to believe in more than one conspiracy, even when the relevant beliefs are mutually inconsistent. In this paper, we bring leading views on conspiracy theoretic beliefs closer together by exploring their rationality under a probabilistic framework. We question the claim that the irrationality of conspiracy theoretic beliefs stems from an inadequate (...) response to disconfirming evidence and internal incoherence. Drawing analogies to Lakatosian research programs, we argue that maintaining a core conspiracy belief can be Bayes-rational when it is embedded in a network of auxiliary beliefs, which can be revised to protect the more central belief from disconfirmation. We propose that the (ir)rationality associated with conspiracy belief lies not in a flawed updating method, but in their failure to converge toward well-confirmed, stable belief networks in the long run. This approach not only reconciles previously disjointed views, but also points toward more specific hypotheses explaining why some agents may be prone to adopting beliefs in conspiracy theories. (shrink)

Stereotypes are often described as being generally inaccurate and irrational. However, for years, a minority of social psychologists has been proclaiming that stereotype accuracy is among the most robust findings in the field. This same minority also opposes the majority by questioning the power of self-fulfilling prophecies and thereby the construction of social reality. The present paper examines this long-standing debate from the perspective of predictive processing, an increasingly influential cognitive science theory. In this theory, stereotype accuracy and self-fulfilling prophecies (...) are two sides of the same coin, namely prediction error minimisation, pointing to a new middle course between the two existing views. On the one hand, predictive processing indicates that depicting stereotypes as generally inaccurate runs counter to their actual purpose of making the social world predictable, which supports the minority view. On the other hand, predictive processing supposes that expectations, including stereotypes, permanently affect perception and behaviour and thereby co-construct social reality, which supports the majority view. Therefore, from a predictive processing perspective, stereotypes are largely rational and not per se inaccurate, and self-fulfilling prophecies are omnipresent and greatly affect social reality. This new middle course appears to fit the empirical data better than the two existing views. (shrink)

There has been much criticism of the idea that Friston's free-energy principle can unite the life and mind sciences. Here, we argue that perhaps the greatest problem for the totalizing ambitions of its proponents is a failure to recognize the importance of evolutionary dynamics and to provide a convincing adaptive story relating free-energy minimization to organismal fitness.

Topics in Cognitive Science, Volume 14, Issue 3, Page 574-601, July 2022.

Theories of relations between language and conceptual development benefit from empirical evidence for concepts available in infancy, but such evidence is comparatively scarce. Here, we examine early representations of specific concepts, namely, sets of dynamic events corresponding either to predicates involving two variables with a reversible, asymmetric relation between them (such as the set of all events that correspond to a linguistic phrase like “a dog is pushing a car,”) or to comparatively simpler, one-variable predicates (such as the set of (...) events corresponding to a phrase like “a dog is jumping.”). We develop a non-linguistic, anticipatory eye-tracking task that can be administered to both infants and adults, and we use this task to gather evidence for the formation and use of such one-and two-place-predicate classes (which we refer to as event sortals) in 12–24-mo-old infants, and in adults with and without concurrent verbal prose shadowing. Using visually similar stimuli for both the simpler (one-place) and the more complex (reversible, asymmetric, two-place) concepts, we find that infants only show evidence for forming and generalizing one-place event sortals, and, while adults succeed with both kinds in the absence of verbal shadowing, shadowing hampers their ability to form and use the asymmetric two-place event sortals. In a subsequent experiment with adults, we find that if the shadowing material is grammatically impoverished, adults now succeed in forming and using both one- and two-place event sortals. We discuss implications of these results for theories of concept acquisition, and the role of language in this process. (shrink)

In Jackendoff's Parallel Architecture, the well-formed expressions of a language are licensed by correspondences between phonology, syntax, and conceptual structure. I show how this architecture can be used to make sense of the existence of parasitic gap constructions. A parasitic gap is one that is rendered acceptable because of the presence of another gap in the same sentence. Compare *a person who everyone who talks to likes Chris, which shows an illicit extraction from a relative clause, and a person everyone (...) who talks to likes, which shows a parasitic gap in the relative clause. Languages differ in terms of the range of configurations in which they allow parasitic gaps; these configurations appear to form a hierarchy.These observations raise some fundamental questions: Why do parasitic gaps exist at all?Why are different syntactic configurations possible for P-gaps and why just these configurations?Why is there a parasitic gap hierarchy?The answers that I propose make crucial use of constructional overgeneralization of across-the-board extraction in coordinate constructions, formulated straightforwardly in the framework of Jackendoff's Parallel Architecture. (shrink)

Human and automated tutors attempt to choose pedagogical activities that will maximize student learning, informed by their estimates of the student's current knowledge. There has been substantial research on tracking and modeling student learning, but significantly less attention on how to plan teaching actions and how the assumed student model impacts the resulting plans. We frame the problem of optimally selecting teaching actions using a decision-theoretic approach and show how to formulate teaching as a partially observable Markov decision process planning (...) problem. This framework makes it possible to explore how different assumptions about student learning and behavior should affect the selection of teaching actions. We consider how to apply this framework to concept learning problems, and we present approximate methods for finding optimal teaching actions, given the large state and action spaces that arise in teaching. Through simulations and behavioral experiments, we explore the consequences of choosing teacher actions under different assumed student models. In two concept-learning tasks, we show that this technique can accelerate learning relative to baseline performance. (shrink)

Lakoff (2016) describes an account of conceptual representation based in part on metaphor. Though promising, this account faces several challenges with respect to learning and development.

This paper introduces a special issue of Cognitive Science initiated on the 25th anniversary of the publication of Parallel Distributed Processing (PDP), a two-volume work that introduced the use of neural network models as vehicles for understanding cognition. The collection surveys the core commitments of the PDP framework, the key issues the framework has addressed, and the debates the framework has spawned, and presents viewpoints on the current status of these issues. The articles focus on both historical roots and contemporary (...) developments in learning, optimality theory, perception, memory, language, conceptual knowledge, cognitive control, and consciousness. Here we consider the approach more generally, reviewing the original motivations, the resulting framework, and the central tenets of the underlying theory. We then evaluate the impact of PDP both on the field at large and within specific subdomains of cognitive science and consider the current role of PDP models within the broader landscape of contemporary theoretical frameworks in cognitive science. Looking to the future, we consider the implications for cognitive science of the recent success of machine learning systems called “deep networks”—systems that build on key ideas presented in the PDP volumes. (shrink)

Recent work in rational probabilistic modeling suggests that a kind of propositional reasoning is ubiquitous in cognition and especially in cognitive development. However, there is no reason to believe that this type of computation is necessarily conscious or resource-intensive.

Bayesian Rationality (Oaksford & Chater 2007) illustrates the strengths of Bayesian models of cognition: the systematicity of rational explanations, transparent assumptions about human learners, and combining structured symbolic representation with statistics. However, the book also highlights some of the challenges this approach faces: of providing psychological mechanisms, explaining the origins of the knowledge that guides human learning, and accounting for how people make genuinely new discoveries.

The commentaries reflect three core themes that pertain not just to our theory, but to the enterprise of connectionist modeling more generally. The first concerns the relationship between a cognitive theory and an implemented computer model. Specifically, how does one determine, when a model departs from the theory it exemplifies, whether the departure is a useful simplification or a critical flaw? We argue that the answer to this question depends partially upon the model's intended function, and we suggest that connectionist (...) models have important functions beyond the commonly accepted goals of fitting data and making predictions. The second theme concerns perceived in-principle limitations of the connectionist approach to cognition, and the specific concerns these perceived limitations raise for our theory. We argue that the approach is not in fact limited in the ways our critics suggest. One common misconception, that connectionist models cannot address abstract or relational structure, is corrected through new simulations showing directly that such structure can be captured. The third theme concerns the relationship between parallel distributed processing (PDP) models and structured probabilistic approaches. In this case we argue that there the difference between the approaches is not merely one of levels. Our PDP approach differs from structured statistical approaches at all of Marr's levels, including the characterization of the goals of cognitive computations, and of the representations and algorithms used. (shrink)

This article argues that understanding human uniqueness requires recognizing that we are a cultural species whose evolution has been driven by the interaction among genes and culture for over a million years. Here, I review the basic argument, incorporate recent findings, and highlight ongoing efforts to apply this approach to more deeply understand both the universal aspects of our cognition as well as the variation across societies. This article will cover (1) the origins and evolution of our capacities for culture, (...) (2) examples of specialized mental abilities such as those related to mechanical causality, prestige, and cooperation, and (3) recent efforts to apprehend our capacities for abstraction. I close by discussing common canards that generate confusion in mapping what makes us human. (shrink)

Over-flexibility in the definition of Friston blankets obscures a key distinction between observational and interventional inference. The latter requires cognizers form not just a causal representation of the world but also of their own boundary and relationship with it, in order to diagnose the consequences of their actions. We suggest this locates the blanket in the eye of the beholder.

The past 15 years occasioned an extraordinary blossoming of research into the cognitive and affective mechanisms that support moral judgment and behavior. This growth in our understanding of moral mechanisms overshadowed a crucial and complementary question, however: How are they learned? As this special issue of the journal Cognition attests, a new crop of research into moral learning has now firmly taken root. This new literature draws on recent advances in formal methods developed in other domains, such as Bayesian inference, (...) reinforcement learning and other machine learning techniques. Meanwhile, it also demonstrates how learning and deciding in a social domain—and especially in the moral domain—sometimes involves specialized cognitive systems. We review the contributions to this special issue and situate them within the broader contemporary literature. Our review focuses on how we learn moral values and moral rules, how we learn about personal moral character and relationships, and the philosophical implications of these emerging models. (shrink)

The prominence of Bayesian modeling of cognition has increased recently largely because of mathematical advances in specifying and deriving predictions from complex probabilistic models. Much of this research aims to demonstrate that cognitive behavior can be explained from rational principles alone, without recourse to psychological or neurological processes and representations. We note commonalities between this rational approach and other movements in psychology – namely, Behaviorism and evolutionary psychology – that set aside mechanistic explanations or make use of optimality assumptions. Through (...) these comparisons, we identify a number of challenges that limit the rational program's potential contribution to psychological theory. Specifically, rational Bayesian models are significantly unconstrained, both because they are uninformed by a wide range of process-level data and because their assumptions about the environment are generally not grounded in empirical measurement. The psychological implications of most Bayesian models are also unclear. Bayesian inference itself is conceptually trivial, but strong assumptions are often embedded in the hypothesis sets and the approximation algorithms used to derive model predictions, without a clear delineation between psychological commitments and implementational details. Comparing multiple Bayesian models of the same task is rare, as is the realization that many Bayesian models recapitulate existing (mechanistic level) theories. Despite the expressive power of current Bayesian models, we argue they must be developed in conjunction with mechanistic considerations to offer substantive explanations of cognition. We lay out several means for such an integration, which take into account the representations on which Bayesian inference operates, as well as the algorithms and heuristics that carry it out. We argue this unification will better facilitate lasting contributions to psychological theory, avoiding the pitfalls that have plagued previous theoretical movements. (shrink)

A theory of conceptual development must specify the innate representational primitives, must characterize the ways in which the initial state differs from the adult state, and must characterize the processes through which one is transformed into the other. The Origin of Concepts (henceforth TOOC) defends three theses. With respect to the initial state, the innate stock of primitives is not limited to sensory, perceptual, or sensorimotor representations; rather, there are also innate conceptual representations. With respect to developmental change, conceptual development (...) consists of episodes of qualitative change, resulting in systems of representation that are more powerful than, and sometimes incommensurable with, those from which they are built. With respect to a learning mechanism that achieves conceptual discontinuity, I offer Quinian bootstrapping. TOOC concludes with a discussion of how an understanding of conceptual development constrains a theory of concepts. (shrink)

Most empirical work in human categorization has studied learning in either fully supervised or fully unsupervised scenarios. Most real-world learning scenarios, however, are semi-supervised: Learners receive a great deal of unlabeled information from the world, coupled with occasional experiences in which items are directly labeled by a knowledgeable source. A large body of work in machine learning has investigated how learning can exploit both labeled and unlabeled data provided to a learner. Using equivalences between models found in human categorization and (...) machine learning research, we explain how these semi-supervised techniques can be applied to human learning. A series of experiments are described which show that semi-supervised learning models prove useful for explaining human behavior when exposed to both labeled and unlabeled data. We then discuss some machine learning models that do not have familiar human categorization counterparts. Finally, we discuss some challenges yet to be addressed in the use of semi-supervised models for modeling human categorization. (shrink)

Learning to understand a single causal system can be an achievement, but humans must learn about multiple causal systems over the course of a lifetime. We present a hierarchical Bayesian framework that helps to explain how learning about several causal systems can accelerate learning about systems that are subsequently encountered. Given experience with a set of objects, our framework learns a causal model for each object and a causal schema that captures commonalities among these causal models. The schema organizes the (...) objects into categories and specifies the causal powers and characteristic features of these categories and the characteristic causal interactions between categories. A schema of this kind allows causal models for subsequent objects to be rapidly learned, and we explore this accelerated learning in four experiments. Our results confirm that humans learn rapidly about the causal powers of novel objects, and we show that our framework accounts better for our data than alternative models of causal learning. (shrink)