Deep neural networks (DNNs) have had extraordinary successes in classifying photographic images of objects and are often described as the best models of biological vision. This conclusion is largely based on three sets of findings: (1) DNNs are more accurate than any other model in classifying images taken from various datasets, (2) DNNs do the best job in predicting the pattern of human errors in classifying objects taken from various behavioral datasets, and (3) DNNs do the best job in predicting (...) brain signals in response to images taken from various brain datasets (e.g., single cell responses or fMRI data). However, these behavioral and brain datasets do not test hypotheses regarding what features are contributing to good predictions and we show that the predictions may be mediated by DNNs that share little overlap with biological vision. More problematically, we show that DNNs account for almost no results from psychological research. This contradicts the common claim that DNNs are good, let alone the best, models of human object recognition. We argue that theorists interested in developing biologically plausible models of human vision need to direct their attention to explaining psychological findings. More generally, theorists need to build models that explain the results of experiments that manipulate independent variables designed to test hypotheses rather than compete on making the best predictions. We conclude by briefly summarizing various promising modeling approaches that focus on psychological data. (shrink)

People are habitual explanation generators. At its most mundane, our propensity to explain allows us to infer that we should not drink milk that smells sour; at the other extreme, it allows us to establish facts (e.g., theorems in mathematical logic) whose truth was not even known prior to the existence of the explanation (proof). What do the cognitive operations underlying the inference that the milk is sour have in common with the proof that, say, the square root of two (...) is irrational? Our ability to generate explanations bears striking similarities to our ability to make analogies. Both reflect a capacity to generate inferences and generalizations that go beyond the featural similarities between a novel problem and familiar problems in terms of which the novel problem may be understood. However, a notable difference between analogy-making and explanation-generation is that the former is a process in which a single source situation is used to reason about a single target, whereas the latter often requires the reasoner to integrate multiple sources of knowledge. This seemingly small difference poses a challenge to the task of marshaling our understanding of analogical reasoning to understanding explanation. We describe a model of explanation, derived from a model of analogy, adapted to permit systematic violations of this one-to-one mapping constraint. Simulation results demonstrate that the resulting model can generate explanations for novel explananda and that, like the explanations generated by human reasoners, these explanations vary in their coherence. (shrink)

Ramscar and colleagues (2010, this volume) describe the “feature-label-order” (FLO) effect on category learning and characterize it as a constraint on symbolic learning. I argue that FLO is neither a constraint on symbolic learning in the sense of “learning elements of a symbol system” (instead, it is an effect on nonsymbolic, association learning) nor is it, more than any other constraint on category learning, a constraint on symbolic learning in the sense of “solving the symbol grounding problem.”.

On several key issues we agree with the commentators. Perhaps most importantly, everyone seems to agree that psychology has an important role to play in building better models of human vision, and (most) everyone agrees (including us) that deep neural networks (DNNs) will play an important role in modelling human vision going forward. But there are also disagreements about what models are for, how DNN–human correspondences should be evaluated, the value of alternative modelling approaches, and impact of marketing hype in (...) the literature. In our view, these latter issues are contributing to many unjustified claims regarding DNN–human correspondences in vision and other domains of cognition. We explore all these issues in this response. (shrink)

Schyns, Goldstone & Thibaut argue that categorization experience results in the learning of new perceptual features that are not derivable from the learner's existing feature set. We explore the meaning and implications of this “nonderivability” claim and relate it to the question of whether perceptual invariants are learnable, and if so, what might be entailed in learning them.

Page argues convincingly for several important properties of localist representations in connectionist models of cognition. I argue that another important property of localist representations is that they serve as the starting point for connectionist representations of symbolic (relational) structures because they express meaningful properties independent of one another and their relations.

Theories of relational concept acquisition based on structured intersection discovery predict that relational concepts with a probabilistic structure ought to be extremely difficult to learn. We report four experiments testing this prediction by investigating conditions hypothesized to facilitate the learning of such categories. Experiment 1 showed that changing the task from a category-learning task to choosing the “winning” object in each stimulus greatly facilitated participants' ability to learn probabilistic relational categories. Experiments 2 and 3 further investigated the mechanisms underlying this (...) “who's winning” effect. Experiment 4 replicated and generalized the “who's winning” effect with more natural stimuli. Together, our findings suggest that people learn relational concepts by a process of intersection discovery akin to schema induction, and that any task that encourages people to discover a higher order relation that remains invariant over members of a category will facilitate the learning of putatively probabilistic relational concepts. (shrink)

Humans, including preschool children, exhibit role-based relational reasoning, of which analogical reasoning is a canonical example. The connectionist model proposed in the target article is only capable of conditional paired-associate learning.

van der Velde & de Kamps argue for the importance of considering the binding problem in accounts of human mental representation. However, their proposed solution fails as a complete account because it represents the bindings between roles and their fillers through associations (or connections). In addition, many criticisms leveled by the authors towards synchrony-based bindings models do not hold.

The fundamental relations that underlie cognitive comparisons—“same” and “different”—can be defined at multiple levels of abstraction, which vary in relational complexity. We compared response times to decide whether or not two sequentially‐presented patterns, each composed of two pairs of colored squares, were the same at three levels of abstraction: perceptual, relational, and system (higher order relations). For both 150 ms and 5 s inter‐stimulus intervals (ISIs), both with and without a masking stimulus, decision time increased with level of abstraction. Sameness (...) at lower complexity levels contributed to decisions based on the higher levels. The pattern of comparison times across levelswas not predictable solely from encoding times. The results indicated that relations at multiple levels of complexity can be abstracted and compared in working memory, with higher complexity levels requiring more processing time. We simulated the impact of relational complexity on response time using Learning and Inference with Schemas and Analogies (LISA), a computational model of relational comparisons based on dynamic binding of elements into roles in a relational working memory. (shrink)

We are big fans of propositions. But we are not big fans of the proposed by Mitchell et al. The authors ignore the critical role played by implicit, non-inferential processes in biological cognition, overestimate the work that propositions alone can do, and gloss over substantial differences in how different kinds of animals and different kinds of cognitive processes approximate propositional representations.

Theories of analogical reasoning have viewed relational structure as the dominant determinant of analogical mapping and inference, while assigning lesser importance to similarity between individual objects. An experiment is reported in which these two sources of constraints on analogy are placed in competition under conditions of high relational complexity. Results demonstrate equal importance for relational structure and object similarity, both in analogical mapping and in inference generation. The human data were successfully simulated using a computational analogy model (LISA) that treats (...) both relational correspondences and object similarity as soft constraints that operate within a limited-capacity working memory; but not with a model (SME) that treats relational structure as pre-eminent. (shrink)

What kind of evidence will lead people to revise their moral beliefs? Moral beliefs are often strongly held convictions, and existing research has shown that morality is rooted in emotion and socialization rather than deliberative reasoning. In addition, more general issues—such as confirmation bias—further impede coherent belief revision. Here, we explored a unique means for inducing belief revision. In two experiments, participants considered a moral dilemma in which an overwhelming majority of people judged that it was inappropriate to take action (...) to maximize utility. Their judgments contradicted a utilitarian principle they otherwise strongly endorsed. Exposure to this scenario led participants to revise their belief in the utilitarian principle, and this revision persisted over several hours. This method provides a new avenue for inducing belief revision. (shrink)

What does Édouard Glissant have to contribute to theorizing decolonization and a philosophy of difference? And how is this contribution tied to rethinking place (from Caribbean to Caribbeanness) and world (comprised of creolized culture and identity)? This essay takes up Glissant’s work in the context of questions of history and memory, with particular focus on how historical experience grounds philosophical work on place and world through articulations of identity, language, cultural production, and thinking after catastrophe. Drawing from a contrast with (...) Martin Heidegger and Walter Benjamin, this essay shows how Glissant’s critical concept “creolization” works as an expression of what I call “the afropostmodern,” and also as an important contribution to decolonial theories of world, culture, memory, language, and subjectivity after the Middle Passage. (shrink)

This essay tracks the movement in Édouard Glissant’s work from thinking relationality as creolisation to Relation as such, to a globalised sense of cultural contact and transformation he ca...

Kant both says that we should recognize our duties as God’s commands, and objects to the theological version of heteronomy, ‘which derives morality from a divine and supremely perfect will’. In this paper I discuss how these two views fit together, and in the process I develop a notion of autonomous submission to divine moral authority. I oppose the ‘constitutive’ view of autonomy proposed by J. B. Schneewind and Christine Korsgaard. I locate Kant’s objection to theological heteronomy against the background (...) of Crusius’s divine command theory, and I compare Kant’s views about divine authority and human political authority. (shrink)

My assigned task is to lay out the shape of my father's life and faith. This is daunting, but it is also a privilege because I loved him and admired him, and his life has been central in shaping my own. I am speaking also on behalf of my mother, my three sisters, Bridget, Louise and Ellie, and our children, Catherine and Andrew, Sam and Anisa, Hannah and Matty.

"I suspect that this will become the book on Schopenhauer's metaphysics for many years to come, and will be required reading for any serious student of Schopenhauer's thought."--David E. Cartwright, University of Wisconsin, Whitewater.

Environmentalism is usually thought to be a liberal political position, but the two primary schools of thought within the conservative intellectual movement support environmentalism as well. The free market perspective has received considerable attention for its potential contributions to environmental protection, but the traditionalist perspective has not. In this essay, I consider several important principles of traditionalist conservatism. The traditionalists are not materialists and are highly critical of our consumer culture. They reject ideology and stress piety toward nature, the intergenerational (...) character of society, and prudence in political and social action. These basic principles are a solid foundation for environmentalism. (shrink)