Cameron Buckner

The Aesthetic Theory of Art—any theory of art claiming that the aesthetic is a descriptively necessary feature of art—has been largely repudiated, in light of what are now considered traditional counter-examples. Mag Uidhir and Buckner argue that the Aesthetic Theory of Art can be more plausibly recast by abandoning aesthetic-feature possession by the artwork for a claim about aesthetic-concept possession by the artist. Aesthetic Theory so re-framed suggests that the aesthetic might have a central and substantial explanatory role to play within both traditional philosophical enquiries and recent, more empirical enquiries into the psychological and cognitive aspects of art and its practice. Finally, they discuss the directions this new work might take—by tying art theory to investigations of the distinctive sensorimotor capacities of expert artists, their specialized aesthetic conceptual schemata, and the ways these distinctive capacities and schemata contribute to the production of artworks.

In this paper, the second of two companion pieces, we explore novel philosophical questions raised by recent progress in large language models (LLMs) that go beyond the classical debates covered in the first part. We focus particularly on issues related to interpretability, examining evidence from causal intervention methods about the nature of LLMs' internal representations and computations. We also discuss the implications of multimodal and modular extensions of LLMs, recent debates about whether such systems may meet minimal criteria for consciousness, and concerns about secrecy and reproducibility in LLM research. Finally, we discuss whether LLM-like systems may be relevant to modeling aspects of human cognition, if their architectural characteristics and learning scenario are adequately constrained.

Large language models like GPT-4 have achieved remarkable proficiency in a broad spectrum of language-based tasks, some of which are traditionally associated with hallmarks of human intelligence. This has prompted ongoing disagreements about the extent to which we can meaningfully ascribe any kind of linguistic or cognitive competence to language models. Such questions have deep philosophical roots, echoing longstanding debates about the status of artificial neural networks as cognitive models. This article -- the first part of two companion papers -- serves both as a primer on language models for philosophers, and as an opinionated survey of their significance in relation to classic debates in the philosophy cognitive science, artificial intelligence, and linguistics. We cover topics such as compositionality, language acquisition, semantic competence, grounding, world models, and the transmission of cultural knowledge. We argue that the success of language models challenges several long-held assumptions about artificial neural networks. However, we also highlight the need for further empirical investigation to better understand their internal mechanisms. This sets the stage for the companion paper (Part II), which turns to novel empirical methods for probing the inner workings of language models, and new philosophical questions prompted by their latest developments.

In this essay, I provide a forward-looking naturalized theory of mental content designed to accommodate predictive processing approaches to the mind, which are growing in popularity in philosophy and cognitive science. The view is introduced by relating it to one of the most popular backward-looking teleosemantic theories of mental content, Fred Dretske’s informational teleosemantics. It is argued that such backward-looking views (which locate the grounds of mental content in the agent’s evolutionary or learning history) face a persistent tension between ascribing determinate contents and allowing for the possibility of misrepresentation. A way to address this tension is proposed by grounding content attributions in the agent’s own ability to detect when it has represented the world incorrectly through the assessment of prediction errors—which in turn allows the organism to more successfully represent those contents in the future. This opens up space for misrepresentation, but that space is constrained by the forward-directed epistemic capacities that the agent uses to evaluate and shape its own representational strategies. The payoff of the theory is illustrated by showing how it can be applied to interpretive disagreements over content ascriptions amongst scientists in comparative psychology and ethology. This theory thus provides a framework in which to make content attributions to representations posited by an exciting new family of predictive approaches to cognition, and in so doing addresses persistent tensions with the previous generation of naturalized theories of content.

Recent breakthroughs in artificial intelligence have primarily resulted from training deep neural networks (DNNs) with vast numbers of adjustable parameters on enormous datasets. Due to their complex internal structure, DNNs are frequently characterized as inscrutable ``black boxes,'' making it challenging to interpret the mechanisms underlying their impressive performance. This opacity creates difficulties for explanation, safety assurance, trustworthiness, and comparisons to human cognition, leading to divergent perspectives on these systems. This chapter examines recent developments in interpretability methods for DNNs, with a focus on interventionist approaches inspired by causal explanation in philosophy of science. We argue that these methods offer a promising avenue for understanding how DNNs process information compared to merely behavioral benchmarking and correlational probing. We review key interventionist methods and illustrate their application through practical case studies. These methods allow researchers to identify and manipulate specific computational components within DNNs, providing insights into their causal structure and internal representations. We situate these approaches within the broader framework of causal abstraction, which aims to align low-level neural computations with high-level interpretable models. While acknowledging current limitations, we contend that interventionist methods offer a path towards more rigorous and theoretically grounded interpretability research, potentially informing both AI development and computational cognitive neuroscience.

In this article, we explore various arguments against the traditional distinction between episodic and semantic memory based on the metaphysical phenomenon of transitional gradation. Transitional gradation occurs when two candidate kinds A and B grade into one another along a continuum according to their characteristic properties. We review two kinds of arguments—from the gradual semanticization of episodic memories as they are consolidated, and from the composition of episodic memories during storage and recall from semantic memories—that predict the proliferation of such transitional forms. We further explain why the distinction cannot be saved from the challenges of transitional gradation by appealing to distinct underlying memory structures and applying our perspective to the impasse over research into ‘episodic-like’ memory in non-human animals. On the whole, we recommend replacing the distinction with a dynamic life cycle of memory in which a variety of transitional forms will proliferate, and illustrate the utility of this perspective by tying together recent trends in animal episodic memory research and recommending productive future directions.

Recent experimental philosophy arguments have raised trouble for philosophers' reliance on armchair intuitions. One popular line of response has been the expertise defense: philosophers are highly-trained experts, whereas the subjects in the experimental philosophy studies have generally been ordinary undergraduates, and so there's no reason to think philosophers will make the same mistakes. But this deploys a substantive empirical claim, that philosophers' training indeed inculcates sufficient protection from such mistakes. We canvass the psychological literature on expertise, which indicates that people are not generally very good at reckoning who will develop expertise under what circumstances. We consider three promising hypotheses concerning what philosophical expertise might consist in: (i) better conceptual schemata; (ii) mastery of entrenched theories; and (iii) general practical know-how with the entertaining of hypotheticals. On inspection, none seem to provide us with good reason to endorse this key empirical premise of the expertise defense

This book provides a framework for thinking about foundational philosophical questions surrounding machine learning as an approach to artificial intelligence. Specifically, it links recent breakthroughs in deep learning to classical empiricist philosophy of mind. In recent assessments of deep learning's current capabilities and future potential, prominent scientists have cited historical figures from the perennial philosophical debate between nativism and empiricism, which primarily concerns the origins of abstract knowledge. These empiricists were generally faculty psychologists; that is, they argued that the active engagement of general psychological faculties-such as perception, memory, imagination, attention, and empathy-enables rational agents to extract abstract knowledge from sensory experience. This book explains a number of recent attempts to model roles attributed to these faculties in deep neural network based artificial agents by appeal to the faculty psychology of philosophers such as Aristotle, Ibn Sina (Avicenna), John Locke David Hume, William James, and Sophie de Grouchy. It illustrates the utility of this interdisciplinary connection by showing how it can provide benefits to both philosophy and computer science: computer scientists can continue to mine the history of philosophy for ideas and aspirational targets to hit on the way to more robustly rational artificial agents, and philosophers can see how some of the historical empiricists' most ambitious speculations can be realized in specific computational systems.

In this essay, I provide a forward-looking naturalized theory of mental content designed to accommodate predictive processing approaches to the mind, which are growing in popularity in philosophy and cognitive science. The view is introduced by relating it to one of the most popular backward-looking teleosemantic theories of mental content, Fred Dretske’s informational teleosemantics. It is argued that such backward-looking views (which locate the grounds of mental content in the agent’s evolutionary or learning history) face a persistent tension between ascribing determinate contents and allowing for the possibility of misrepresentation. A way to address this tension is proposed by grounding content attributions in the agent’s own ability to detect when it has represented the world incorrectly through the assessment of prediction errors—which in turn allows the organism to more successfully represent those contents in the future. This opens up space for misrepresentation, but that space is constrained by the forward-directed epistemic capacities that the agent uses to evaluate and shape its own representational strategies. The payoff of the theory is illustrated by showing how it can be applied to interpretive disagreements over content ascriptions amongst scientists in comparative psychology and ethology. This theory thus provides a framework in which to make content attributions to representations posited by an exciting new family of predictive approaches to cognition, and in so doing addresses persistent tensions with the previous generation of naturalized theories of content.

The last 5 years have seen a series of remarkable achievements in deep-neural-network-based artificial intelligence research, and some modellers have argued that their performance compares favourably to human cognition. Critics, however, have argued that processing in deep neural networks is unlike human cognition for four reasons: they are (i) data-hungry, (ii) brittle, and (iii) inscrutable black boxes that merely (iv) reward-hack rather than learn real solutions to problems. This article rebuts these criticisms by exposing comparative bias within them, in the process extracting some more general lessons that may also be useful for future debates.

Deep learning is currently the most prominent and widely successful method in artificial intelligence. Despite having played an active role in earlier artificial intelligence and neural network research, philosophers have been largely silent on this technology so far. This is remarkable, given that deep learning neural networks have blown past predicted upper limits on artificial intelligence performance—recognizing complex objects in natural photographs and defeating world champions in strategy games as complex as Go and chess—yet there remains no universally accepted explanation as to why they work so well. This article provides an introduction to these networks as well as an opinionated guidebook on the philosophical significance of their structure and achievements. It argues that deep learning neural networks differ importantly in their structure and mathematical properties from the shallower neural networks that were the subject of so much philosophical reflection in the 1980s and 1990s. The article then explores several different explanations for their success and ends by proposing three areas of inquiry that would benefit from future engagement by philosophers of mind and science.

Our prominent definitions of cognition are too vague and lack empirical grounding. They have not kept up with recent developments, and cannot bear the weight placed on them across many different debates. I here articulate and defend a more adequate theory. On this theory, behaviors under the control of cognition tend to display a cluster of characteristic properties, a cluster which tends to be absent from behaviors produced by non-cognitive processes. This cluster is reverse-engineered from the empirical tests that comparative psychologists use to determine whether a behavior was generated by a cognitive or a non-cognitive process. Cognition should be understood as the natural kind of psychological process that non-accidentally exhibits the properties assessed by these tests (as well as others we have not yet discovered). Finally, I review two plausible neural accounts of cognition's underlying mechanisms—one based in localization of function to particular brain regions and another based in the more recent distributed networks approach to neuroscience—which would explain why these properties non-accidentally cluster. While this notion of cognition may be useful for a number of debates, I here focus on its application to a recent crisis over the distinction between cognition and association in comparative psychology.

A surge of empirical research demonstrating flexible cognition in animals and young infants has raised interest in the possibility of rational decision‐making in the absence of language. A venerable position, which I here call “Classical Inferentialism”, holds that nonlinguistic agents are incapable of rational inferences. Against this position, I defend a model of nonlinguistic inferences that shows how they could be practically rational. This model vindicates the Lockean idea that we can intuitively grasp rational connections between thoughts by developing the Davidsonian idea that practical inferences are at bottom categorization judgments. From this perspective, we can see how similarity‐based categorization processes widely studied in human and animal psychology might count as practically rational. The solution involves a novel hybrid of internalism and externalism: intuitive inferences are psychologically rational (in the explanatory sense) given the intensional sensitivity of the similarity assessment to the internal structure of the agent's reasons for acting, but epistemically rational (in the justificatory sense) given an ecological fit between the features matched by that assessment and the structure of the agent's environment. The essay concludes by exploring empirical results that show how nonlinguistic agents can be sensitive to these similarity assessments in a way that grants them control over their opaque judgments.

In artificial intelligence, recent research has demonstrated the remarkable potential of Deep Convolutional Neural Networks (DCNNs), which seem to exceed state-of-the-art performance in new domains weekly, especially on the sorts of very difficult perceptual discrimination tasks that skeptics thought would remain beyond the reach of artificial intelligence. However, it has proven difficult to explain why DCNNs perform so well. In philosophy of mind, empiricists have long suggested that complex cognition is based on information derived from sensory experience, often appealing to a faculty of abstraction. Rationalists have frequently complained, however, that empiricists never adequately explained how this faculty of abstraction actually works. In this paper, I tie these two questions together, to the mutual benefit of both disciplines. I argue that the architectural features that distinguish DCNNs from earlier neural networks allow them to implement a form of hierarchical processing that I call “transformational abstraction”. Transformational abstraction iteratively converts sensory-based representations of category exemplars into new formats that are increasingly tolerant to “nuisance variation” in input. Reflecting upon the way that DCNNs leverage a combination of linear and non-linear processing to efficiently accomplish this feat allows us to understand how the brain is capable of bi-directional travel between exemplars and abstractions, addressing longstanding problems in empiricist philosophy of mind. I end by considering the prospects for future research on DCNNs, arguing that rather than simply implementing 80s connectionism with more brute-force computation, transformational abstraction counts as a qualitatively distinct form of processing ripe with philosophical and psychological significance, because it is significantly better suited to depict the generic mechanism responsible for this important kind of psychological processing in the brain.

The Darwinian protolanguage hypothesis is one of the most popular theories of the evolution of human language. According to this hypothesis, language evolved through a three stage process involving general increases in intelligence, the emergence of grammatical structure as a result of sexual selection on protomusical songs, and finally the attachment of meaning to the components of those songs. The strongest evidence for the second stage of this process has been considered to be birdsong, and as a result researchers have investigated the existence of various forms of grammar in the production and comprehension of songs by birds. Here, we argue that mating dances are another relevant source of sexually-selected complexity that has until now been largely overlooked by proponents of Darwinian protolanguage, focusing especially on the dances of long-tailed manakins. We end by sketching several lines of research that should be pursued to determine the relevance of mating dances to the evolution of language.

A surge of empirical research demonstrating flexible cognition in animals and young infants has raised interest in the possibility of rational decision-making in the absence of language. A venerable position, which I here call “Classical Inferentialism”, holds that nonlinguistic agents are incapable of rational inferences. Against this position, I defend a model of nonlinguistic inferences that shows how they could be practically rational. This model vindicates the Lockean idea that we can intuitively grasp rational connections between thoughts by developing the Davidsonian idea that practical inferences are at bottom categorization judgments. From this perspective, we can see how similarity-based categorization processes widely studied in human and animal psychology might count as practically rational. The solution involves a novel hybrid of internalism and externalism: intuitive inferences are psychologically rational (in the explanatory sense) given the intensional sensitivity of the similarity assessment to the internal structure of the agent's reasons for acting, but epistemically rational (in the justificatory sense) given an ecological fit between the features matched by that assessment and the structure of the agent's environment. The essay concludes by exploring empirical results that show how nonlinguistic agents can be sensitive to these similarity assessments in a way that grants them control over their opaque judgments.

The application of digital humanities techniques to philosophy is changing the way scholars approach the discipline. This paper seeks to open a discussion about the difficulties, methods, opportunities, and dangers of creating and utilizing a formal representation of the discipline of philosophy. We review our current project, the Indiana Philosophy Ontology (InPhO) project, which uses a combination of automated methods and expert feedback to create a dynamic computational ontology for the discipline of philosophy. We argue that our distributed, expert-based approach to modeling the discipline carries substantial practical and philosophical benefits over alternatives. We also discuss challenges facing our project (and any other similar project) as well as the future directions for digital philosophy afforded by formal modeling.

Our prominent definitions of cognition are too vague and lack empirical grounding. They have not kept up with recent developments, and cannot bear the weight placed on them across many different debates. I here articulate and defend a more adequate theory. On this theory, behaviors under the control of cognition tend to display a cluster of characteristic properties, a cluster which tends to be absent from behaviors produced by non-cognitive processes. This cluster is reverse-engineered from the empirical tests that comparative psychologists use to determine whether a behavior was generated by a cognitive or a non-cognitive process. Cognition should be understood as the natural kind of psychological process that non-accidentally exhibits the properties assessed by these tests (as well as others we have not yet discovered). Finally, I review two plausible neural accounts of cognition's underlying mechanisms?one based in localization of function to particular brain regions and another based in the more recent distributed networks approach to neuroscience?which would explain why these properties non-accidentally cluster. While this notion of cognition may be useful for a number of debates, I here focus on its application to a recent crisis over the distinction between cognition and association in comparative psychology

I here critique the application of the traditional, similarity-based account of natural kinds to debates in psychology. A challenge to such accounts of kindhood—familiar from the study of biological species—is a metaphysical phenomenon that I call ‘transitional gradation’: the systematic progression of slightly modified transitional forms between related candidate kinds. Where such gradation proliferates, it renders the selection of similarity criteria for kinds arbitrary. Reflection on general features of learning—especially on the gradual revision of concepts throughout the acquisition of expertise—shows that even the strongest candidates for similarity-based kinds in psychology exhibit systematic transitional gradation. As a result, philosophers of psychology should abandon discussion of kindhood, or explore non-similarity based accounts.

The functionalist approach to kinds has suffered recently due to its association with law-based approaches to induction and explanation. Philosophers of science increasingly view nomological approaches as inappropriate for the special sciences like psychology and biology, which has led to a surge of interest in approaches to natural kinds that are more obviously compatible with mechanistic and model-based methods, especially homeostatic property cluster theory. But can the functionalist approach to kinds be weaned off its dependency on laws? Dan Weiskopf has recently offered a reboot of the functionalist program by replacing its nomological commitments with a model-based approach more closely derived from practice in psychology. Roughly, Weiskopf holds that the natural kinds of psychology will be the functional properties that feature in many empirically successful cognitive models, and that those properties need not be localized to parts of an underlying mechanism. I here skeptically examine the three modeling practices that Weiskopf thinks introduce such non-localizable properties: fictionalization, reification, and functional abstraction. In each case, I argue that recognizing functional properties introduced by these practices as autonomous kinds comes at clear cost to those explanations’ counterfactual explanatory power. At each step, a tempting functionalist response is parochialism: to hold that the false or omitted counterfactuals fall outside the modeler’s explanatory aims, and so should not be counted against functional kinds. I conclude by noting the dangers this attitude poses to scientific disagreement, inviting functionalists to better articulate how the individuation conditions for functional kinds might outstrip the perspective of a single modeler.

How should we determine the distribution of psychological traits—such as Theory of Mind, episodic memory, and metacognition—throughout the Animal kingdom? Researchers have long worried about the distorting effects of anthropomorphic bias on this comparative project. A purported corrective against this bias was offered as a cornerstone of comparative psychology by C. Lloyd Morgan in his famous “Canon”. Also dangerous, however, is a distinct bias that loads the deck against animal mentality: our tendency to tie the competence criteria for cognitive capacities to an exaggerated sense of typical human performance. I dub this error “anthropofabulation”, since it combines anthropocentrism with confabulation about our own prowess. Anthropofabulation has long distorted the debate about animal minds, but it is a bias that has been little discussed and against which the Canon provides no protection. Luckily, there is a venerable corrective against anthropofabulation: a principle offered long ago by David Hume, which I call “Hume’s Dictum”. In this paper, I argue that Hume’s Dictum deserves a privileged place next to Morgan’s Canon in the methodology of comparative psychology, illustrating my point through a discussion of the debate over Theory of Mind in nonhuman animals.

Philosophers and cognitive scientists have worried that research on animal mind-reading faces a ‘logical problem’: the difficulty of experimentally determining whether animals represent mental states (e.g. seeing) or merely the observable evidence (e.g. line-of-gaze) for those mental states. The most impressive attempt to confront this problem has been mounted recently by Robert Lurz. However, Lurz' approach faces its own logical problem, revealing this challenge to be a special case of the more general problem of distal content. Moreover, participants in this debate do not agree on criteria for representation. As such, future debate should either abandon the representational idiom or confront underlying semantic disagreements.

In her target article, Jennifer McMahon argues that we understand art not by explicitly interpreting “raw percepts,” but rather by engaging with our implicit tendencies to interpret complex stimuli in terms of culturally-engrained preconceptions and narratives. These attributions of order require a shared conceptual and cultural background, and thus one might worry that in denying access to raw percepts, the view dulls art’s critical edge. Against this worry, McMahon argues that art can continue to create and innovate by inviting us to critically reflect upon the very preconceptions on which our engagement with it necessarily depends. In this commentary, I place these attributions of order in historical and empirical context. In addition, I discuss a lingering, related mystery — the possibility of the occasionally punctuated character of artistic evolution, in which prevailing aesthetic conventions are replaced with almost entirely new ones. I suggest that such radical breaks with the past are possible even given the concept-ladeness of perception, but are only likely to succeed when they tap into a culturally-invariant bedrock of more basic attributions of order.