-
83021.239174
What do large language models actually model? Do they tell us something about human capacities, or are they models of the corpus we’ve trained them on? I give a non-deflationary defence of the latter position. Cognitive science tells us that linguistic capabilities in humans rely supralinear formats for computation. The transformer architecture, by contrast, supports at best a linear formats for processing. This argument will rely primarily on certain invariants of the computational architecture of transformers. I then suggest a positive story about what transformers are doing, focusing on Liu et al. (2022)’s intriguing speculations about shortcut automata. I conclude with why I don’t think this is a terribly deflationary story. Language is not (just) a means for expressing inner state but also a kind of ‘discourse machine’ that lets us make new language given appropriate context. We have learned to use this technology in one way; LLMs have also learned to use it too, but via very different means.
-
185683.239224
Interactions between agents are supported through a continuous process of detecting and responding to behaviors that are contingent upon the other agent’s behavior. Here, we explore the temporal dependence of these mechanisms, focusing on the role of timescale compatibility in inter-agent interactions. Using continuous-time recurrent neural networks (CTRNNs) to control embodied agents in a minimal social interaction task, we demonstrate that effective interactions require agents to operate on compatible timescales. Our results indicate that time scale mismatches disrupt agents’ ability to distinguish other agents from non-social entities, revealing a timescale threshold beyond which agents begin mis-classifying slower agents as static objects and faster agents as non-social animate objects.
-
329271.239232
This paper is divided in two parts. In part I, I argue against two attempts to naturalise the notion of scientific representation, by reducing it to isomorphism and similarity. I distinguish between the means and the constituents of representation, and I argue that isomorphism and similarity are common (although not universal) means of representation; but that they are not constituents of scientific representation. I look at the prospects for weakened versions of these theories, and I argue that only those that abandon the aim to naturalise scientific representation are likely to be successful. In part II of the paper, I present a deflationary conception of scientific representation, which minimally characterises it by means of two necessary conditions: representation is essentially intentional and it has the capacity to allow surrogate reasoning and inference. I then defend this conception by showing that it successfully meets the objections and difficulties that make its competitors, such as isomorphism and similarity, untenable. In addition the inferential conception explains the success of various means of representation in their appropriate domains, and it sheds light on the truth and accuracy of scientific representations.
-
353180.239238
Economists have long scoffed at know-it-all business and financial gurus with the rhetorical question, “If you’re so smart, why aren’t you rich?” And sometimes the gurus use the same question to scoff at know-it-all economists. …
-
513400.239244
It’s widely held that we perceive not only low-level properties, such as colors and shapes, but also high-level properties, such as the property of being a dog or of being a moving train. Debate about which types of property we perceive has recently eclipsed the question of how perceiving itself operates. We focus here on that latter question, proposing an account on which perception of low-level properties occurs by way of mental qualities alone, whereas perception of high-level properties occurs by way of mental qualities together with conceptual content of the type that figures in thinking. It is central to our account that mental qualities have a type of representational character unique to them, so that mental qualities can interact representationally with conceptual content in perceiving. We present a number of advantages of this account, including how it fits with a range of experimental findings, and address several objections to it.
-
529108.23925
This paper critically analyses the “attention economy” within the framework of cognitive science and techno-political economics, as applied to both human and machine interactions. We explore how current business models, particularly in digital platform capitalism, harness user engagement by strategically shaping attentional patterns. These platforms utilize advanced AI and massive data analytics to enhance user engagement, creating a cycle of attention capture and data extraction. We review contemporary (neuro)cognitive theories of attention and platform engagement design techniques and criticize classical cognitivist and behaviourist theories for their inadequacies in addressing the potential harms of such engagement on user autonomy and wellbeing. 4E approaches to cognitive science, instead, emphasizing the embodied, extended, enactive, and ecological aspects of cognition, offer us an intrinsic normative standpoint and a more integrated understanding of how attentional patterns are actively constituted by adaptive digital environments. By examining the precarious nature of habit formation in digital contexts, we reveal the techno-economic underpinnings that threaten personal autonomy by disaggregating habits away from the individual, into an AI managed collection of behavioural patterns. Our current predicament suggests the necessity of a paradigm shift towards an ecology of attention. This shift aims to foster environments that respect and preserve human cognitive and social capacities, countering the exploitative tendencies of cognitive capitalism.
-
877061.239255
Cognitive scientists ascribe inferential processes to (neuro)cognitive systems to explain many of their capacities. Since these ascriptions have different connotations, philosophical accounts of inference could help clarify their assumptions and forestall potential confusion. However, many existing accounts define inference in ways that are out of touch with successful scientific practice – ways that overly intellectualise inference, construe inference in complete opposition to association, and imply that inferential processes prevent minds from being in contact with the outside world. In this chapter, we combine Siegel’s (2017) Response Hypothesis with insights from basal cognition and ecological rationality to sketch a philosophically viable, updated account of inference in (neuro)cognitive systems. According to this view, inference is a kind of rationally evaluable transition from some inputs or current representations to some conclusion or output representation. This notion of inference aligns with and can illuminate scientific practices in disparate fields, while eschewing a commitment to a consciously accessible language-like neural code or a formal system of mental logic, highlighting the continuity between inferential and associative processes, and allowing for a non-indirect mind-world relationship, where minds are genuinely open and responsive to their environment.
-
1074746.239264
Despite its growing appeal for the study of consciousness, the notion of entropy has yet to lead to widely supported new insights about the nature of phenomenal experience. Typically, entropy measures of brain activity are found to correlate with cognitive functions that are assumed to index consciousness. Taking a very different approach, this theoretical framework does not conflate consciousness with any function. It presents a series of premises to argue that consciousness is fundamentally characterized as inactionable perception, i.e. that does not give rise to macrophysical action. This is then fitted in a framework of perception and action as informational changes in a dynamical neural state space. In this model, inactionable perception naturally arises as the prediction-driven increase of concept-related entropy. This entails an increase of (Shannon) information while its efficacy to produce macrophysical effects decreases, which is here referred to as information dissipation, analogously to energy dissipation in thermodynamic systems. It results from inefficient sensorimotor coupling with the environment, which occurs when behavior is not fixed relative to the stimulus. Despite the posited inefficacy of conscious perception, it consists of action-specific information and can therefore be interpreted as potential behavior. Starting from fundamental properties, this framework may provide a new and coherent conceptual basis for a fuller understanding of what consciousness is and how it relates to the physical world. Although many of its implications remain to be explored, it appears consistent with empirical findings, and prompts subtle reinterpretations of some classical results in perception research.
-
1190039.23927
Perspectival realism claims that scientific knowledge is always situated into a vantage point. We argue that ecological psychology offers a suitable framework to develop perspectival epistemologies. Ecological psychology stresses that perception is focused on affordances, i.e. the possibilities of interactions afforded by reality given the abilities of an organism. We call the integrating view as ecological perspectivism. It claims that science offers knowledge of reality in terms of affordances, which are relational to the instruments and abilities of scientific communities. Cognition is of affordances, and what a domain affords for scientists depends on which skills and technologies they avail. We connect this proposal with the main arguments for perspectivism. First, regarding instrumental detections, ecological perspectivism offers a realist account of perception that treats the use of instruments as tools that scaffold and extend embodied cognition. Second, regarding model pluralism, ecological perspectivism supports an artifactualist account of modelling as embodied cognition extended by tools.
-
1294877.239278
Poseidon who has the power to inflict a vengeful wrath. Yet Odysseus is overcome by pride at his own cleverness and shouts his own name from the prow of his ship, carelessly jeopardizing the safety of his crew. The parable of Odysseus and the Cyclops is a uniquely rich and compelling story, but it involves an utterly ordinary kind of failure to respond to reasons: ego eclipses prudence. In this moment, Odysseus is irrational and is responsible for this irrationality. His irrationality stems from the fact that he has violated the rational requirement to respond to his reasons. While we often meet this requirement in everyday life, we also often violate it by failing to respond to reasons due to ego, closed-mindedness, carelessness, or other poor epistemic habits. In such cases, our failures render us irrational.
-
1294959.239284
moment or not at all. Nonetheless, Lessing thought that there is at the disposal of the poet an indirect means to capture the beauty of material objects. Homer would have put it to good use in the Iliad, where the beauty of Helen of Troy was conveyed not by a description of her beauty-making features, but by a description of the effect of her beauty: “What Homer could not describe in detail he makes us understand by the effect: oh! poets paint for us the pleasure, inclination, love, rapture, which beauty causes, and you will have painted beauty itself” (Lessing 1836[1766], ). At the very least, what this passage makes clear is that
-
1394940.239289
Christopher Devlin Brown’s The Hope and Horror of Physicalism works through different ways of understanding the content of physicalism, evaluates the “existential consequences” of physicalism so understood, and attempts to defend one form of physicalism – “Russellian physicalism” – from consciousness-based objections. I first raise some minor-but-not-too-minor concerns about Brown’s historical account of physicalism. Second, I discuss one version of physicalism (the “theory-based version”) that Brown works with in assessing physicalism’s existential consequences. Third, I raise some questions about Brown’s preferred way of understanding physicalism, which he labels “Russellian physicalism”, and which is a version of “via negativa physicalism”. My discussions are offered in a constructive spirit.
-
1582918.239295
The self represents a multifactorial entity made up of several interrelated constructs. It is suggested that self-talk orchestrates interactions between most self-processes—especially those entailing self-reflection. A review of the literature is performed, specifically looking for representative studies (n = 12) presenting correlations between self-report measures of self-talk and self-reflective processes. Self-talk questionnaires include the Self-Talk Scale, the Varieties of Inner Speech Questionnaire, the General Inner Speech Questionnaire, and the Inner Speech Scale. The main self-reflection measures are the Rumination and Reflection Questionnaire, the Self-Consciousness Scale, and the Philadelphia Mindfulness Scale. Most measures comprise subscales which are also discussed. Findings include: (1) positive significant correlations between self-talk used for self-management/assessment and self-reflection, arguably because the latter entails self-regulation, which itself relies on self-directed speech; (2) positive significant correlations between critical self-talk and self-rumination, as both may recruit negative, repetitive, and uncontrollable self-thoughts; (3) negative associations between self-talk and the self-acceptance aspect of mindfulness, likely because thinking about oneself in the present in a non-judgmental way is best achieved by repressing one’s inner voice. Limitations are discussed, including the selective nature of the reported correlations. Experimentally manipulating self-talk would make it possible to further explore causal associations with self-processes.
-
1651931.2393
Daniel Dennett’s view about consciousness in nonhuman animals has two parts. One is a methodological injunction that we rely on our best theory of consciousness to settle that issue, a theory that must initially work for consciousness in humans. The other part is Dennett’s application of his own theory of consciousness, developed in Consciousness Explained (1991), which leads him to conclude that nonhuman animals are likely never in conscious mental states. I defend the methodological injunction as both sound and important, and argue that the alternative approaches that dominate the literature are unworkable. But I also urge that Dennett’s theory of consciousness and his arguments against conscious states in nonhuman animals face significant difficulties. Those difficulties are avoided by a higher-order-thought theory of consciousness, which is close to Dennett’s theory, and provides leverage in assessing which kinds of mental state are likely to be conscious in nonhuman animals. Finally, I describe a promising experimental strategy for showing that conscious states do occur in some nonhuman animals, which fits comfortably with the higher-order-thought theory but not with Dennett’s.
-
1824915.239306
Hypotheses about how and why animals behave the way they do are frequently labelled as either associative or cognitive. This has been taken as evidence that there is a fundamental distinction between two kinds of behavioural processes. However, there is significant disagreement about how to define this distinction whether it ought to be rejected entirely. Rather than seeking a definition of the associative-cognitive distinction, or advocating for its rejection, I argue that it is an artefact of the way that comparative psychologists generate hypotheses. I suggest that hypotheses for non-human animal behaviour are often generated by analogy with hypotheses drawn from human psychology and associative learning theory, a justifiable strategy since analogies help to establish the pursuit-worthiness of a hypothesis. Any apparent distinction is a misleading characterisation of what is a complex web of hypotheses that explain diverse behavioural phenomena. The analogy view of the distinction has three advantages. It motivates the apparent existence of the distinction based on a common inference strategy in science, analogical reasoning. It accounts for why the distinction has been difficult to articulate, because of the diversity of possible analogies. Finally, it delimits the role of the distinction in downstream inferences about animal behaviour.
-
1857809.239312
This commentary aims to support Tim Crane’s account of the structure of intentionality by showing how intentional objects are naturalistically respectable, how they pair with concepts, and how they are to be held distinct from the referents of thought. Crane is right to reject the false dichotomy that accounts of intentionality must be either reductive and naturalistic or non-reductive and traffic in mystery. The paper has three main parts. First, I argue that the notion of an intentional object is a phenomenological one, meaning that it must be understood as an object of thought for a subject. Second, I explain how Mark Sainsbury’s Display Theory of Attitude Attribution pairs well with Crane’s notion of an intentional object and allows for precisification in intentional state attributions while both avoiding exotica and capturing the subject’s perspective on the world. Third, I explain the reification fallacy, the fallacy of examining intentional objects as if they exist independently of subjects and their conceptions of them. This work helps to bring out how intentionality can fit in the natural world while at the same time not reducing aboutness to some non-intentional properties of the natural world.
-
1872308.23932
— We present a reformulation of the model predictive control problem using a Legendre basis. To do so, we use a Legendre representation both for prediction and optimization. For prediction, we use a neural network to approximate the dynamics by mapping a compressed Legendre representation of the control trajectory and initial conditions to the corresponding compressed state trajectory. We then reformulate the optimization problem in the Legendre domain and demonstrate methods for including optimization constraints. We present simulation results demonstrating that our implementation provides a speedup of 31-40 times for comparable or lower tracking errors with or without constraints on a benchmark task.
-
1943372.239325
Modern generative AI systems have shown the capacity to produce remarkably fluent language, prompting debates both about their semantic understanding and, less prominently, about whether they can perform speech acts. This paper addresses the latter question, focusing on assertion. We argue that to be capable of assertion, an entity must meet two requirements: it must produce outputs with descriptive functions, and it must be capable of being sanctioned by agents with which it interacts. The second requirement arises from the nature of assertion as a norm-governed social practice. Pre-trained large language models that have not been subject to fine-tuning fail to meet the first requirement. Language models that have been fine-tuned for “groundedness” or “correctness” may meet the first requirement, but fail the second. We also consider the significance of the point that AI systems can be used to generate proxy assertions on behalf of human agents.
-
1943463.239331
Suppose Socrates is looking at a bright red apple in good viewing conditions, so that it looks to him the colour it is. Schematically, Aristotle’s explanation of this “Good Case” is that the apple looks bright red to Socrates because he has taken on the perceptual form of bright red without the matter. But what happens if Socrates misperceives the apple instead and it looks purple? It is not at all clear how to apply Aristotle’s account of perception to such a “Bad Case.” Does Socrates still take on the perceptual form of the actual—bright red—colour of the apple in the Bad Case? Of purple? Neither? I argue that applying Aristotle’s account of perception to this sort of Bad Case requires that there are different ways of being in perceptual contact with perceptible qualities like the colour of an apple, depending on how that perceptual contact is mediated by changes in the sense organs and perceptual medium.
-
2171279.239336
In this article, I develop the idea of theoretical complexes to characterize large-scale theoretical movements in the cognitive sciences, such as classical computational cognitivism, connectionism, embodied cognition, and predictive processing. It is argued that these theoretical movements should be construed as groups of closely connected individual theories and models of cognitive processes that share similar general hypotheses about the nature of cognition. General hypotheses form conceptual cores of complexes of cognitive theories, giving them their structure and functional properties. The latter are said to consist primarily of helping practitioners of theoretical complexes further develop their individual accounts of cognitive phenomena. It is claimed that the theoretical diversity fostered in this way has already benefited the cognitive sciences in a number of important ways and has the potential to further advance the field.
-
2401985.239342
This article uses especially Sartre’s existential philosophy (also drawing from Scheler, Husserl, and Descartes) to investigate pathogenetic issues in psychopathology from a first-person perspective. Psychosis is a “total experience” that points to orientating changes in subjectivity, supported by evidence regarding self-disorders in the schizophrenia spectrum. This article proposes that schizophrenia is essentially characterized (and distinguished) by specific structural alterations of (inter)subjectivity around the relationship between self and Other, which all its seemingly disparate signs and symptoms eventually point to. Two reciprocal distortions are present in psychotic schizophrenia patients: (A) an encroaching and substantialized Other, and (B) a self transformed into being-for-the-Other. Under the altered conditions of (A & B), delusional mood is the presence but inaccessibility of the Other; a delusional perception is an eruption or surfacing of objectification of self by Other; a delusion is an experience of the Other, which fulfills certainty, incorrigibility, and potentially falsehood.
-
2461210.239347
Sorry for the long blog-hiatus! I was completely occupied for weeks, teaching an intensive course on theoretical computer science to 11-year-olds (! ), at a math camp in St. Louis that was also attended by my 8-year-old son. …
-
2515945.239352
philosophical logic may also interest themselves with the logical appendices, one of which presents modal logic as a subsystem of the logic of counterfactuals. Last but not least, the work also includes an afterword that is both a severe reprimand to the analytic community for a certain sloppiness and an exhortation to all colleagues to apply more rigor and patience in addressing metaphysical issues. People familiar with Williamson’s work will not be surprised by the careful and detailed (sometimes a bit technical) argumentation, which demands careful attention from the reader. As expected, this is a most relevant contribution to an increasingly popular topic by one of today’s leading analytic philosophers.
-
2517337.239358
Cognitive neuroscientists typically posit representations which relate to various aspects of the world, which philosophers call representational content. Anti-realists about representational content argue that contents play no role in neuroscientific explanations of cognitive capacities. In this paper, I defend realism against an anti-realist argument due to Frances Egan, who argues that for content to be explanatory it must be both essential and naturalistic. I introduce a case study from cognitive neuroscience in which content is both essential and naturalistic, meeting Egan’s challenge. I then spell out some general principles for identifying studies in which content plays an explanatory role.
-
2517383.239363
Representations appear to play a central role in cognitive science. Capacities such as face recognition are thought to be enabled by internal states or structures representing external items. However, despite the ubiquity of representational terminology in cognitive science, there is no explicit scientific theory outlining what makes an internal state a representation of an external item. Nonetheless, many philosophers hope to uncover an implicit theory in the scientific literature. This is the project of the current thesis. However, all such projects face an obstacle in the form of Frances Egan’s argument that content plays no role in scientific theorising. I respond that, in some limited regions of cognitive science, content is crucial for explanation. The unifying idea is that closer attention to the application of information theory in those regions of cognitive neuroscience enables us to uncover an implicit theory of content. I examine the conditions which must be met for the cognitive system to be modelled using information theory, presenting some constraints on how we apply the mathematical framework. For example, information theory requires identifying probability distributions over measurable outcomes, which leads us to focus specifically on neural representation. I then argue that functions are required to make tractable measures of information, since they serve to narrow the range of possible contents to those potentially explanatory of a cognitive capacity. However, unlike many other teleosemanticists, I argue that we need to use a non-etiological form of function. I consider whether non-etiological functions allow for misrepresentation, and conclude that they do. Finally, I introduce what I argue is the implicit theory of content in cognitive neuroscience: maxMI. The content of a representation is that item in the environment with which the representation shares maximal mutual information.
-
2860010.239369
Accuracy plays an important role in the deployment of machine learning algorithms. But accuracy is not the only epistemic property that matters. For instance, it is well-known that algorithms may perform accurately during their training phase but experience a significant drop in performance when deployed in real-world conditions. To address this gap, people have turned to the concept of algorithmic robustness. Roughly, robustness refers to an algorithm’s ability to maintain its performance across a range of real-world and hypothetical conditions. In this paper, we develop a rigorous account of algorithmic robustness grounded in Robert Nozick’s counterfactual sensitivity and adherence conditions for knowledge. By bridging insights from epistemology and machine learning, we offer a novel conceptualization of robustness that captures key instances of algorithmic brittleness while advancing discussions on reliable AI deployment. We also show how a sensitivity-based account of robustness provides notable advantages over related approaches to algorithmic brittleness, including causal and safety-based ones.
-
2921170.239374
Why are quantum correlations so puzzling? A standard answer is that they seem to require either nonlocal influences or conspiratorial coincidences. This suggests that by embracing nonlocal influences we can avoid conspiratorial fine-tuning. But that’s not entirely true. Recent work, leveraging the framework of graphical causal models, shows that even with nonlocal influences, a kind of fine-tuning is needed to recover quantum correlations. This fine-tuning arises because the world has to be just so as to disable the use of nonlocal influences to signal, as required by the no-signaling theorem. This places an extra burden on theories that posit nonlocal influences, such as Bohmian mechanics, of explaining why such influences are inaccessible to causal control. I argue that Everettian Quantum Mechanics suffers no such burden. Not only does it not posit nonlocal influences, it operates outside the causal models framework that was presupposed in raising the fine-tuning worry. Specifically, it represents subsystems with density matrices instead of random variables. This allows it to sidestep all the results (including EPR and Bell) that put quantum correlations in tension with causal models. However, this doesn’t mean one must abandon causal reasoning altogether in a quantum world. After all, quantum systems can clearly stand in causal relations. When decoherence is rampant and there’s no controlled entanglement, Everettian Quantum Mechanics licenses our continued use of standard causal models. When controlled entanglement is present—such as in Bell-type experiments—we can employ recently proposed quantum causal models that are consistent with Everettian Quantum Mechanics. We never need invoke any kind of nonlocal influence or any kind of fine-tuning.
-
2921223.23938
We take a fresh look at Daniel Dennett’s naturalist legacy in philosophy, focusing on his rethinking of philosophical methods. Critics sometimes mistake Dennett for promoting a crude naturalism or dismissing philosophical tools like first-person intuition. We present his approach as more methodologically radical, blending science and philosophy in a way that treats inquiry as an evolving process. Concepts and intuitions are tested and adjusted in light of empirical findings and broader epistemic aims. For Dennett, science isn’t a limitation on philosophy, but a tool that sharpens it, with empirical data helping to refine our understanding both of concepts and philosophical phenomena alike. By exploring Dennett’s methodological contributions, we underscore the ongoing importance of his naturalist perspective in today’s philosophical landscape.
-
2921292.239387
In this paper, we argue that a perceiver’s contributions to perception can substantially affect what objects are represented in perceptual experience. To capture the scalar nature of these perceiver-contingent contributions, we introduce three grades of subject-dependency in object perception. The first grade, “weak subject-dependency,” concerns attentional changes to perceptual content like, for instance, when a perceiver turns their head, plugs their ears, or primes their attention to a particular cue. The second grade, “moderate subject-dependency,” concerns changes in the contingent features of perceptual objects due to action-orientation, location, and agential interest. For instance, being to the right or left of an object will cause the object to have a corresponding locative feature, but that feature is non-essential to the object in question. Finally, the third grade, “strong subject-dependency,” concerns generating perceptual objects whose existence depends upon their perceivers’ sensory contributions to perception. For this final grade of subject-dependency the adaptive perceptual system shapes diverse representations of sensory information by contributing necessary features to perceptual objects. To exemplify this nonstandard form of object perception we offer evidence from the future-directed anticipation of perceptual experts, and from the feature binding of synesthetes. We conclude that strongly subject-dependent perceptual objects are more than mere material objects, but are rather a necessary combination of material objects with the contributions of a perceiving subject.
-
3036672.239393
This paper introduces "pseudo-consciousness" as a novel framework for understanding and classifying advanced artificial intelligence (AI) systems that exhibit sophisticated cognitive behaviors without possessing subjective awareness or sentience.