Are there ‘basic’ tastes and, if so, how many are there? While, to date, this question has mostly been addressed by sensory scientists, it would seem ripe for contemporary philosophical consideration (i.e., after Plato and Aristotle’s early discussion of the matter). Consider only the fact that the majority of those scientists who have written on the subject appear unable to make up their minds as to whether or not umami should be classed as a basic taste, alongside sweet, bitter, sour, and salty. The emergence of various other taste qualities in recent years, such as fatty acid (or ‘oleogustus’) and kokumi, has led (at least one group of) researchers to add another category, namely ‘alimentary’ tastes alongside the so-called basic tastes. Several different lines of evidence have been used to support the existence of basic tastes, including phenomenological (i.e., introspection or perception), cross-cultural linguistic/anthropological, neurophysiological, and neuroimaging data. Perhaps unsurprisingly, though, the answers provided sometimes differ. In this narrative historical review, we critically evaluate the evidence focusing, in particular, on the problematic case of umami, as well as investigating whether an embodied account of basic taste can be supported. Based on the reviewed evidence, we argue for a three-tiered interpretation of ‘basic’, ‘possibly basic’, and ‘ancillary taste qualities’.
In this article I respond to the commentaries written by Adams and Browning, Constantinou et al, Drayson, Hinrichs, Momennejad, Nemati, and Williams on The Brain Abstracted. I divide my responses into three broad themes: 1) Epistemology of science, 2) Metaphysical concerns, and 3) The disciplinary relationships – science, technology and philosophy.
Vehicles, the carriers of representational content, are an important, if somewhat undertheorized, posit in cognitive science. In this paper we argue that generating and maintaining representational vehicles is not a trivial problem, even more clearly so when we are dealing with cognition in complex systems such as brains, or brain-body-environment aggregates. We discuss various vehicle-building operations and strategies that can be applied to complex systems (such as underutilizing and enlarging encoding space), all of them connected to what information theorists call channel coding. We then show how these strategies map onto various prominent vehicle candidates in neuroscience.
This paper defends a stuff theory of the chemical senses: smell and taste, unlike vision, touch, and hearing, are directly oriented toward stuffs rather than individual objects. I first argue that stuff constitutes an irreducible ontological category — distinct from both individuals and universals. I then contend, drawing on recent anti-reductionist work in the philosophy of chemistry, that chemistry itself is best understood as a science of stuffs rather than of atoms or molecules. Finally, building on my earlier work (Mizrahi 2014), I show that the distinctive phenomenology of smell and taste — including mixture, concentration, and the sense of presence — is best explained by a stuff theory of their proper objects. The paper concludes by considering the implications of this view for the long-standing question of how to individuate the senses of smell and taste.
Philosophers often discuss zombies and inverts, hypothetical cases in which perfect physical/functional duplicate of us either lack conscious experience or else have their experiences “shuffled” such that they correlate with different stimuli and physiological states. The idea that there may also be physical/functional duplicates of us who have more kinds of experiences than we do, however, has gone unnoticed. These phenomenal smugglers, as I call them, raise novel questions on what sorts of restrictions there may be on how many or what sorts of extra experiences our smuggler counterparts could conceivably have. I appeal to the dimensional structure, the precision structure, the combinatorial structure, the syntactic structure, and the integration structure of consciousness in order to clarify and answer the questions raised by smugglers about the conceivable upper bounds of consciousness.
Vision has a dual life or possibly more. On the one hand, vision describes the environment as being a certain way. On the other hand, vision guides the execution of actions. Vision is, thus, Janus-faced: it is tied to both describing properties, this description being related also to spatial properties relevant for action, as well as to guiding the motor execution of that action. Philosophers and cognitive scientists have long studied this dual aspect of vision. However, this raises a fundamental question: how do descriptive functions and guiding functions offered by vision integrate with intentions? The motor acts, within visual guidance, complementing the properties visually described are, indeed, most of the time, intentional motor acts. So far, philosophers have either explored the relationship between vision’s descriptive and guiding roles (its Janus-faced nature) or the link between guidance and intention (the so-called Interface Problem). Here, we propose a framework that unifies these three domains, visual description, visual guidance of action, and intentionality, by offering a comprehensive account that connects the Janus-faced nature of vision with a solution to the Interface Problem. This makes vision more than Janus-faced.
In this paper, I argue that two plausible claims for olfactory experience, (i) that modality is spatially indeterminate regarding the spatial location of what it presents and (ii) that whatever it presents, it is experienced as external to the perceiver’s body, are in conflict with one another. I argue that a satisfactory answer to this puzzle requires us to accept as part of olfactory phenomenology aspects of experience that do not obviously belong to olfactory experience. I claim that this ultimately leads to a modification of the idea that olfactory experience is spatially indeterminate.
This special issue unites original theoretical and empirical research on two topics that are gaining traction in cognitive neuroscience and psychology, but so far have small philosophical footprints: dreaming and waking mind wandering. While the fields of dream and mind wandering research are largely separate, phenomenological and neurophysiological similarities between waking mind wandering and sleep-related experiences suggest that these phenomena are intimately connected. Together, they raise important questions about the nature and functions of spontaneous mental phenomena and their relation to wakefulness and sleep, as well as for theories of attention, action, and consciousness.
The Newman problem is a fundamental problem that threatens to undermine structural assumptions and structural theories throughout philosophy and science. Here, we consider the problem in the context of consciousness science. We introduce and discuss the problem, and explain why it is detrimental not only to structuralist assumptions, but also to theories of consciousness, if left unconsidered. However, we show that if phenomenal spaces, and mathematical structures of conscious experience more generally, are understood in the right way, the Newman problem does not arise. The upshot of this paper is that consciousness science needs to be careful in which definition of structure to consider, but if it is, the Newman problem disappears.
In recent years, more and more papers have critically analyzed the concept of ‘representation’ in neuroscience and concluded that its ambiguity and imprecision constitute serious, perhaps even fatal, flaws (e.g. Baker et al., 2022; Favela & Machery, 2023; Pohl et al., 2025). Here we use the literature on ambiguous concepts in philosophy of science (e.g. Brigandt, 2010; Haueis, 2024; Novick, 2023) to motivate a very different conclusion. We first step back and ask why such an ambiguous concept would be so widely used among scientists. We suggest that it serves a primary epistemic goal of neuroscience: to explain how cognitive capacities arise. We note that it is still unclear what form(s) successful explanations will take, and chart several philosophical and scientific debates on this matter. Because ‘representation’ denotes findings that contribute to such explanations, its ambiguity necessarily reflects the uncertainty currently inherent in the field. Our analysis of ‘representation’ suggests that its ambiguity cannot be resolved through philosophical analysis or substitution with other concepts alone. Instead, more empirical work is needed to help clarify through successful research how to achieve neuroscience’s epistemic goals.
We describe a form of structuralism that focuses on the mereological parts of conscious states, drawing on the metaphor of a symphony where the removal or alteration of any contributor affects the whole in a holistic but determinate way. This variational method is applied to what we call the "minimal experience of self", a composite encompassing (1) a feeling of agency, (2) a feeling of privacy, and (3) a feeling of “me-ness”, each of which partially fuses with our internal thoughts and our bodily sense. We support this view by showing how (1)-(3) can be dissociated in all possible combinations, supporting our analysis with empirical and clinical cases.
An increasing amount of work in AI aims to build computational systems that are not just tools for human users but agents in their own right. In AI, agency is often taken to consist simply in the capacity to pursue and achieve goals. However, different kinds of sophistication in representational processing produce different degrees or varieties of goal-directedness. Realist representational accounts of goal-directedness usually omit or fail to highlight a requirement which is central to instrumentalist representational accounts, namely that there is a certain coherence or unity of purpose amongst the different goals that the system pursues. That is a commitment of Dennett’s Intentional Stance, for example. The same requirement operates when biologists adopt the rational agent heuristic to make sense of the evolved phenotypes of an organism. This paper argues that this requirement should be included when specifying what it is for an AI system to be an agent. If the degree to which an AI system is an agent is captured in terms of what it represents and what computations it performs, mechanisms that help achieve unity of purpose are an important ingredient. One dimension along which an AI system becomes more agentive is increased sophistication in such mechanisms. Conversely, when the objective is to build AI systems that are agents, satisfying this additional requirement will endow machine learning systems with a deeper kind of goal-directedness or agency.
Conscious experiences have many structural features. Consider how your color experiences have dimensions of variation corresponding to hue, saturation, and brightness, how your visual acuity decreases in precision from the center of your visual field to the periphery, how your pain experiences come in different magnitudes, or how your temporal experience seems to flow in a continuous stream. ‘Structuralism’, in the most general sense, may be defined as an approach to consciousness research where the central aim is to investigate the structures of conscious experiences. Given this broad definition, there are many varieties of structuralism. As examples, methodological structuralists think that scientific methods give us knowledge only about structural features of consciousness, while ontic structuralists think that all there is to consciousness is structure. But even those who deny those claims might still think that the proper aim of the science of consciousness is to investigate structure. The aim of this special volume is to bring these dispersed discussions together by organizing a collection of articles about the structures of conscious experiences and the roles that structure ought to play in consciousness research. The goal is to set a foundation and an agenda for a structuralist research program in the science of consciousness.
The rise of artificial intelligence (AI) raises the question of whether we should introduce a new category of representations, next to mental and scientific representations. We argue that AI ‘representations’, in particular of deep neural networks, differ significantly from the mental and scientific representations central to the philosophy of (cognitive) science. These systems lack essential aspects, such as semantic content, the ability to misrepresent, and a clear use condition guiding behavior; it is often unclear why and what they represent. They also lack the capacity to form or identify misrepresentations which makes it impossible to assess their accuracy. Furthermore, AI systems do not satisfy a use condition in the same way as mental and scientific representations. We conclude that, while AI systems can, under certain conditions, be useful tools for scientific discovery, their internal states should not be mistaken for mental and scientific representations.
An adequate theory of representation should distinguish between the structure of a representation and the structure of what it represents. I argue that the simplest sorts of transformers (the architecture that underlies most familiar Large Language Models) have only a very lightweight structure for their representations: insofar as they work with the structure of language, they represent it but do not use it. In addition to being interesting in its own right, this also shows how we may use high-level invariants at the computational level to place constraints on representational formats at the algorithmic level.
Can neural representations be naturalized? Current debates surrounding the naturalization of representation in neuroscience typically characterize this project in terms of one of three options: Methodological Naturalism, Ontological Naturalism, or the belief that both types of naturalism provide support for, and constraints on, each other to drive inquiry. In this paper, I argue that all three of these options are problematic. The two projects of naturalism cannot be pulled apart from one another, nor can one act as effective support/constraint on the other. The relationship between these projects of naturalism is far more complex, nuanced, and interdependent than is typically thought. I highlight how this influences current debates regarding the nature of representation in neuroscientific practice.
Large language models (LLMs) produce seemingly meaningful outputs, yet they are trained on text alone without direct interaction with the world. This leads to a modern variant of the classical symbol grounding problem in AI: can LLMs' internal states and outputs be about extra-linguistic reality, independently of the meaning human interpreters project onto them? We argue that they can. We first distinguish referential grounding—the connection between a representation and its worldly referent—from other forms of grounding and argue it is the only kind essential to solving the problem. We contend that referential grounding is achieved when a system's internal states satisfy two conditions derived from teleosemantic theories of representation: (1) they stand in appropriate causal-informational relations to the world, and (2) they have a history of selection that has endowed them with the function of carrying this information. We argue that LLMs can meet both conditions, even without multimodality or embodiment.
It is common to hear neural states and processes described in representational terms, where it is alleged that neurons function to represent stimuli, categories, motor commands and various other things. Evidence for such a role is typically based upon the ways neurons reliably respond to specific stimuli. In a recent paper, Pohl et. al. (2024) have offered a more formalized account of neural representation evidence that utilizes Shannon’s information theory. Despite the ingenuity of their theory, we argue that Pohl et. al. fail to present a successful account of evidence for neural representation, and indeed, repeat common mistakes that other researchers make in supporting neural representations. We offer a more in-depth and rigorous analysis of such evidence, and argue for a more skeptical conclusion: What commonly gets presented as evidence for neural representations actually isn’t. While we support the reality of cognitive representations at higher levels of analysis, we do not believe compelling evidence has been presented that supports assigning a representational role for neural states and structures.
Philosophical work on representation has largely focused on descriptive content, which concerns what the world is like. Neuroscientists and psychologists regularly make other sorts of content ascriptions, however, including to forward models, motor commands, and error signals. Here I focus on directive states – states that shape what a system does – in relation to the question: what makes a directive state a representation? I begin by recharacterizing Ramsey’s (2007) Job Description Challenge (JDC) so that it applies to directive states. After arguing that existing answers to the JDC do not answer the “directive JDC,” I develop my own account by asking: what is the explanatory payoff of appeal to directive representations? I argue that directive representations help explain how a system achieves a particular outcome in the face of varying internal or external conditions. To be a representation, then, a directive state must make a specific causal contribution to bringing about an outcome in a way that is decouplable from any particular behavior of the system. This account captures both what is distinctive about directive representation and what it has in common with descriptive representation. It also implies that we should be much more sparing in attributing content to directive states.
Philosophers sometimes observe that some scientists believe in representations because those representations explain behavior. Philosophers also sometimes observe that some scientists believe that single-unit (single-cell, single-neuron) firing rates represent environmental features, in part, because of correlations between the environmental features and the single-unit firing rates. Less frequently, philosophers have observed that sometimes scientists combine the results of behavioral experiments and neuroscientific experiments. Recent philosophical work discussing the evidence for representations has not recognized this picture. In addition, this recent philosophical work has undertheorized the abductive reasoning underlying the postulation of representations to explain behavior.
