Nicholas Shea

Experimental psychology often relies on a distinction between top-down and bottom-up effects. This distinction is problematic because top-down effects are poorly defined. Specifically, top-down effects are defined as effects of previously stored information on processing of current input, which is far too broad since it includes dispositions to transition from some types of representational states to others, which are implicit in the operation of any psychological process. This chapter suggests a way to distinguish information stored in that way from the kind of influence of prior information that psychologists are concerned to classify as a top-down effect. The distinction drawn here illuminates discussions of the cognitive penetration of perception, as well as discussions of the theoretical usefulness of a perception-cognition distinction.

Commentary on BBS target article: S. M. Fleming & M. Michel, ‘Sensory Horizons and the Functions of Conscious Vision’, Behavioural and Brain Sciences (forthcoming) The authors make a timely argument that the temporal profile of consciousness is an under-exploited constraint on theories of consciousness. However, the exact timing matters for the evolutionary hypothesis. A rival hypothesis is that in aquatic environments consciousness already had the function of representing distal objects, with a time-delay to consciousness in the 100-200ms range.

When we collaborate with others to tackle novel problems, we anticipate how they will perform their part of the task to coordinate behavior effectively. We might estimate how well someone else will perform by extrapolating from estimates of how well we ourselves would perform. This account predicts that our metacognitive model should make accurate predictions when projected onto people as good as, or worse than, us but not on those whose abilities exceed our own.We demonstrate just such a pattern and that it leads to worse coordination when working with people more skilled than ourselves. Metacognitive projection is associated with a specific activity pattern in anterior lateral prefrontal cortex (alPFC47). Manipulation of alPFC47 activity altered metacognitive projection and impaired interpersonal social coordination. By contrast, monitoring of other individuals’ observable performance and outcomes is associated with a distinct pattern of activity in the posterior temporal parietal junction (TPJp).

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.

Our hope and aim was to provoke debate and research on the hypothesis that conscious experiences form quality spaces [1], so we were very pleased to receive letters from Dołęga, Mentec and Cleeremans [2] and Song [3] making constructive suggestions for taking this enquiry in new directions. Our focus was on how various computational theories of consciousness can accommodate the quality space hypothesis. Dołęga et al. make the helpful observation that this should also be investigated diachronically – both developmentally, and during learning. Song points to valuable work that expands our methods for investigating quality spaces. She also argues that non-conventional approaches, like Integrated Information Theory (IIT), will be needed to account for the qualitative character of consciousness. Work on IIT has been valuable in pointing to the importance of the phenomenon we grapple with in our paper – the potential interconnectedness of conscious experiences – which other theories of consciousness have tended to overlook. However, we respectfully disagree that tackling the phenomenon necessitates a non-conventional scientific approach.

In the target article, Christie, Brusse, et al. argue that selected effect functions do not, in general, explain why a trait exists in a population and, therefore, theories of representational content should not rely on selected effect functions. This response focuses on the claim about functions-for-representation. The role of evolutionary functions in a theory of content is to pick out outcomes that have been systematically stabilized by natural selection. Correctness conditions are conditions involved in explaining how that happened. Selected effect functions can play that role in the complex equilibria that Christie, Brusse, et al. identify. Non-equilibrium cases are also discussed.

Research on concepts has concentrated on the way people apply concepts online, when presented with a stimulus. Just as important, however, is the use of concepts offline, when planning what to do or thinking about what is the case. There is strong evidence that inferences driven by conceptual thought draw heavily on special-purpose resources: sensory, motoric, affective, and evaluative. At the same time, concepts afford general-purpose recombination and support domain-general reasoning processes—phenomena that have long been the focus of philosophers. There is a growing consensus that a theory of concepts must encompass both kinds of process. This book shows how concepts are able to act as an interface between general-purpose reasoning and special-purpose systems. Concept-driven thinking can take advantage of the complementary costs and benefits of each. The book lays out an empirically-based account of the different ways in which thinking with concepts takes us to new conclusions and underpins planning, decision-making, and action. It also spells out three useful implications of the account. First, it allows us to reconstruct the commonplace idea that thinking draws on the meaning of a concept. Second, it offers an insight into how human cognition avoids the frame problem and the complementary, less discussed, ‘if-then problem’ for nested processing dispositions. Third, it shows that metacognition can apply to concepts and concept-driven thinking in various ways. The framework developed in the book elucidates what it is that makes concept-driven thinking an especially powerful cognitive resource.

The quality space hypothesis about conscious experience proposes that conscious sensory states are experienced in relation to other possible sensory states. For instance, the colour red is experienced as being more like orange, and less like green or blue. Recent empirical findings suggest that subjective similarity space can be explained in terms of similarities in neural activation patterns. Here, we consider how localist, workspace, and higher-order theories of consciousness can accommodate claims about the qualitative character of experience and functionally support a quality space. We review existing empirical evidence for each of these positions, and highlight novel experimental tools, such as altering local activation spaces via brain stimulation or behavioural training, that can distinguish these accounts.

A reasoning process is more than an unfolding causal chain. Although some thoughts cause others in virtue of their contents, paradigmatic cases of personal-level inference involve something more, some appreciation that the conclusion follows from the premises. Both first-order processes and second-order beliefs have proven problematic or inadequate to account for the phenomenon. Thus, here I argue for an intermediate position, according to which an epistemic feeling, a form of procedural metacognition, plays this role. Extensive psychological research has shown that epistemic feelings are involved in monitoring many kinds of cognitive process, affecting how the processes unfold. Inferences may be no different. Inferences are also plausibly accompanied by an epistemic feeling, in particular a feeling of reliability or unreliability. Such a feeling accounts for the phenomenological datum. It can also play a significant epistemic role for the thinker.

Peter Godfrey-Smith recently introduced the idea of representational ‘organization’. When a collection of representations form an organized family, similar representational vehicles carry similar contents. For example, where neural firing rate represents numerosity (an analogue magnitude representation), similar firing rates represent similar numbers of items. Organization has been elided with structural representation, but the two are in fact distinct. An under-appreciated merit of representational organization is the way it facilitates computational processing. Representations from different organized families can interact, for example to perform addition. Their being organized allows them to implement a useful computation. Many of the cases where organization has seemed significant, but which fall short of structural representation, are cases where representational organization underpins a computationally useful processing structure.

A basic deep neural network (DNN) is trained to exhibit a large set of input–output dispositions. While being a good model of the way humans perform some tasks automatically, without deliberative reasoning, more is needed to approach human‐like artificial intelligence. Analysing recent additions brings to light a distinction between two fundamentally different styles of computation: content‐specific and non‐content‐specific computation (as first defined here). For example, deep episodic RL networks draw on both. So does human conceptual reasoning. Combining the two takes advantage of the complementary costs and benefits of each. It also offers a better model of human cognitive competence.

The question of whether non-human animals are conscious is of fundamental importance. There are already good reasons to think that many are, based on evolutionary continuity and other considerations. However, the hypothesis is notoriously resistant to direct empirical test. Numerous studies have shown behaviour in animals analogous to consciously-produced human behaviour. Fewer probe whether the same mechanisms are in use. One promising line of evidence about consciousness in other animals derives from experiments on metamemory. A study by Hampton (Proc Natl Acad Sci USA 98(9):5359–5362, 2001 ) suggests that at least one rhesus macaque can use metamemory to predict whether it would itself succeed on a delayed matching-to-sample task. Since it is not plausible that mere meta-representation requires consciousness, Hampton’s study invites an important question: what kind of metamemory is good evidence for consciousness? This paper argues that if it were found that an animal had a memory trace which allowed it to use information about a past perceptual stimulus to inform a range of different behaviours, that would indeed be good evidence that the animal was conscious. That functional characterisation can be tested by investigating whether successful performance on one metamemory task transfers to a range of new tasks. The paper goes on to argue that thinking about animal consciousness in this way helps in formulating a more precise functional characterisation of the mechanisms of conscious awareness.

Ruth Millikan has hypothesised that human cognition contains ‘consistency testers’. Consistency testers check whether different judgements a thinker makes about the same subject matter agree or conflict. Millikan’s suggestion is that, where the same concept has been applied to the world via two routes, and the two judgements that result are found to be inconsistent, that makes the thinker less inclined to apply those concepts in those ways in the future. If human cognition does indeed include such a capacity, its operation will be an important determinant of how people use concepts. It will have a major impact on which concepts they deploy and which means of application (conceptions) they rely on. Since consistency testers are a selection mechanism at the heart of conceptual thinking, they would be crucial to understanding how concepts are selected – why some are retained and proliferate and others die out. Hence, whether consistency testers for concepts exist, and how they operate, is an important question for those seeking to understand the cultural evolution of concepts, and of the words we use to express them.

The human mind is extraordinary in its ability not merely to respond to events as they unfold but also to adapt its own operation in pursuit of its agenda. This ‘cognitive control’ can be achieved through simple interactions among sensorimotor processes, and through interactions in which one sensorimotor process represents a property of another in an implicit, unconscious way. So why does the human mind also represent properties of cognitive processes in an explicit way, enabling us to think and say ‘I’m sure’ or ‘I’m doubtful’? We suggest that ‘system 2 metacognition’ is for supra-personal cognitive control. It allows metacognitive information to be broadcast, and thereby to coordinate the sensorimotor systems of two or more agents involved in a shared task.

Metacognition – the ability to represent, monitor and control ongoing cognitive processes – helps us perform many tasks, both when acting alone and when working with others. While metacognition is adaptive, and found in other animals, we should not assume that all human forms of metacognition are gene-based adaptations. Instead, some forms may have a social origin, including the discrimination, interpretation, and broadcasting of metacognitive representations. There is evidence that each of these abilities depends on cultural learning and therefore that cultural selectionmight shape humanmetacognition. The cultural origins hypothesis is a plausible and testable alternative that directs us towards a substantial new programme of research.

Research on concepts has focused on categorization. Categorization starts with a stimulus. Equally important are episodes that start with a thought. We engage in thinking to draw out new consequences from stored information, or to work out how to act. Each of the concepts out of which thought is constructed provides access to a large body of stored information. Access is not always just a matter of retrieving a stored belief (semantic memory). Often it depends on running a simulation. Simulation allows conceptual thought to draw on information in special-purpose systems, information stored in special-purpose computational dispositions and special-purpose representational structures. While the utility of simulation, prospection or imagination is widely appreciated, the role of concepts in the process is not well understood. This paper turns to cognitive and computational neuroscience for a model of how simulations enable thinkers to reach novel conclusions. Carried over to conceptual thought, the model suggests that concepts are ‘plug & play’ devices. The distinctive power of thought-driven simulation derives from the ability of concepts to plug into two kinds of structure at once: the combinatorial structure of a thought at one end and special-purpose structural representations at the other. This article is part of the theme issue ‘Concepts in interaction: social engagement and inner experiences’.

The ability of the mind to conceptualize what is not present is essential. It allows us to reason counterfactually about what might have happened had events unfolded differently or had another course of action been taken. It allows us to think about what might happen – to perform 'Gedankenexperimente' (thought experiments) – before we act. However, the cognitive and neural mechanisms mediating this ability are poorly understood. We suggest that the frontopolar cortex (FPC) keeps track of and evaluates alternative choices (what we might have done), whereas the anterior lateral prefrontal cortex (alPFC) compares simulations of possible future scenarios (what we might do) and evaluates their reward values. Together, these brain regions support the construction of suppositional scenarios.

A recent study has established that thinkers reliably engage in epistemic appraisals of concepts of natural categories. Here, five studies are reported which investigated the effects of different manipulations of category learning context on appraisal of the concepts learnt. It was predicted that dimensions of concept appraisal could be affected by manipulating either procedural factors or declarative factors. While known effects of these manipulations on metacognitive judgements such as category learning judgements and confidence at test were replicated, procedural factors had no reliable effects on the dimensions of concept appraisal. Effects of declarative manipulations on some forms of concept appraisal were observed.

It is a rare privilege to have such eminent and insightful reviewers. Their kind words about the book are much appreciated – perhaps more than they realise. And I'm grateful to all three for having read the book so constructively. Each has given me several things to think about. In the space available here I will focus on the objections that seem most critical. Robert Rupert argues that I rely on an overly narrow understanding of what the cognitive sciences explain (x1). Elisabeth Camp presses me on what precisely it takes to qualify as a structural representation and raises questions about holism (x2). John Krakauer makes a fundamental objection to positing representations when they seem not to be needed to explain behaviour (x3). Rupert has also provided a useful introduction to the book, so I will jump straight in with my replies.

We have various everyday measures for identifying the presence of consciousness, such as the capacity for verbal report and the intentional control of behavior. However, there are many contexts in which these measures are difficult to apply, and even when they can be applied one might have doubts as to their validity in determining the presence/absence of consciousness. Everyday measures for identifying consciousness are particularly problematic when it comes to ‘challenging cases’—human infants, people with brain damage, nonhuman animals, and AI systems. There is a pressing need to identify measures of consciousness that can be applied to challenging cases. This paper explores one of the most promising strategies for identifying and validating such measures—the natural-kind strategy. The paper is in two broad parts. Part I introduces the natural-kind strategy, and contrasts it with other influential approaches in the field. Part II considers a number of objections to the approach, arguing that none succeeds.

This paper reports the first empirical investigation of the hypothesis that epistemic appraisals form part of the structure of concepts. To date, studies of concepts have focused on the way concepts encode properties of objects and the way those features are used in categorization and in other cognitive tasks. Philosophical considerations show the importance of also considering how a thinker assesses the epistemic value of beliefs and other cognitive resources and, in particular, concepts. We demonstrate that there are multiple, reliably judged, dimensions of epistemic appraisal of concepts. Four of these dimensions are accounted for by a common underlying factor capturing how well people believe they understand a concept. Further studies show how dimensions of concept appraisal relate to other aspects of concepts. First, they relate directly to the hierarchical organization of concepts, reflecting the increase in specificity from superordinate to basic and subordinate levels. Second, they predict inductive choices in category-based induction. Our results suggest that epistemic appraisals of concepts form a psychologically important yet previously overlooked aspect of the structure of concepts. These findings will be important in understanding why individuals sometimes abandon and replace certain concepts; why social groups do so, for example, during a “scientific revolution”; and how we can facilitate such changes when we engage in deliberate “conceptual engineering” for epistemic, social, and political purposes.

In their constructive reviews, Frances Egan, Randy Gallistel and Steven Gross have raised some important problems for the account of content advanced by Nicholas Shea in Representation in Cognitive Science. Here the author addresses their main challenges. Egan argues that the account includes an unrecognised pragmatic element; and that it makes contents explanatorily otiose. Gallistel raises questions about homomorphism and correlational information. Gross puts the account to work to resolve a dispute about probabilistic contents in perception, but argues that a question remains about whether probabilities are found in the content or instead in the manner of representation.

There has been little investigation to date of the way metacognition is involved in conceptual change. It has been recognised that analytic metacognition is important to the way older children acquire more sophisticated scientific and mathematical concepts at school. But there has been barely any examination of the role of metacognition in earlier stages of concept acquisition, at the ages that have been the major focus of the developmental psychology of concepts. The growing evidence that even young children have a capacity for procedural metacognition raises the question of whether and how these abilities are involved in conceptual development. More specifically, are there developmental changes in metacognitive abilities that have a wholescale effect on the way children acquire new concepts and replace existing concepts? We show that there is already evidence of at least one plausible example of such a link and argue that these connections deserve to be investigated systematically.