Andrew Clark

  The paper considers the problems involved in getting neural networks to learn about highly structured task domains. A central problem concerns the tendency of networks to learn only a set of shallow (non-generalizable) representations for the task, i.e., to miss the deep organizing features of the domain. Various solutions are examined, including task specific network configuration and incremental learning. The latter strategy is the more attractive, since it holds out the promise of a task-independent solution to the problem. Once we see exactly how the solution works, however, it becomes clear that it is limited to a special class of cases in which (1) statistically driven undersampling is (luckily) equivalent to task decomposition, and (2) the dangers of unlearning are somehow being minimized. The technique is suggestive nonetheless, for a variety of developmental factors may yield the functional equivalent of both statistical AND informed undersampling in early learning

Strange inversions occur when things work in ways that turn received wisdom upside down. Hume offered a strangely inverted story about causation, and Darwin, about apparent design. Dennett suggests that a strange inversion also occurs when we project our own reactive complexes outward, painting our world with elusive properties like cuteness, sweetness, blueness, sexiness, funniness, and more. Such properties strike us as experiential causes, but they are really effects—a kind of shorthand for whole sets of reactive dispositions rooted in the nuts and bolts of human information processing. Understanding the nature and origins of that strange inversion, Dennett believes, is thus key to understanding the nature and origins of human experience itself. This paper examines this claim, paying special attention to recent formulations that link that strange inversion to the emerging vision of the brain as a Bayesian estimator, constantly seeking to predict the unfolding sensory barrage.

How do intelligent agents spawn and exploit integrated processing regimes spanning brain, body, and world? The answer may lie in the ability of the biological brain to select actions and policies in the light of counterfactual predictions – predictions about what kinds of futures will result if such-and-such actions are launched. Appeals to the minimization of ‘counterfactual prediction errors’ (the ones that would result under various scenarios) already play a leading role in attempts to apply the basic toolkit of the neurocomputational theory known as ‘predictive processing’ to higher cognitive functions such as policy selection and planning. In this paper, I show that this also leads naturally to the discovery and use of extended processing regimes defined across heterogeneous mixtures of biological and non-biological resource. This solves a long-standing puzzle concerning the ‘recruitment’ of the right non-neural processing resources at the right time. It reveals how (and why) human brains spawn and maintain extended human minds.

Predictive processing has begun to offer new insights into the nature of conscious experience—but the link is not straightforward. A wide variety of systems may be described as predictive machines, raising the question: what differentiates those for which it makes sense to talk about conscious experience? One possible answer lies in the involvement of a higher-order form of prediction error, termed expected free energy. In this paper we explore under what conditions the minimization of this new quantity might underpin conscious experience. Our suggestion is that the minimisation of Expected Free Energy is not in itself sufficient for the occurrence of conscious experience. Instead, it is relevant only insofar as it helps deliver what Ward et al. ( 2011 ) have previously described as a _sense of our own poise over an action space_. Perceptual experience, we will argue, is nothing other than the process that puts current actions in contact with goals and intentions, enabling some creatures to know the space of options that their current situation makes available. This proposal fits with recent work suggesting a deep link between conscious contents and contents computed at an ‘intermediate’ level of processing, apt for controlling action.

Abstract The paper considers two recent accounts of the difference between human and animal thought. One deflationary account, due to Daniel Dennett, insists that the only real difference lies in our ability to use words and sentences to give artificial precision and determinacy to our mental contents. The other, due to Paul Smolensky, conjectures that we at times deploy a special purpose device (the Conscious Rule Interpreter) whose task is to deal with public, symbolically coded data and commands. Both these accounts make a crucial error. They offer what is in effect an extra top?level processor to soothe our realist/classical prejudices. But in each case the extra ingredient turns out to be explanatorily hollow. Appealing to language use and language processing alone mistakes a cognitive effect for a cognitive cause. I argue instead that we need to seek a more profound architectural condition which may ground our conscious linguistic abilities but also explains a variety of deeper facts. I sketch a picture which seems to meet those needs and draw out its implications for the debates about belief and about classical Artificial Intelligence

Abstract Cognitive Science, it is argued, comprises two distinct projects. One is an Engineering project whose goal is understanding the in?the?head computational activities which ground intelligent behaviour. The other is a Descriptive project whose goal is the mapping of relations between thoughts as ascribed using the (sentential) apparatus of the propositional attitudes. Some theorists (e.g. Fodor, 1987) insist that the two projects are (in a sense to be explained) deeply related. This view is contested, and the consequences of its abandonment examined. Such consequences are seen to include (i) the irrelevance of scientific arguments for Eliminative Materialism, (ii) a view concerning the proper roles of classical and connectionist work in Artificial Intelligence and (iii) the failure of an allegedly damning argument against connectionism (the so?called systematicity argument)

Summarizes and illuminates two decades of research Gathering important papers by both philosophers and scientists, this collection illuminates the central themes that have arisen during the last two decades of work on the conceptual foundations of artificial intelligence and cognitive science. Each volume begins with a comprehensive introduction that places the coverage in a broader perspective and links it with material in the companion volumes. The collection is of interest in many disciplines including computer science, linguistics, biology, information science, psychology, neuroscience, iconography, and philosophy. Examines initial efforts and the latest controversies The topics covered range from the bedrock assumptions of the computational approach to understanding the mind, to the more recent debates concerning cognitive architectures, all the way to the latest developments in robotics, artificial life, and dynamical systems theory. The collection first examines the lineageof major research programs, beginning with the basic idea of machine intelligence itself, then focuses on specific aspects of thought and intelligence, highlighting the much-discussed issue of consciousness, the equally important, but less densely researched issue of emotional response, and the more traditionally philosophical topic of language and meaning. Provides a gamut of perspectives The editors have included several articles that challenge crucial elements of the familiar research program of cognitive science, as well as important writings whose previous circulation has been limited. Within each volume the papers are organized to reflect a variety of research programs and issues. The substantive introductions that accompany each volume further organize the material and provide readers with a working sense of the issues and the connection between articles.

[Andy Clark] What is the relation between perceptual experience and the suite of sensorimotor skills that enable us to act in the very world we perceive? The relation, according to 'sensorimotor models' is tight indeed. Perceptual experience, on these accounts, is enacted via skilled sensorimotor activity, and gains its content and character courtesy of our knowledge of the relations between movement and sensory stimulation. I shall argue that this formulation is too extreme, and that it fails to accommodate the substantial firewalls, dis-integrations, and special-purpose streamings that form the massed strata of human cognition. In particular, such strong sensorimotor models threaten to obscure the computationally potent insensitivity of key information-processing events to the full subtleties of embodied cycles of sensing and moving. /// [Naomi Eilan] The strong sensorimotor account of perception gives self-induced movements two constitutive roles in explaining visual consciousness. The first says that self-induced movements are vehicles of visual awareness, and for this reason consciousness 'does not happen in the brain only'. The second says that the phenomenal nature of visual experiences is consists in the action-directing content of vision. In response I suggest, first, that the sense in which visual awareness is active should be explained by appeal to the role of attention in visual consciousness, rather than self-induced movements; and second, that the sense in which perceptual consciousness does not happen in the brain only should be explained by appeal to the relational nature of perceptual consciousness, appeal to which also shows why links with action cannot exhaust phenomenal content.