Cognitive Architectures

Abstract

This special issue on Cognitive Architectures is a collaboration between the Neuron editorial team and Stanislas Dehaene (Collège de France) and Yadin Dudai (Weizmann Institute, Israel). The seed of the idea for this issue originally took form while some of the authors in this series collaborated during the initial phase of the European Flagship Project, the “Human Brain Project.” These early discussions centered on the critical importance of systems and cognitive neuroscience approaches for truly being able to understand and ultimately model brain function. As the idea of organizing a special issue around this theme evolved, we broadened the circle of participation and invited a stellar group of international researchers to contribute to this collection. Cognitive architecture specifies the infrastructure for an intelligent system: the set of internal representations, algorithms, and hardware choices that allow it to operate. The architecture should preferably include those aspects of the cognitive agent that remain mostly stable over time. In theory, cognitive architecture is agnostic to the hardware in which it is implemented (thus making contact with the fields of artificial intelligence and computer science). In brain research, however, it is absolutely not. Understanding the biological hardware supporting a given cognitive architecture is a key step, and so is understanding how development leads to the mature functional system. Brains have been shaped over eons by selective pressures in which environments, capacities, biological constraints, and chance events interacted. As a result, the biological substrate, its structure, in situ activity, phylogenetic history, and ontogenetic unfolding are all critical for understanding the brain’s computational goals as well as the operation and implementation of the algorithms subserving these goals. A reverse-engineering approach—starting from the function, analyzing it into its component parts, and then finding how these map onto brain circuits and neurons—is what unites the contributions to this special issue. Using otherwise different, though complementary, approaches, they all attempt to elucidate aspects of the cognitive architecture of the brain. Such attempts are not an easy task to start with, and not merely because of the need to bridge multiple levels of analysis in iterations of top-down and bottom-up expeditions, but also because this type of research raises fundamental issues concerning the way our brain understands itself. For example, when we identify a process or candidate computational goal that we deem reflected in a certain spatiotemporal signature of brain activity or in behavior, is this a natural kind, or only our anthropomorphic taxonomy? What are the critical levels of analysis for a given architecture and the relevant inter-level translation rules? Is there a shared “language” of neurophysiology that supports our different cognitive abilities and allows them to integrate into fluent behavior? These questions clearly exceed the scope of this foreword, but they do seem to emerge as a subtext in this special issue. The original aim of this special issue was to provide a survey of selected core areas in cognitive neuroscience and, for each of these areas, to critically review the key principles and experimental data at the behavioral, neural, and network levels that need to be considered and incorporated into any realistic theoretical model of the brain. We hope that these Perspectives will not only provide a useful summary for theoreticians and cognitive neuroscientists, but also that they will be interesting to readers who may be coming from other areas and disciplines. We intentionally aimed the scope of the issue to cover a broad range of cognitive domains, from visual perception, memory, decision making, language, navigation, learning, cognitive development, to the social brain. Each Perspective provides an overview of a particular cognitive domain, with the aim of illustrating not only what we know and where recent advances have shed light, but also to highlight what we do not yet know and where we are still in the dark. To add to the educational element, each Perspective contains two Boxes highlighting key advances and current status of the field, as well as key unknowns and future directions. We do not, of course, claim any originality in assuming that pinpointing what we do not know is at least as important as pinpointing what we think we know. Nevertheless, we hope that these Boxes will provide a useful tool for stimulating discussion and new work in the field. We are indebted to the many authors and reviewers who thoughtfully contributed to this special issue. Cognitive neuroscience is such a rich research area that we could not have possibly covered the full spectrum of this exciting field in one issue of Neuron. We hope you enjoy these articles and that you share our eager anticipation for the future of this exciting research area. On a final note, this special issue will be featured at the Society for Neuroscience annual meeting in Chicago, Illinois. We invite you to stop by the Cell Press/Elsevier booth (#130) to pick up your free copy.