Connectionism

see cognitive theory

Connectionism and the Mind: Parallel Processing Dynamics and Evolution in Networks by William Bechtel, Adele Abrahamsen (Blackwell) provides a clear and balanced introduction to connectionist networks and explores their theoretical and philosophical implications. As in the first edition, the first few chapters focus on network architecture and offer an accessible treatment of the equations that govern learning and the propagation of activation, including a glossary for reference. The reader is walked step-by-step through such tasks as memory retrieval and prototype formation. The middle chapters pursue the implications of connectionism's focus on pattern recognition and completion as fundamental to cognition. Some proponents of connectionism have emphasized these functions to the point of rejecting any role for linguistically structured representations and rules, resulting in heated debates with advocates of symbol processing accounts of cognition. The coverage of this controversy has been updated and augmented by a new chapter on modular networks. Finally, three new chapters discuss the relation of connectionism to three emerging research programs: dynamical systems theory, artificial life, and cognitive neuroscience.

Author summary: Connectionism is what happened when certain cognitive scientists began using neural networks as a means of modeling cognition. This was probably the first strong indicator that cognitive science is a shape‑changer that can endure by repeatedly incorporating new ideas and techniques. In the 1970s cognitive science received its name and a distinct identity as the intellectual home of researchers who produced new symbolic models of cognition by blending key developments of the 1950s and 1960s: the cognitive revolution in psychology, the Chomskian revolution in linguistics, and the heady early days of artificial intelligence. By the 1980s what is now called "classic cognitive science" became more diverse in its disciplinary influences (e.g., certain philosophers, anthropologists, and sociologists became involved), but cognitive scientists continued to find unity in their allegiance to core assumptions of the symbolic approach and their commitment to interdisciplinary work. At the same time, a challenge to this unity was mounted by researchers who reached back to a slightly earlier era (the 1940s and 1950s) in which crosstalk between neuroscience and an emerging science of computation had yielded neurally inspired networks of units which achieved computation by propagating activation. The cognitive scientists who revived them in the early 1980s pressed the case that such networks should be embraced as a subsymbolic alternative to symbolic models of cognition, and referred to them as connectionist, neural network, artificial neural network (ANN), or parallel distributed processing (PDP) models.

The first edition of Connectionism and the Mind appeared in 1991, approximately one decade after connectionism made its entrance. Battle for dominance between symbolic and connectionist modeling was at a peak. The first edition provided a primer on the computational basics of networks, then reviewed the battle, and in a final chapter discussed how a variety of disciplines might be affected by the rise of connectionism. Another decade has passed as this second edition goes to press, and the landscape has again been considerably altered. What is now sometimes called "classic connectionism" has become more established within cognitive science and some of its contributing disciplines, but it has faced new external challenges from transdisciplinary trends towards dynamical systems, artificial life research, and cognitive neuroscience. Some network modelers have been inspired to incorporate these  trends in their work, resulting in less classic varieties of connectionism to which we give ample attention to in this new edition.

Connectionism and the Mind (2002), like its predecessor, is written primarily for those who are curious but not yet knowledgeable about connectionism. For individuals who simply want to know what all the fuss is about and to navigate occasional encounters with networks, this book should serve as a one‑stop shopping emporium. We envisage such readers to include advanced undergraduates with above‑average interest in cognitive science, graduate students and faculty whose interests intersect those of connectionists, and individuals outside the academic world who have been hooked by a casual encounter with networks and want to know more. We also have kept in mind those graduate students in the cognitive sciences for whom this book may be among their first entry points into a field they intend to explore in much greater depth and perhaps make their career. The boxes and appendices offer them a little more detail than the text alone, and the cross‑references to specific modeling software as well as the "Sources and Suggested Readings" at the end of each chapter should help guide their next steps. Finally, large portions of the book are relevant to those who want to update knowledge that is no longer current. We worked especially hard at finding the best avenues of explanation and examples of research for such challenging topics as dynamical systems theory. We hope we came close to meeting the goal that every reader find these topics both accessible and exciting.

Both editions are distinctive in the extent to which they cover conceptual issues and philosophical inquiry into the mind while also introducing the nuts and bolts of connectionist modeling. This new edition of Connectionism and the Mind retains and updates the first three chapters of the first edition, in which the stage is set and a variety of connectionist architectures and learning procedures are described (including worked‑out examples of basic computations). As before, the next few chapters focus on theoretical claims and counterclaims. Certain well‑known network models provided the original context for these arguments, but a number of proposals and models from the 1990s have been added in this new edition. Specifically, chapter 4 (on pattern recognition) combines parts of the original chapters 4 and 5 and adds a new network model of logical derivation. Chapter 5 (on using networks rather than rules to perform such tasks as past‑tense acquisition) combines parts of the original chapters 6 and 7 and adds more recent past‑tense simulations. Chapter 6 (on issues of representation) combines parts of the original chapter 6 with a variety of new material. The final four chapters are entirely new. Each focuses on a different context in which network research moved beyond classical connectionism in the 1990s: modular networks and feature maps in chapter 7, dynamical systems in chapter 8, artificial life research in chapter 9, and cognitive neuroscience in chapter 10.

Headline 3

insert content here

WT Main | About WT | Review Links | Contact | Review Sources | Search