Social psychology is the study of how people’s cognitive states or behaviors are influenced by the presence of others. Its methods are predominantly experimental, but as with any experimental science, the meaning of empirical results depends on the underlying theory that the experiments are meant to test. The more expressive and formal the underlying theory, the more insight one can gain from experiments.
Van Overwalle offers a general approach to constructing theories in social psychology and testing them against experimental evidence. The formalism in which he constructs his theories is a connectionist model (a computational neural network). These networks are inspired by the neural basis of natural cognition and are widely used to solve practical problems in pattern recognition and machine intelligence. Van Overwalle argues that since their mechanisms resemble those involving human neurons, they offer an attractive framework for modeling and studying cognitive processes. The claim should be taken cautiously: the actual dynamics of real neural networks are frequency-based and far more complex than the amplitude-based dynamics of a connectionist network with nonlinear threshold functions, and there is abundant evidence that, in natural networks, concepts take the form not of individual nodes (as in Van Overwalle’s localist nets), but of patterns of activation among many neurons that individually have no semantics. Nevertheless, the quantitative nature of connectionist networks is more realistic than logic-based symbolic formalisms (a feature that it shares with a few other formalisms, such as dynamic field theory), and it offers a formal way to construct falsifiable models.
The book is organized to encourage readers to construct and experiment with connectionist representations of classical models in social psychology. Van Overwalle has produced a program (FIT) for constructing and running connectionist networks; it is freely available on the Web. The appendix of the book is a detailed manual to the FIT system, and each chapter provides exercises based on FIT.
The three chapters in Part 1 provide an introduction to connectionism in general and feedforward and recurrent networks in particular, describing how the delta learning rule can be used to fit a network to experimental data.
The balance of the book (about 360 pages) presents a series of connectionist models of phenomena of interest to social psychologists: three models of causal attribution, four of the formation of impressions of groups and individuals, and one on attitude formation and change. One of these models (on the fundamental attribution bias) is previously unpublished, but the others are adapted from journal publications by Van Overwalle and his colleagues. Each chapter shows how a particular phenomenon can be modeled in a connectionist network, and compares results from classical psychological experiments with those from the model. Each has been extended with a set of exercises in FIT that allow the reader to replicate the original modeling work and extend it in various ways. By working through these exercises, the reader not only gains an understanding of connectionist modeling, but is also exposed to a number of central results in intrapersonal social psychology.
This volume would be invaluable in support of a laboratory component of a social psychology course. Because the text is drawn from published articles, it presumes a basic competence in professional vocabulary and thought idioms, making it difficult for a new student to follow; however, it would be suitable for self-study if accompanied with an introductory textbook, and it would certainly be valuable for students in a classroom setting where the teacher can introduce each chapter’s concepts. The book does not discuss interpersonal concepts--in every case, the connectionist model is of what happens inside one person’s head. It would be very useful to extend it with exercises using multiple networks to model how different cognitive beings interact with one another.
