Reviewed by
Brian Castellani
Kent State University
Can complexity scientists bridge, in the words of C. P. Snow, the two cultures of academia - the humanities and the sciences - to create a more thoroughgoing explanation of human cognition? More specifically, can the tools of hermeneutics, mathematics and computer simulation be integrated to assemble better and more useful models of human social understanding than currently exist? These are the two provocative and ambitious questions - the former the broader, and the latter the more specific - that frame the intent and focus of Klüver and Klüver's recent book, Social Understanding.Some readers will say "no" to both questions. Klüver and Klüver, however, make a convincing case that there is little other choice than "yes." The premise for their conviction can be outlined quickly: Artificial intelligence (AI) has become estranged from its original goal; namely, to model human cognition. AI's estrangement is due, in large measure, to the mathematical formalisms it pursues, such as rational choice, neural nets, computational modeling and game theory, which do not really model, globally speaking, how humans think. They do not model thinking because they fail to address the problem of everyday understanding; that is, how humans, as social creatures, make meaning and interpret their worlds. Everyday understanding is, however, the domain of hermeneutics. Therefore, the models of AI need to be integrated with the tools of hermeneutics if they are to advance our understanding of understanding. Such integration, however, requires an equal mixture of methodological innovation and conceptual clarification. Methodological innovation replaces/augments differential equations, network models and overly complicated algorithms with something simpler: network topology and its according characteristics. Conceptual clarification requires an exegetical tour through the history of AI and hermeneutics, with many stops along the way to clarify, dust off, re-think, limit, discard or integrate the various ideas central to these two traditions. The methodological and conceptual result is what Klüver and Klüver call a hermeneutical AI, principled on the idea that human understanding is not a calculating optimization algorithm or a differential equation; instead, it is fundamentally an interpretive act involving social, symbolic interactions amongst people and their worlds.
To demonstration the utility of their hermeneutical AI, Klüver and Klüver organize their book into three sections. The first section builds for the reader a two-culture (think C. P. Snow) vocabulary, based on a tour of artificial intelligence (chapter 1) and hermeneutics (chapter 2). Chapter 1 is one of the best, concise readings I have seen on AI and its recent extensions into complexity science. Klüver and Klüver do a nice job defining concepts that are often misused or fuzzily explained, such as attractor points, topology of networks, meta-rules. Chapter 2's quick tour through the history of hermeneutics is also well done, primarily because Klüver and Klüver steer clear of 'old-school' or 'post-modern' obsessions with formal philosophical or scientific understanding, opting instead to focus on everyday understanding, the stuff of sociology and social physics.
Having built their vocabulary for the understanding of everyday understanding, the second section (Chapters 3 through 5) engages in a series of exploratory studies, meant to address a variety of sub-topics such as mental models, meaning making and information processing (Chapter 3), and social learning and rule following (Chapter 4). Klüver and Klüver's research strategy for these studies is straightforward enough: take a topic such as rule-following; bracket a small segment of it; generate an empirical question; design a cellular automata or neural net simulation; test the results by exploring the model's topology; decide what worked, what they learned; and, finally, acknowledge limitations and try again. This research strategy is taken to its conclusion in Chapter 5, as Klüver and Klüver explore, from a reverse engineering perspective, how well their hermeneutical AI models fit real-world data.
This is honest research and it is to be commended. The only drawback, however, is that one can get somewhat lost trying to figure out how their various studies come together to generate a larger picture. Klüver and Klüver, however, respond. They are clear from the first page of the book. Their goal was never to offer a unified hermeneutical AI of everyday understanding. Instead, their goal was to test if such a thing is possible by providing "some preliminary results of our research in this direction" (p. 1). I think this was a very smart strategy, because, given the incredible provocation upon which this book is based, their modest approach allowed them to accomplish what, at first, looked impossible. For example, their topological approach, grounded in a variety of neural net models, offers a lot of promise. As such, I came to the end of the book saying, along with Klüver and Klüver, "while the preliminary results are hesitant, tentative and cautious, a hermeneutical AI is possible." For such a difficult accomplishment Klüver and Klüver are to be commended.
Return to Contents
of this issue