Neuroscience and operations research: a two-way street

Abstract

In 1986, my brother Hubert, a professor of philosophy, and I wrote the book "Mind over Machine: The Power of Human Intuition and Expertise in the Era of the Computer" [Dreyfus and Dreyfus 1986] in which we argued that the brain produces skillful coping behavior in familiar types of situations by using involved intuition rather than by detached thinking. By "thinking" we meant the kind of reasoning, symbol-manipulating, rule-following, theorybased procedures, etc. that we are consciously aware of as we face novel problems. Our primary goal was to argue that the belief held by most researchers designing expert systems at that time — that experts use reasoning and rules — was misguided. Our argument was phenomenological, meaning based on careful observation of both novice and expert naturalistic behavior. In chapter 6 on Managerial Art and Management Science, we applauded the construction of O.R. models of structured domains such as inventory control or queueing phenomena and of novel situations, but we questioned the advisability of developing models of unstructured situations such as business managerial or public policy issues that are based on the interrogation of experts about what they considered the important facts describing the situation (state variables), the rules by which they would change over time given decisions (dynamics), and a measure of quality of the resulting sequence of events (criterion). We also questioned the use in familiar types of situations of decision analysis requiring that experts furnish probabilities of events and utilities of skeletally described outcomes. We could, however, offer no convincing refutation of the belief prevalent in artificial intelligence research and implicitly held in operations research that, while experienced experts in familiar types of situations make intuitive decisions rapidly and effortlessly, they must be doing so by unconscious thinking, presumably based on shortcuts and rule compilations acquired during their experience. With trepidation we offered the conjecture that the intuitive brain may store a large repertoire of remembered situations that had been successfully handled in the past, and may somehow access one similar to the current situation and then use that information to produce its decisions. By 1988, when a paperback version of our book was published, we had learned enough about neural networks to renounce our separately remembered situation (i) view in favor of synaptic-based pattern discrimination and association, but we in no way anticipated the neuroscientific events described below. While this explanation of intuition survives today [Klein 2003], modern behavioral neuroscience is finding otherwise [Dreyfus 2004], and operations research has played a fundamental role in this conclusion.