An excerpt from Massimo Piattelli-Palmarini’s book (which I strongly recommend) “Inevitable Illusions: How Mistakes of Reason Rule Our Mind”:
The seeds of doubt about this “ideal” and the legitimacy of classical theory were sown when it began to be seen that the supposedly rational subjects-those who would react rationally to decision making-were in fact not only quantitatively but also qualitatively different from real subjects. What brought this home in striking fashion was the massive development of computational equipment in the decision-making process. A single famous example makes this plain.
In 1957, L. B. Lusted, a clinical researcher at the National Institute of Health, and R. S. Ledley, a dentist at the National Bureau of Standards, sought to automate on computers, and thus improve, the decision-making process by which doctors, with clinical data at their disposal, made their diagnoses. Their approach was about as classical as you could find. They based themselves on classical logic in the strictest of ways (that is, on the so-called tables of logical functions, such as negation, con junction, disjunction, and the conditional “if . . . then”) and on a hierarchy of hypotheses and subhypotheses depicted in flow charts, the arrows or “directions” of which logically linked, in a connected graph of increasing details, the probabilistically weighted hypotheses. They almost immediately realized that these automated diagnoses of theirs gave rather different results from those that the best clinicians might have made from the same data fed to the computer. The discrepancy between real subjects and ideal subjects in this classical case of ‘judgment under uncertainty” emerged with dramatic force. The dilemma that surfaced was whether it would be more rational to follow the conclusions of the best clinical minds or those of the computer.
A few years earlier, in 1954, the University of Minnesota psychologist Paul E. Meehl had published an explosive article, “Clinical versus Statistical Prediction: A Theoretical Analysis and a Review of the Evidence.” Meehl’s thesis was that statistical prediction was more reliable than intuitive prediction, even that of the best doctors. According to Meehl’s data (which were based on an impressive body of research), the results of psychological tests analyzed by the computer were better able to predict outcomes (e.g., who might give up his studies or quit college, who might make a good pilot, who might fall back into crime, who might attempt suicide) than the personal judgments of eminent professional psychologists.