“How do you decide when to replace your car?”

In the early 80s, computer scientist Stuart Dreyfus was at RAND working with the US government and big businesses on formal models of optimal decision-making. During those days of the infancy of operations research and digital computers, the average person at a cocktail party had not heard that machines could think, learn, and create. In social situations where Stuart had to explain his work, he favored an example of the decisions behind buying a new car.

Stuart told people they could use a computer to estimate the costs of operating an aging car and the cost of buying a new one; throw in other factors like reliable performance, deprecation, and the pleasure derived from ownership; weigh all those factors appropriately; and let the computer determine the most desirable sequence of decisions to replace.

One night, he was asked if his explanation was how HE comes to make the decision to replace a car.

“Of course not,” Stuart replied without hesitation. “That was only an example of how to use the formal procedure. Buying a new car is much too important to be left to a mathematical model. I mull it over for awhile, and buy a new car when it feels right.”

The next morning Stuart began to reflect. How could he tell generals, businessmen, and policy-makers that they should use a decision-making technique that he himself wouldn’t use in his own personal life?

Hunches and intuitions, and even systematic illusions, are the very core of expert-decision making, so whether one seeks to use a digital computer to model heuristic rules behind actual problem solving, or whether one tries to find optimal algorithms, the result fails to capture the insight and ability of the expert decision maker.

While operations research had successes in modeling operational problems in the military and industry, that is no reason to believe that the same mathematical modeling techniques can tell experienced generals what military strategies are optimal, or business executives whether to diversify their companies.

Problems involving deep understanding built up on the basis of vast experience will not yield – as do simple, well-defined problems that exist in isolation from much of human experience – to formal mathematical or computer analysis.

—- from “Mind Over Machine” by Hubert & Stuart Dreyfus