Friday, Apr. 18, 1969

Decision Theory: Guide to Choice-Making

TO beleaguered man, life is an uphill obstacle course. Between the cradle and the grave stretches an endless multitude of problems, every one of which cries out for decision. At what moment is it safe for the pedestrian to challenge the hazards of wheeled traffic? If a picnic is scheduled for tomorrow, what are the odds against rain? By what mysterious standards does a man pick a friend, a pastime, a profession or a wife?

More often than not, human intelligence seeks to resolve these and a host of similar riddles by intuition alone. And intuition makes an excellent guide. But it is the cardinal premise of a division of social science called decision theory that intuition is often not enough. Decision theory is based on the premise that man's capacity to solve life's problems correctly is limited by two factors: in extremely complex situations, he is not always capable of mastering all the information, and he does not always decide as logic and reason tell him he should. Beyond human intuition, says Psychologist Ward Edwards, lies an individual's personal calculation of the odds in favor or against. This personal factor, which measures the individual's will to win rather than the mathematical probabilities, must be counted into the risk and the odds.

Money Game. Edwards, 42, is head of the University of Michigan's Engineering Psychology Laboratory. Since last July, he has been operating an ingenious gambling experiment called "Stakes & Odds" at the Four Queens Hotel in Las Vegas. With the full encouragement of the house and the Nevada State Gaming Commission, a computer has recorded the decision-making patterns of some 250 volunteers. The game that they are asked to play (with real money) has two parts: in the first, a player must select two bets, one good and one bad, from four that are offered him; in the second, he has the option to keep or get rid of a bet, depending upon how he judges its value to him.

From the computer's carefully recorded data, Edwards has learned that people on the whole make remarkably rational decisions. Nevertheless, more than a third of the participants become befogged by superstitions, biases and logical incoherences. Most people, for example, regard an event as more likely to occur if they stand to lose by its occurrence rather than gain by it. Also, they tend to inflate the value of the money they stand to win--that is, a $10 bet means more to them emotionally than five $2 bets.

Unaffected by such emotional factors, a computer does better at the game than people do--which does not mean that decision theorists have contempt for man. In fact, Edwards has a profound respect for the logical abilities of the human mind. One of the inexplicable wonders of life is that a normal man can, with almost ridiculous ease, solve in an instant problems of theoretically great complexity. Take for example, ticktacktoe. Theoretically, in five moves alone this childishly simple game can be played 15,120 different ways. Nonetheless, man easily cuts his way through these impenetrable thickets of choice to make X's and O's in the right combinations in order to win.

It is an obvious fact that life is not as simple as a game of ticktacktoe, and in more complex situations man, despite the powers of his intellect, all too frequently makes mistakes. For many years, science has been grappling with the problem of how to assist man in coping with the vagaries of choice. One of the best-known of these efforts is the "theory of games" evolved by Mathematician John von Neumann and Economist Oskar Morgenstern in 1944, which attempts to translate human decision-making into pure mathematics.

Decision theory goes one step further by trying to accommodate the elusive factors of chance and human inconsistency. One of the basic principles is the distinction between value and utility--the difference in zeal, for instance, with which a rich man and a poor man will chase a windblown dollar. Like game theory, decision theory accepts human rationality. Unlike game theory, which holds that the optimal strategy is to lose no more than necessary, decision theory argues that the human contestant ought to be moved to win all he possibly can.

Game theory posits opponents of equal intelligence and craftiness, while decision theory accepts individual variation and unpredictability--including human failure to calculate risks in a logical manner. It is now known that in many situations the human subject will consistently undervalue the probabilities. In one war-problem experiment at Michigan, Psychologist Edwards carefully loaded the odds against peace at 99 to 1. His subjects, consulting the evidence, intuitively set the odds at less than 5 to 1. This human conservatism toward risk, repeatedly confirmed in the laboratory, has led decision theorists to reliance on a mathematical principle known as Bayes' theorem. Formulated by Thomas Bayes, an 18th century British cleric, it states how probabilities, or opinions about how likely an event is to occur, ought to be appraised in the light of new information. In effect, the decision theorists propose that while man can rationally reach conclusions from an original set of circumstances, he tends to cling stubbornly to these conclusions even if they are contradicted by subsequent evidence.

In an increasingly complex world, this conservative attitude toward new facts may well have significant impact on the decisions that affect human survival. Edwards cites the hypothetical example of a U.S. general digesting information from American intelligence in Europe: The general learns that Soviet troops in unprecedented number have crossed the East German border. An agent reports the boast of a Russian colonel, drunk in a Berlin Bierstube, that Chancellor Kiesinger has only a month to live. From Black Sea stations, Russian submarines move out in unfamiliar formations. The U.S. general must decide in a very short time whether these ominous data require a response --and, if so, what sort of response.

Decision theorists would argue that the general ought to pool judgment to the unemotional logic of the computer. Why? To begin with, a computer is better able to assimilate a great many variables, such as a drunken indiscretion and the movement of troops, to weigh them rapidly and to come up with a statement of mathematical probability, taking them all into account. Human judgment, faced with the same variables in a highly charged and fluid situation, simply cannot equal the machine's precise and unbiased capacity to calculate the probabilities and odds.

At a more ordinary level, the human with $40,000 to spend on a house wagers that sum on the real estate market. He has little idea what it will buy him, but he has already decided what he wants. He usually underestimates such changing influences on the price as the locale, the season and even the necessities of the individual house seller. This is the kind of judgment that, in Ward Edwards' opinion, the computer can make better than man.

No Heart. But there are other kinds of situations. A father teaching his son how to play chess will throw game after game to his opponent rather than discourage an apprentice skill. This is obviously not a probability decision. The computer can tell the father how to win. It might even tell him that letting the son win will help the boy learn only up to a point, and then achieve the opposite results. But it can scarcely tell him how important it might be to show love by throwing a game. Such considerations go beyond logic and probability.

Because he is an incorrigible humanist, Edwards does not foresee the day when man will entirely yield decision-making to the machine. But he is certain that together they would make a good team. "What is crucially human in decision-making is the evaluation process," he says. "Once the human being has assigned the values, the task of making the best decisions--that is, of computing man's own values in the most efficient way--is merely computational in nature and is best left to the machine." He is convinced, for instance, that a computer, swallowing data from a physician, would come up with a better clinical diagnosis than the doctor.

If there are defects in decision theory, one of them may be in nominating the computer to be more than it is: an adjunct of human intelligence. Whatever rudimentary reason a machine possesses is owed entirely to its creator and cannot exceed it. To propose that the builder pass judgment to his artifact is in itself an act of risk-taking--and scarcely any of the probabilities have been calculated.

This file is automatically generated by a robot program, so reader's discretion is required.