Konrad -- You wrote
"If humans make decisions without using a detailed analysis of the problem, then certainly these decisions will often not be consistent with the choices they would have made if they used such an analysis, or the choices prescribed by expected utility if applied to such an analysis. But saying that such examples are inconsistent with expected utility is a little harsh. Surely here one should be considering the problem where the human is acting under uncertainty, not having performed a detailed analysis of the situation, rather than the problem where such an analysis is available." Maxi-min (or mini-max of loss) is a strategy that many find palatable -- look at how long mathematicians and others considered it to be a reasonable solution to a 2 person game. But it can't be derived from expected utility theory. The same is true in Allais paradox. Perhaps one can get someone to think a little harder or point out how one could be easily taken to the cleaners if that someone adhered to maxi-min or the non-independence in Allais example and then they would change their mind. Even so, Kahneman and Tversky would still find human experts making the same mistakes. These examples are not exceptional. By taking the next step, using Expected Utility Maximization (EUM) to solve the problem of decision making under constrained cognitive and computational resources you have increased the level of complexity by orders of magnitude. So how is our poor cognitively deprived human to master that one? Rather, I believe the answer is learning how to avoid errors in decision making over time, developing short cuts along the way, in repetitive situations, much the same way a child learns to walk. Perhaps then, once the decision making apparatus is burned in, we would (and do) find experts in decision making fields acting AS IF they were using EUM. Are we ever going to find someone who is good at decision making under uncertainty and under stress tell us that he uses EUM? (Here's a question to the Aristotelian rationality crowd -- would you ask a child to explain how he decides to fire his neurons and balances himself when walking?) What Kahneman and Tversky and others discovered is that those short cuts learned from training don't always work, especially when taken out of context in which they were learned or in otherwise strange situations. But I suspect we have no disagreement -- as a prescriptive theory of decision making I find Expected Utility quite satisfying, especially for hard decisions where the stakes are high enough to make the cost of careful decision making worthwhile. Problem is that EUM is not very descriptive, is in some sense unnatural. Some would argue that therefore EUM is not relevant because it is not the natural process of making decisions (in fact there is a new field developing called "natural decision making" ). But humans do lot's of unnatural things to avoid human error: relying on instruments when flying a plane in a storm when your own senses "know better"; or Ulysses tying himself to the mast. Bob.
