simplify the following so anyone with a highschool education can understand: We study a sufficiently general regret criterion for choosing between two probabilistic lotteries. For independent lotteries, the criterion is consistent with stochastic dominance and can be made transitive by a unique choice of the regret function. Together with additional (and intuitively meaningful) super-additivity property, the regret criterion resolves the Allais' paradox including the cases were the paradox disappears, and the choices agree with the expected utility. This superadditivity property is also employed for establishing consistency between regret and stochastic dominance for dependent lotteries. Furthermore, we demonstrate how the regret criterion can be used in Savage's omelet, a classical decision problem in which the lottery outcomes are not fully resolved. The expected utility cannot be used in such situations, as it discards important aspects of lotteries.
GPT-3.5
GPT-6 Astra
We study a way to choose between two uncertain options, or “lotteries,” based on the regret someone might feel if the option they rejected turned out better.
When the two options’ outcomes are independent—meaning one does not affect the other—this approach favors an option that offers better chances of good results. There is also exactly one way, within our framework, to measure regret that keeps choices logically consistent: if someone prefers A to B and B to C, they should prefer A to C.
We also add an intuitive assumption: regret from several disappointments together can be at least as strong as the regret from each disappointment added up. With this assumption, our approach explains the Allais paradox, where people’s choices conflict with standard decision theory. It also explains when that conflict disappears. The same assumption helps ensure sensible choices when the options’ outcomes are linked rather than independent.
Finally, we show how the approach works in Savage’s omelet problem: deciding whether to add an egg that might be rotten to an omelet. This illustrates a decision where some possible outcomes remain unknown. We argue that standard expected utility—which judges an option by averaging how desirable its possible outcomes are, weighted by their probabilities—leaves out information that matters in such cases, while a regret-based approach can account for it.
What changed
GPT-3.5’s version reads easily but simplifies by deleting. It says the method is “consistent with other methods that consider probability,” a vague stand-in for stochastic dominance. It drops the unique transitive regret function and never mentions the super-additivity property. Its claim about situations “where we don’t have all the information” appears twice, once out of place, while Savage’s omelet goes unnamed. A high schooler would understand every sentence and still not know what the paper found.
Astra keeps nearly every claim and translates each one. Transitivity becomes “if someone prefers A to B and B to C, they should prefer A to C.” Super-additivity becomes regret from several disappointments together being “at least as strong” as their sum. The omelet gets its concrete egg that “might be rotten,” and expected utility is finally defined. It is roughly twice as long, but the extra words carry the content GPT-3.5 left out. Astra also notes that the paradox sometimes “disappears,” a detail GPT-3.5 dropped.