Bounded rationality leads to optimal decision-making and learning under uncertainty: Satisficing, prospect theory, and comparative valuation breaking the speed-accuracy tradeoff


Some classically rational standards for actions such as optimization are simply intractable. We often instead satisfice a certain reference level that is good enough for us. We can use some heuristics but they may lead to biases. Though rational analysis by Anderson (1990) can argue the adaptive rationality of biases in relation to the environmental structure, heuristics and biases have been mostly studied in isolation from other factors in conformity with the tradition in psychology. To show the efficacy of the subrational heuristics in union, we execute computer simulations adopting the framework of reinforcement learning that models iterative decision-making under uncertainty. We implement three characteristics representative of human behavior: Satisficing (Simon, 1952), risk attitudes and reflection (Tversky & Kahneman, 1981), and comparative valuation (Kahneman & Tversky, 1979). We show that they, combined together, exhibit an adaptively optimal behavior with an extremely easy parameter tuning.

Back to Table of Contents