Non-Rational Discrete Choice Based On Q-Learning And The Prospect Theory

When modelling human discrete choice the standard approach is to adopt the rational model. This has been shown, however, to fail systematically under some conditions, which makes evident the need for a better approach. The choice model is however only part of the problem because it does not say how to deal with uncertainty, where learning is necessary. In this regard, some evidences support the claim that the brain employs a reinforcement learning strategy and the Q-Learning is a model that copes with it. The Q-Learning, nevertheless, is based on the rational choice defined by Bernoulli principle and von Neumann’s axioms. Then to cope with a non-rational learning algorithm, for modelling human discrete choice, a novel modification to the Q-Learning algorithm is here presented.