How has evolution shaped our decision making? A Neural Network Agent Based Model for the development of heuristics

This work focuses on understanding how evolutionary forces, in a complex environment, have shaped the human decision making processes. Within the field of choice under uncertainty, it is experimentally observed that people tend to make irrational and/or controversial choices (e.g. Allais Paradox, Ellesberg Paradox), adopting simple heuristics rather than following rational principles established by the expected utility framework. Through this project, we aim to retrieve such observed irrational preferences as an evolutionary emergent phenomenon. Specifically, the environment we live in is extremely complex, often characterized by highly non-linear and time-evolving conditions. Evolutionary forces, both exogenous (environment) and endogenous (group interaction), made our ancestors develop certain heuristics, which everyday help us taking quick and efficient decisions. The hypothesis we would like to prove is that these "mental shortcuts" work optimally in a real-world-complexity scenario, but only sub-optimally in abstract and oversimplified laboratory setups, leading to the emergence of irrational choice patterns and paradoxes. To demonstrate our point, we developed an Agent Based Model, where agents are first faced with "non-probabilized" uncertainty, with the aim of naturally facilitating the development of heuristics and strategies. Once the training phase has ended, they are then confronted with much simpler tasks (probabilized uncertainty, risky lotteries), to observe if some empirically reported patterns would arise. In order to include intelligence in our model, we chose as agents Artificial Neural Networks, and we then trained the agents by means of an evolutionary algorithm, to mimic the actual Darwinian selection. As results, we observed different decision-making attitudes, depending on the degree of risk and uncertainty of the choices faced by the agents, both in the training evolutionary phase and in the out of sample analysis. In light of such results, possible extensions of the present work are outlined, in order to gain a deeper understanding of how decision making works.