DIEP seminar by Marta C. Couto
Evolution of boundedly rational learning in games
People constantly make strategic decisions, which are essential in negotiation, coordination, or cooperation problems. The resulting dynamics of strategy adaptation can be studied with evolutionary game theory. A crucial quantity in such models is the strength of selection, which regulates how likely individuals are to switch to a better strategy. The larger the selection strength, the more biased the learning process is toward strategies with large payoffs. Thus, selection strength is often interpreted as a measure of rationality or learning effectiveness.
Most previous models assume a fixed selection strength for all players. As a result, the effect of heterogeneous selection strengths – and whether higher selection strength leads to a strategic advantage – remains unclear. To address this, we study settings where (i) individuals have different selection strengths and (ii) strategies and selection strengths co-evolve. One might expect evolution to favor ever-increasing values of selection strengths. Remarkably, however, higher selection strength does not always lead to better long-term outcomes – depending on the strategic interaction in place, we can observe convergence to a finite value, or even evolutionary branching. This work sheds light on how evolution may shape learning mechanisms for social behavior, suggesting that boundedly rational learning might evolve not only as a by-product of cognitive constraints but as a way to gain strategic advantages.
If you wish to attend this seminar online, please send an email to s.banerjee2@uva.nl to receive the zoom-link.