Evolutionary Adaptation to Social Information Use Without Learning

Social information can provide information about the presence, state and intentions of other agents; therefore it follows that the use of social information may be of some adaptive benefit. As with all information, social information must be interpretable and relatively accurate given the situation in which it is derived. In both nature and robotics, agents learn which social information is relevant and under which circumstances it may be relied upon to provide useful information about the current environmental state. However, it is not clear to what extent social information alone is beneficial when decoupled from a within-lifetime learning process, leaving evolution to determine whether social information provides any long term adaptive benefits. In this work we assess this question of the adaptive value of social information when it is not accompanied by a within-lifetime learning process. The aim here is to begin to understand when social information, here expressed as a form of public information, is adaptive; the rationale being that any social information that is adaptive without learning will be a good base to allow the learning processes associated with social information to evolve and develop later. Here we show, using grounded neuroevolutionary artificial life simulations incorporating simulated agents, that social information can in certain circumstances provide an adaptive advantage to agents, and that social information that more accurately indicates success confers more reliable information to agents leading to improved success over less reliable sources of social information.