- Emergence of emotional expressions (and emotional states) as a communication system in a multi agent environment. - Ecological simulation to explore niche development - Sexual selection to drive improved fitness - evolve selection criteria in parallel with fitness - When is SS beneficial? - is the runaway phenomenon useful in some contexts but not others? - is there a complexity threshold that must be met before SS is beneficial? - Is SS ever beneficial? - Approach - first show benefit on some task - then ablate task to simplify it until SS is no longer beneficial - Sexual selection to drive speciation and create niches - if runaway SS drives the species into a new domain, is part of the benefit that it’s a domain isolated from other species? - How does SS relate to novelty search? - if SS drives the species into new areas, is this a form of novelty search? - Do the selectors in SS evolve to look for novelty? - What features do the selectors in SS look for? * Mate selection is action selection! * is there benefit in life-time learning? * What’s the reward function? * quality / viability of offspring? * Start with RL infrastructure in the agent * Mate selection is an objective function that the agent’s genome defines * Mating is an action selection over the course of a lifetime * Wait! * Version 1: the genome defines target mate characteristics. * Version 2: the genome defines an algorithm for determining target mate characteristics * What does an RL approach look like? * Life-time learning. * Multi mating chances during a lifetime * Feedback based on offspring characteristics * very similar problem to defining a desirable mate! Defining a desirable child! * species-wide learning, control net in the genome