How to train your Artificial Agents to be more human-like?

Deep reinforcement learning (Deep RL) methods can be used to train artificial agents (AAs) to reach or exceed human-level performance in various
task contexts. This provides a potential opportunity for researchers and practitioners to augment and enhance individual and team functioning with the
use of AAs as teammates or opponents. However, although AAs may exceed human-level performance, this does not guarantee that AA behaviours are
representative of human-like behaviours. This talk will present findings comparing the differential behaviours of AAs and human experts across
several perceptual-motor task contexts, as well as explore methods to augment the training of AAs with the use of human-inspired models. We
demonstrate how the use of hybrid, human-inspired Deep RL methods can ensure the effective realization of human-like behaviours in AA agents.