Recent advances in Deep Learning (DL) and ReinforcementLearning (RL) make it possible to train neural network agentswith raw, first-person visual perception to execute language-like instructions in 3D simulated worlds. Here, we inves-tigate the application of such deep RL agents as cognitivemodels, specifically as models of infant word learning. Wefirst develop a simple neural network-based language learningagent, trained via policy-gradient methods, which can inter-pret single-word instructions in a simulated 3D world. Tak-ing inspiration from experimental paradigms in developmentalpsychology, we run various controlled simulations with the ar-tificial agent, exploring the conditions in which established hu-man biases and learning effects emerge, and propose a novelmethod for visualising and interpreting semantic representa-tions in the agent. The results highlight the potential util-ity, and some limitations, of applying state-of-the-art learningagents and simulated environments to model human cognition.

Education has a profound impact on human cognition. People who have participated in education are better at solvingabstract reasoning tasks, can flexibly transfer knowledge across domains and are better at explaining their solutions.However, the properties of education that are responsible for these cognitive changes are poorly understood. We explorethe hypothesis that a structured education consisting of a cumulative, compositional curricular learning regime usingculturally constructed concepts and tools can account for many of these observations. In particular, we demonstrate thata connectionist model that learns to solve difficult analogical reasoning problems using a structured education is betterat knowledge reuse, while simultaneously providing explanations for solutions. We predict that premature progressionthrough a curriculum, before proficiency in a foundational stage has been established can fundamentally limit the potentialfor subsequent abstract reasoning performance or knowledge transfer ability.