Large language models excel in some cognitive domains but still fall short of human abilities in others. This project explores whether we can close these gaps by building AI systems that mirror established human cognitive architectures. The team will create networks of interacting LLM “modules,” each prompted to play the role of a specific cognitive subsystem—such as memory, attention, or decision-making. Using a large dataset of human behavioral experiments, they will test how well these architectures reproduce human-like performance compared to standard LLMs. They will also evolve and refine these architectures, potentially.