Dual Process

Dual-process theory posits that cognition involves both fast, intuitive processes (System 1) and slower, deliberative processes (System 2). Current research focuses on applying this framework to improve artificial intelligence, particularly large language models (LLMs), by integrating both reactive and proactive reasoning capabilities. This is achieved through architectures combining heuristic methods with more analytical approaches like Monte Carlo Tree Search, or by dynamically controlling the balance between in-context learning and in-weights learning within a single model. Such advancements hold significant potential for enhancing AI performance in complex tasks, such as dialogue planning, autonomous driving, and human-AI collaboration.