AI Needs Human-Like Metacognition for Broader Intelligence

Current AI advancements are narrow, focusing on specific tasks like image recognition or natural language processing. These successes rely on vast datasets and computational power but lack broader human-like intelligence. The paper proposes a multi-agent AI architecture inspired by Kahneman’s ‘thinking fast and slow’ theory. System 1 (fast) agents react based on past experience, while System 2 (slow) agents engage in deliberate reasoning for optimal solutions. Both types of agents use a model of the world and a self-model to guide their actions.