Melanie Mitchell: Concepts, Analogies, Common Sense & Future of AI

Intelligence remains a deeply complex, emergent phenomenon that resists simple reductionist definitions. Melanie Mitchell, a professor of computer science and expert in complex systems, argues that human cognition relies fundamentally on analogy-making, where concepts are fluidly applied across different contexts to generate expectations and mental models. While current deep learning models achieve remarkable success in narrow tasks, they lack the common sense, intuitive physics, and dynamic feedback loops required for true human-level intelligence. The "long tail" of unpredictable edge cases in real-world environments, such as autonomous driving, highlights the limitations of purely data-driven approaches. Achieving artificial general intelligence requires moving beyond brute-force scaling toward architectures that can form and manipulate concepts, mirroring the way humans integrate sensory input with internal models to navigate an open-ended world.