AI Must Embrace Specialization via Superhuman Adaptable Intelligence

In a new paper titled 'AI Must Embrace Specialization via Superhuman Adaptable Intelligence,' a team including Meta's Chief AI Scientist Yann LeCun challenges the industry's core pursuit of Artificial General Intelligence (AGI). The authors argue that AGI is a flawed and overloaded concept, noting that even humans are not truly 'general' in capability. They propose a new, more practical goal for AI development: Superhuman Adaptable Intelligence (SAI). This framework shifts the focus from building a single, human-like general intelligence to creating systems that can specialize to achieve superhuman performance in critical domains.

The paper defines SAI as intelligence that can learn to exceed humans at any important task and fill skill gaps where humans are inherently incapable. This move from generality to targeted superhuman specialization has major implications for research priorities, safety discussions, and product development. It suggests a future where we deploy a constellation of highly capable, specialized AI agents rather than a monolithic AGI. This conceptual shift aims to provide clearer goals for the field and could redirect billions in research funding toward building adaptable, expert systems for science, medicine, and engineering.