A Definition of AGI
Dan Hendrycks, Dawn Song, Christian Szegedy, Honglak Lee, Yarin Gal, Erik Brynjolfsson, Sharon Li, Andy Zou, Lionel Levine, Bo Han, Jie Fu, Ziwei Liu, Jinwoo Shin, Kimin Lee, Mantas Mazeika, Long Phan, George Ingebretsen, Adam Khoja, Cihang Xie, Olawale Salaudeen, Matthias Hein, Kevin Zhao
2025-10-27
Summary
This paper tackles the tricky question of what Artificial General Intelligence, or AGI, actually *is*, and provides a way to measure how close we are to achieving it.
What's the problem?
Currently, there's no agreed-upon definition of AGI. This makes it hard to tell how much progress is really being made with AI. We know today's AI is good at specific tasks, but it's unclear if it's getting closer to having the broad, flexible intelligence of a human being. It's like trying to measure height without a ruler – you don't know where you stand.
What's the solution?
The researchers created a framework for defining and measuring AGI based on how the human brain works. They used a well-established psychological theory called Cattell-Horn-Carroll theory, which breaks down human intelligence into ten different areas like reasoning, memory, and how we perceive things. They then adapted tests used to measure these abilities in people to test AI systems. This gives each AI a score showing how it stacks up against a well-educated adult in each area, and overall.
Why it matters?
This work is important because it moves the discussion of AGI from vague ideas to something concrete and measurable. By giving AI systems an 'AGI score', we can track progress more accurately and identify specific areas where AI still needs to improve. The results show current AI, even advanced models like GPT-4 and GPT-5, still have significant gaps in fundamental cognitive abilities, especially long-term memory, despite being strong in knowledge-based tasks.
Abstract
The lack of a concrete definition for Artificial General Intelligence (AGI) obscures the gap between today's specialized AI and human-level cognition. This paper introduces a quantifiable framework to address this, defining AGI as matching the cognitive versatility and proficiency of a well-educated adult. To operationalize this, we ground our methodology in Cattell-Horn-Carroll theory, the most empirically validated model of human cognition. The framework dissects general intelligence into ten core cognitive domains-including reasoning, memory, and perception-and adapts established human psychometric batteries to evaluate AI systems. Application of this framework reveals a highly "jagged" cognitive profile in contemporary models. While proficient in knowledge-intensive domains, current AI systems have critical deficits in foundational cognitive machinery, particularly long-term memory storage. The resulting AGI scores (e.g., GPT-4 at 27%, GPT-5 at 58%) concretely quantify both rapid progress and the substantial gap remaining before AGI.