A Definition of AGI (www.agidefinition.ai)

Researchers from the Center for AI Safety and multiple academic partners proposed a concrete, quantifiable definition of AGI: an AI that matches or exceeds the cognitive versatility and proficiency of a well‑educated adult. To operationalize this, they map Cattell‑Horn‑Carroll (CHC) human cognitive theory onto ten equally weighted (10% each) core cognitive components—including acquired knowledge, perception, central executive function, memory, and output—and adapt established human psychometric batteries to evaluate AI systems. The framework produces granular "AGI scores" that quantify progress (examples given: GPT‑4 ≈ 27%, GPT‑5 ≈ 58%) and expose where models concentrate capabilities versus where they falter. Significance lies in turning AGI from a vague goal into a measurable engineering target and benchmark: the framework highlights a highly “jagged” cognitive profile in current models (strong in knowledge‑rich tasks but weak in foundational cognitive machinery), with pronounced deficits in long‑term memory storage and other executive functions. That has immediate implications for research priorities (e.g., memory architectures, sustained reasoning), model evaluation, safety/validation standards, and policy discussions—enabling more objective tracking of progress toward generality and clearer identification of failure modes that matter for robustness and alignment.