Artificial General Intelligence (AGI)
Artificial General Intelligence (AGI) is a hypothetical AI system capable of performing any intellectual task a human can, matching or exceeding human cognitive flexibility across all domains without task-specific retraining. No system meeting a rigorous consensus definition of AGI has been verified as of 2026.
AGI is a theoretical milestone in AI development at which a machine exhibits general cognitive ability — the capacity to learn, reason, plan, and solve novel problems across arbitrary domains without task-specific retraining. This distinguishes it from today's narrow AI systems, which excel within the specific distributions they were trained on but degrade or fail outside them. The term entered widespread use in the early 2000s when researchers began formally distinguishing long-run capability goals from near-term machine learning progress.
How AGI would work mechanistically remains an open research question. Current leading candidates include large-scale neural networks extended with persistent memory, structured reasoning modules, tool use, and reinforcement learning from feedback — the general approach pursued by OpenAI, DeepMind, and Anthropic. Alternative approaches include symbolic AI, neurosymbolic hybrids, and whole-brain emulation. An informal operational test for AGI is sustained performance above human level across a broad, validated benchmark suite covering language, mathematics, coding, scientific reasoning, and open-ended planning, without any task-specific fine-tuning.
The significance of AGI is both technical and societal. A system that could autonomously direct its own improvements or conduct independent scientific research would compress the pace of knowledge generation substantially. At the societal level, the concentration of such a system within a single organization or state raises governance and security questions that have elevated AGI timelines to a topic in international AI policy, including discussions at G7 summits and dedicated AI safety conferences.
As of 2026, leading labs disagree on whether AGI has been reached or is imminent. Frontier reasoning models achieve superhuman scores on benchmarks such as MMLU, MATH, and competitive programming, but all current systems fail to generalize reliably across the full breadth of human cognitive tasks. OpenAI's leadership has publicly stated that AGI could arrive within a few years; many academic researchers note the absence of a rigorous consensus definition or standardized evaluation. The debate is partly definitional: no agreed threshold specifies what level of cross-domain generalization qualifies as general intelligence.