Textbook of AI

Also known as: AGI, strong AI, general AI, human-level AI

Artificial General Intelligence (AGI), also called strong AI, refers to a hypothetical AI system that would possess the flexible, adaptive intelligence characteristic of human cognition: the ability to learn any intellectual task without being explicitly reprogrammed, to reason about novel situations, and to transfer knowledge fluidly across domains. Whereas narrow AI excels at specific tasks, AGI would match or exceed human performance on the full range of cognitive work.

There is no consensus definition of AGI, nor agreement on how to measure progress toward it. Some researchers focus on performance across a broad suite of benchmarks; others emphasise autonomy, self-improvement, or subjective experience. The Turing test was an early proposal for operationalising "human-like" intelligence, but modern researchers regard it as neither necessary nor sufficient. More recent proposals emphasise the ability to perform economically valuable work, pass professional examinations, or solve novel scientific problems.

Opinions on AGI timelines range widely, from those who believe it is decades or centuries away to those who expect it within years. As of the mid-2020s, no deployed system is AGI, though large language models have sparked vigorous debate about whether they represent meaningful progress or merely very capable narrow systems. AGI is central to discussions of AI alignment and existential risk: if a system far more capable than humans were pursuing objectives misaligned with human welfare, the consequences could be profound and difficult to reverse.