Definition of artificial general intelligence

Summary

Artificial general intelligence (AGI) is a hypothetical type of artificial intelligence designed to match or surpass human capabilities across virtually all cognitive tasks. Unlike artificial narrow intelligence (ANI), which is limited to specific, well-defined jobs, an AGI system would be able to generalize knowledge, transfer skills between different domains, and solve novel problems without needing task-specific reprogramming. Beyond AGI, artificial superintelligence (ASI) is often described as AI that would outperform the best human abilities across essentially every domain by a wide margin. Because there is no single universally agreed definition of “intelligence” for computers, AGI is commonly characterized by a set of broad cognitive abilities. Researchers generally expect an AGI system to reason and use strategy, solve problems under uncertainty, represent knowledge (including common-sense knowledge), plan, learn, and communicate in natural language, with the ability to integrate these skills to achieve goals. Some definitions also emphasize additional traits such as imagination and autonomy, and consider sensory and action capabilities (e.g., seeing, hearing, manipulating objects) as desirable for expressing intelligence in the physical world. Human-level AGI has been discussed in relation to tests such as the Turing test, as well as other “human-level” benchmarks like the Ikea test and the coffee test.

Key Points

• 1

  AGI is defined as hypothetical AI that can perform at or above human capability across a wide range of cognitive tasks, unlike ANI which is task-limited.

• 2

  AGI is typically characterized by broad abilities such as reasoning, planning, learning, knowledge representation, and natural-language communication, often integrated to achieve goals.

• 3

  There is no single agreed-upon definition of computer intelligence, so AGI is discussed through traits and proposed benchmarks (e.g., Turing test, Ikea test, coffee test).