Artificial General Intelligence (AGI)
Artificial General Intelligence (AGI) refers to a hypothetical form of AI that can understand, learn, and apply knowledge across a wide range of tasks at a level comparable to, or exceeding, human cognitive ability.
Definition
Artificial General Intelligence (AGI) refers to a hypothetical form of AI that can understand, learn, and apply knowledge across a wide range of tasks at a level comparable to, or exceeding, human cognitive ability.
Overview
Most AI systems in production today, including large language models like the ones behind ChatGPT, are considered "narrow" AI — extremely capable at pattern recognition, language, and specific tasks, but without genuine general reasoning, autonomous goal-setting, or transfer of understanding across arbitrarily different domains the way a human can. AGI describes a system that could match or exceed human performance across virtually any intellectual task, adapting to novel situations without task-specific retraining. There is no agreed-upon technical definition or test for when AGI has been achieved, and estimates from researchers and labs vary widely — some believe current scaling of deep learning and transfer learning techniques could approach AGI-like capabilities within years, while others argue fundamentally new architectures or theories of reasoning are required. The term is often used loosely in industry and media, which makes it useful to distinguish AGI from more measurable milestones like passing specific benchmarks or exhibiting broad few-shot learning. AGI is closely tied to discussions of AI safety and alignment, since a system with general, autonomous capability raises materially different risk and governance questions than today's narrow, deployed models. The blog post AI Safety and Ethics: What Every Developer Should Know covers these considerations in more detail.
Key Concepts
- Aims for human-level or greater performance across arbitrary tasks, not one narrow domain
- No industry-agreed technical definition or objective test for achievement
- Distinct from today's narrow AI systems, including large language models
- Central to debates about AI safety, alignment, and governance
- Estimated timelines vary widely across researchers and organizations
- Closely discussed alongside concepts like autonomous reasoning and self-improvement