What Is Artificial General Intelligence?

Direct Answer

Artificial general intelligence (AGI) is the hypothetical capacity of an AI system to perform any intellectual task that a human being can — not by memorizing patterns for each task, but through general-purpose reasoning, learning, and problem-solving that transfer across domains. A system with AGI could write a novel, prove a theorem, run a laboratory, negotiate a contract, and learn to pilot a spacecraft, adapting to novel situations without special retraining. This is contrasted with narrow AI, which excels at one well-defined task: chess engines beat world champions but cannot hold a conversation; language models converse but cannot reliably plan or reason about the physical world. AGI remains a concept more than a reality. Definitions vary — some emphasize human-level performance across most economically valuable tasks, others emphasize flexibility and generality as such — and whether AGI will be achieved, when, and by what methods are among the most contested questions in the philosophy and science of AI.

Historical Context

The goal of general machine intelligence is as old as AI itself. Turing's 1950 paper asked whether machines could think, and the founders of the field — Minsky, McCarthy, Simon, Newell — believed that a few decades of work would yield machines with full human-level intelligence. The 1956 Dartmouth workshop, which named the field, assumed that "every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it." That optimism crashed repeatedly: the "AI winters" of the 1970s and 1980s showed that general reasoning was far harder than expected, while expert systems were brittle. The field fragmented into narrow successes — computer vision, speech recognition, game playing — and the phrase "artificial general intelligence" was coined in the 1990s-2000s to name what the field had quietly given up pursuing. The 2010s deep learning revolution, culminating in large language models, has revived the question of whether general intelligence is achievable by scaling learning systems, and researchers now debate it with renewed intensity.

Key Arguments & Debates

One debate concerns feasibility. Optimists argue that intelligence is largely a matter of pattern learning at scale, so current trends — bigger models, more data, better architectures — will converge on AGI, possibly within decades. Skeptics argue that current systems lack genuine understanding, causal reasoning, and embodiment, and that scaling statistical prediction will never produce a mind that truly grasps the world. The philosophical dimension is equally contested. Computationalists hold that mind is a kind of computation, so a sufficiently powerful general system would be a mind; anti-computationalists, from Searle to Penrose, argue that meaning, understanding, or consciousness requires something computation cannot supply. A further debate concerns the path: some argue AGI must be engineered deliberately with new architectures; others, that it will emerge unpredictably from scaling and self-improvement. And a third debate concerns the timeline and the risks: Bostrom argues that a self-improving AGI could rapidly become a superintelligence posing existential risk, while others treat AGI as distant science fiction that distracts from current harms.

Contemporary Relevance

By 2026, the debate is no longer academic. Large language models and multimodal systems have crossed thresholds once considered markers of general intelligence — passing professional exams, generating code, solving novel reasoning puzzles — leading some researchers to argue that AGI is near, and others to retort that benchmark scores are not intelligence. Governments and companies now publish AGI timelines and safety plans; regulators debate when a system counts as general-purpose and what obligations follow. The philosophical questions sharpen the policy questions. If AGI systems are merely powerful tools, the key issues are alignment, safety, and control. If they could be conscious minds, the key issues become welfare and moral status. And if they could be superintelligent, the stakes become existential. Understanding AGI therefore requires philosophy as much as engineering: what counts as intelligence, whether minds can be artificial, and what we owe to minds, however they are made.

Related Concepts

• What Is the Turing Test? — the classic criterion for machine intelligence
• Can AI Be Conscious? — whether AGI would have inner experience
• What Is the Technological Singularity? — the hypothesized intelligence explosion
• What Is the Alignment Problem? — keeping powerful AI aligned with human values
• What Is AI Ethics? — the moral framework for AI development
• Topic: Technology — the hub for technology-related questions

Further Learning

• Stanford Encyclopedia of Philosophy: Artificial Intelligence — the definitive survey of AI and its philosophy
• Internet Encyclopedia of Philosophy: Artificial Intelligence — accessible introduction
• Stanford Encyclopedia of Philosophy: The Computational Theory of Mind — the thesis that minds are computations
• Stanford Encyclopedia of Philosophy: Turing Test — the history and limits of behavioral tests