Glossary · AGI and alignment
AGI (artificial general intelligence)
AGI, or artificial general intelligence, refers to the idea of AI that can learn and perform a broad range of intellectual tasks beyond a narrow specialty. Definitions differ over required performance, autonomy, and learning efficiency, so the term does not provide a universally accepted technical classification.
AGI (artificial general intelligence): the explanation
The central idea is an ability to transfer skills and adapt to varied problems. A specialized system may excel at one activity while being unable to learn another. AGI would be assessed across a much wider scope. However, proposed definitions emphasize different things: task diversity, acquiring new skills, or performing economically relevant work. When you encounter an AGI claim, first ask which definition is being used and what evidence would satisfy its requirements.
The term is used to discuss a research objective and the possible consequences of more general systems. It does not name one architecture or a feature you can simply switch on in an assistant. A language model, an agent, and AGI describe different concepts. An agent is organized to pursue tasks, often with tools and feedback. That organization alone does not demonstrate general intelligence, regardless of how many services it can connect to.
No single test settles every issue raised by the term. Strong performance may depend on familiar data, specialized tools, or a large computing budget. Relevant evidence includes capabilities, limitations, and evaluation conditions. Consciousness is a separate question: a functional definition of AGI can concern what a system can accomplish without claiming that it has subjective experience. Likewise, broad capability does not automatically establish reliability, appropriate objectives, or safe behavior in practical applications.
A concrete example
You watch an assistant draft a letter, analyze a spreadsheet, and organize a trip. This versatility is useful, but it does not by itself demonstrate AGI. Supporting that claim would require clear criteria and evidence about learning genuinely new tasks, adapting to unfamiliar settings, and succeeding without hidden assistance or task specific preparation. A polished demonstration cannot answer all those questions on its own.
Why it matters
AGI can carry considerable weight in product presentations and public debate. When choosing a tool, translate the label into practical questions: which tasks does it complete, with what reliability, permissions, and human oversight? Evidence about your actual workflow matters more than a broad classification. An AGI announcement does not, on its own, guarantee that an application is useful, trustworthy, or safe.
Common misconceptions
- An assistant that discusses many subjects is not automatically AGI, because breadth of visible conversation does not establish general adaptation to unfamiliar situations or reliable competence across all the activities mentioned in its fluent answers.
- AGI does not necessarily imply consciousness, because capability based definitions concern learning and problem solving without establishing subjective experience, feelings, or an independent will in the system being evaluated under those definitions.
Related terms: agi and alignment
- ARC-AGI
- ARC-AGI is a benchmark series designed to study how AI systems solve unfamiliar problems from limited examples or interactions.
- ARC Prize
- ARC Prize is a research and competition initiative that encourages open progress on AI generalization problems.
- generalization
- Generalization is an AI system’s ability to use what it has learned to handle new cases.
- fluid intelligence
- Fluid intelligence is the ability to reason through unfamiliar problems without relying mainly on previously learned knowledge.
Where this site covers it
The term appears on 24 pages of this site. A few of them:
- The history of AGI: from Turing to today's labs Guide
- When will AGI arrive? Published forecasts, their methods, and why they differ Guide
- AGI risks: what researchers say, without alarmism Guide
- How We Get To AGI: François Chollet’s talk, chapter by chapter Guide
- Measuring progress toward AGI: benchmarks, limits, saturation, and contamination Guide
- ARC-AGI: the test, its versions, and published scores Guide
- Program synthesis: principles, history, and a possible path to AGI Guide
- ARC Prize: the competition, its rules, editions, results, and lessons Guide
Go further
Frequently asked questions
Is there one accepted definition?
No. Definitions differ in task scope, performance level, autonomy, and learning ability. A meaningful claim should specify its criteria, evaluation methods, and observed limitations instead of relying on the acronym as though it settled those questions.
Would AGI always be better?
General capability does not guarantee superiority on every task. A specialized tool may remain more accurate, faster, or easier to control for a particular need. Compare demonstrated performance and practical constraints rather than assuming that broader means better.
Can its arrival be predicted?
An announced date depends on assumptions about technical progress and on the definition being used. It is not an established fact. To understand a forecast, examine its criteria, uncertainties, and the evidence that could count against it.
Neutrality. No affiliate links. No vendor pays to be reviewed, rated or ranked. Reviews are directed by François Chollet, neutral and factual, and every data point is dated and sourced. How we score