Bling AGI

Dossier entry

What people mean by AGI, and why they disagree

Bling AGI editorial ยท

The plain answer

AGI usually means an AI system that can do a wide range of tasks at least as well as people, rather than one narrow task. But there is no agreed definition or test. Some definitions focus on economic work, some on human-like learning, and some on graded levels of capability, so two people can disagree about whether AGI exists while looking at the same system.

"AGI" appears in product launches, investor letters, and headlines, often without explanation. The term is useful, but slippery. Before deciding whether a claim about AGI is right, it helps to know which AGI is meant.

The core idea

Most uses share one contrast: narrow AI is built or trained for a particular task, while general AI can handle many different tasks, including unfamiliar ones, at roughly human level or better. Everything beyond that is contested.

Four common ways AGI is defined

Framing Roughly says AGI is... What it emphasises
Economic A system that can do most economically valuable work as well as or better than people Usefulness and impact
Human-like cognition A system that can learn, reason, and adapt across domains the way people do How it thinks, not just output
Levels or tiers A scale of generality and performance, from emerging to superhuman Progress as degrees, not a finish line
Test-based A system that passes a specific test or set of tests Measurability

A well-known economic example is the OpenAI Charter, which describes AGI as "highly autonomous systems that outperform humans at most economically valuable work." A levels example is a 2023 paper by Google DeepMind researchers proposing graded levels of AGI that consider both how general a system is and how well it performs.

Why the definition changes the answer

Take the same hypothetical system: it writes code, drafts essays, and answers questions across many subjects, but makes confident errors and cannot reliably carry out long tasks on its own.

  • Under an economic definition, someone might say it is close, because it does useful work.
  • Under a human-like cognition definition, someone might say it is far, because it does not learn or reason reliably like a person.
  • Under a levels definition, someone might place it at an early level of general capability.

None of these people is necessarily wrong. They are answering different questions.

Questions to ask about any AGI claim

  1. Which definition is being used? Is it stated?
  2. What evidence would show the definition is met, and was it measured?
  3. Who benefits if the claim is believed, or disbelieved?
  4. Has anyone independent checked it?

The hype check turns these into a scorecard.

What AGI is not

  • Not the same as superintelligence, which means far beyond the best humans in nearly every domain.
  • Not the same as consciousness or sentience. A system's capability says nothing settled about inner experience.
  • Not a single product feature. It is a contested description of capability.

Limitations

This page summarises how the term is used; it does not take a position on whether or when AGI will be achieved. Definitions continue to shift. For how capability is measured, read How to read an AI benchmark claim.

Sources and further reading

We link to primary or official sources wherever we can. Guidance changes, so check the source before acting on anything important.