GENERALITY
How broadly a system can transfer competence across different kinds of tasks and domains.
Being superhuman at chess or protein folding is impressive but narrow.FIELD GUIDE · AGI, DEFINITIONS & HYPE
WHEN IS AI ACTUALLY AGI?
There is no AGI referee. Labs disagree about whether the finish line is breadth, autonomy, human-level performance, or replacing valuable work.
UPDATED 2026-09-21
NO PHD REQUIRED
ELI5
AGI is not a single scientific measurement with one universally accepted finish line. Different definitions emphasize different things: how many kinds of tasks a system can do, how well it performs compared with humans, how independently it can work, or whether it can replace people across a large share of the economy. That is why one spectacular benchmark result can be real without settling whether AGI has arrived.
PERFORMANCE
How well the system performs relative to people or other systems on the tasks being measured.
Breadth and depth are separate: a system can be broad but mediocre, or narrow and extraordinary.AUTONOMY
How long and how reliably the system can pursue goals without step-by-step human guidance.
Some AGI definitions treat autonomy as central; others treat it as a deployment property layered on top of capability.ECONOMICALLY VALUABLE WORK
Tasks people and organizations pay humans to perform across the economy.
OpenAI's longstanding definition explicitly uses this economic benchmark rather than consciousness or humanlike personality.WHY THIS BECOMES A FIGHT
WHY ARE PEOPLE FIGHTING ABOUT THIS?
Calling something AGI changes how people talk about safety, regulation, labor, investment, national power, and whether a technology milestone has been crossed. But the label can also compress several different questions into one dramatic word. A system might be broadly capable without being reliable, economically transformative without being autonomous, or superhuman on many tests without matching humans in ordinary life.
GET THESE OFF THE TABLE
THE STRAW MEN
The bad arguments first. Nobody gets to win by beating these.
THE ROBOT-GOD VERSION
“AGI MEANS A SENTIENT ROBOT GOD. IF IT ISN'T CONSCIOUS, IT DOESN'T COUNT.”
THE LEADERBOARD VERSION
“IT BEAT HUMANS ON THREE BENCHMARKS. CONGRATS, AGI ACHIEVED.”
NOW MAKE THE GOOD ARGUMENT
STEEL MAN THE CASE
Give the people you disagree with the version they would actually defend.
THE CAPABILITY MAP
GENERALITY SHOULD BE THE CORE
A useful AGI definition should distinguish breadth from depth. Google DeepMind's Levels of AGI framework explicitly separates how general a system is from how well it performs, while treating autonomy as a related but distinct deployment dimension. This makes AGI more like a multidimensional capability map than a single finish-line score.
A decathlete is defined by breadth across events, not by setting the world record in one sprint.
THE WORK TEST
ECONOMIC REPLACEMENT IS THE CONSEQUENTIAL THRESHOLD
OpenAI defines AGI as highly autonomous systems that outperform humans at most economically valuable work. This definition focuses on social and economic consequence: once systems can independently do most valuable work better than people, the distinction matters even if they do not think, feel, or behave like humans.
The economy does not care whether the machine feels human if it can do the job.
FOLLOW THE MONEY
WHO PAYS? WHO WINS?
LABS + INVESTORS
A looser AGI label can make progress sound historic and justify enormous capital spending. A stricter definition postpones the milestone and raises the evidentiary bar.
WORKERS + EMPLOYERS
Economic definitions matter because they tie AGI to substitution and productivity rather than philosophical resemblance to humans.
REGULATORS
Rules triggered by vague labels can be gamed or misunderstood. Capability thresholds are easier to govern when the measurable dimensions are explicit.
THE PUBLIC
People get a clearer picture when headlines separate 'superhuman at this task,' 'broadly capable,' 'highly autonomous,' and 'economically transformative' instead of calling all four AGI.
RECEIPTS, NOT VIBES
WHAT DO WE ACTUALLY KNOW?
THERE IS NO SINGLE AGREED DEFINITION
Google DeepMind's Levels of AGI paper begins by analyzing multiple existing definitions and proposes a framework built around performance and generality, with autonomy considered separately.
The disagreement is not merely media confusion; researchers and labs operationalize the term differently.OPENAI USES AN ECONOMIC + AUTONOMY DEFINITION
OpenAI's Charter defines AGI as highly autonomous systems that outperform humans at most economically valuable work.
That definition does not require consciousness, a humanoid body, or perfect performance at every possible task.REAL-WORLD WORK EVALS ARE TRYING TO MAKE THE ECONOMIC VERSION MEASURABLE
OpenAI's GDPval evaluates models on economically valuable real-world tasks drawn from 44 occupations.
A benchmark like GDPval can measure part of the economic-capability question without by itself establishing AGI.ONE AMAZING RESULT DOES NOT ESTABLISH GENERALITY
DeepMind's framework explicitly distinguishes narrow superhuman capability from broader general performance.
The relevant question is not only 'how high did it score?' but 'across how many genuinely different kinds of work?'WHAT WOULD SETTLE SOME OF THIS?
WHAT WOULD CHANGE THE ARGUMENT?
- A stable, independently reproducible battery of real-world tasks spanning many occupations where one system performs at or above skilled-human level rather than only on curated benchmarks.
- Evidence that the same system transfers to unfamiliar domains with little task-specific retraining or bespoke scaffolding.
- Long-horizon evaluations showing reliable autonomous completion of economically meaningful work, including recovery from mistakes and changing conditions.
- Broad agreement among labs, researchers, and policymakers on which dimensions should trigger the label, even if they still disagree on the exact threshold.
TAKE THIS TO DINNER: Stop asking whether one score proves AGI. Ask what the system can do broadly, independently, reliably and for real money.
RECEIPTS
The guide is the map. These are the sources behind the substantive claims.