AI 101 — What is AGI?
AGI — short for artificial general intelligence — is a hypothetical AI system that could do any intellectual job a person can do, instead of being good at one narrow task. Every AI you can actually use today is narrow: it writes, translates, spots patterns in scans, plays games — each system built for its lane. AGI names the destination where one system covers all the lanes. It is a goal nobody has reached, not a product anyone can buy.
Why it matters right now
"AGI" is one of the most-used words in AI coverage, and usually left unexplained. Labs write it into their founding documents — OpenAI's charter defines AGI as "highly autonomous systems that outperform humans at most economically valuable work." Safety researchers organize around it: DeepMind opened an AGI institute this September, which we covered at the time — DeepMind opens an AGI institute, starting with readable thinking. Regulators and investors treat milestones tied to AGI as meaningful. So when a CEO, a safety paper, and a government report all say "AGI," knowing whether they mean the same thing is basic reading comprehension for AI news.
The one-paragraph mental model
Today's AI works like a roster of specialists hired one at a time: a translation model, a coding model, a forecasting model, each trained for its job and retrained when the job changes. AGI would be a single generalist who learns a new task in minutes without retraining, carries what worked in one domain into the next, and plans long chains of work toward a goal it wasn't specifically briefed on. Because "AGI" invites arguments, Google DeepMind's Levels of AGI paper turns it into a ladder — emerging, competent, expert, exceptional, superhuman — with separate axes for how general and how autonomous a system is, so people can place an actual model on an actual scale instead of arguing about a single finish line.
The analogy
Picture a kitchen full of appliances — a toaster perfect at toasting, a blender perfect at blending, a microwave perfect at reheating. That is today's AI: each machine excellent inside its job, useless outside it. AGI would be a trained cook who walks into any kitchen — yours, a restaurant's, a field camp's — and makes whatever meal is asked, using whatever is there, picking up the kitchen's quirks in minutes. The appliances don't become the cook, and the cook doesn't need a new machine for every new dish.
Common misconceptions
"AGI means it's conscious or alive." No mainstream definition says so. Every serious AGI definition is about capability and autonomy — OpenAI's is explicitly economic: outperforming humans at valuable work, autonomously. A system could in principle hit that bar while feeling nothing. Whether machines can be conscious is a separate, older philosophical question that "AGI" doesn't settle.
"AGI is the same as superintelligence." It isn't. AGI means human-level across the board; superintelligence means beyond human level. DeepMind's ladder keeps them apart: "exceptional" means better than most experts in most domains, "superhuman" is the rung above that. Much public confusion comes from treating the two as one event.
"Someone will flip a switch and it arrives overnight." Progress is measured in benchmarks and task lengths, not announcements. METR's research group reported in March 2025 that the length of real software tasks its agents can complete has been doubling roughly every seven months — but researchers openly argue about what that trend implies, and models that ace professional exams still stumble on weeks-long autonomous work. Even people who accept the measurements disagree on the arrival date; treat any confident calendar as a guess.
Where to learn more
Google DeepMind's "Measuring Progress Toward AGI: A Cognitive Framework" post explains its cognitive-taxonomy approach in plain language and is the friendliest entry point. For the formal vocabulary, the Levels of AGI paper is dense but the tables are worth the effort. And METR's time-horizon tracker shows the measurement debate with live charts instead of opinions.
Related reading: AI vs machine learning: what's the difference? draws the line between narrow AI and the generalist dream · What is AI alignment? covers why building AGI safely is its own research field · What is an AI agent? explains the autonomous task-doers often cited as early steps toward it.
If OpenAI, DeepMind, and the researchers who study AGI all define it differently, whose definition should regulators — and your own planning — use? Tell us in the comments.
