Artificial general intelligence (AGI) refers to an AI capable enough to replace more or less anyone with a desk job. A yet more advanced concept is superintelligence, meaning an AI so smart no human can understand it. The term AGI was coined by Shane Legg, co-founder of Google DeepMind.
In 2025 large language models from OpenAI and Google DeepMind won gold in the International Mathematical Olympiad, 18 years sooner than experts had predicted in 2021. By 2027 it should be possible to train a model using 1,000 times the computing resources that built GPT-4.
Big names in the AI industry have predicted AGI's arrival within a few years. Jack Clark, Anthropic's co-founder, sees "many trend lines up to 2027". Demis Hassabis, Google DeepMind's co-founder, thinks AI could cure all diseases "within the next decade or so". Mark Zuckerberg of Meta has said, "Superintelligence is in sight." Liang Wenfeng of DeepSeek reckons AGI may arrive in as little as two years. Sam Altman of OpenAI has predicted that AI systems will probably start producing "novel insights" by 2026. Dario Amodei of Anthropic has said "powerful AI" could arrive on the same timeframe.
The AI Futures Project, a research group, predicted in April 2026 that by the beginning of 2027 the top AI models should be as capable as a programmer at an AI lab, and by the end of that year they would be able to run the lab's research. These forecasts assume that one of the first areas of research to get a big boost from AI will be the development of AI itself—so-called "recursive self-improvement".
The Longitudinal Expert AI Panel (LEAP), run by Ezra Karger, an economist at the Federal Reserve Bank of Chicago (in a capacity not affiliated with his day job), aims to build systematic forecasts of AI development. LEAP asks specific, testable questions—such as when self-driving cars will account for 20% of American ride-hailing trips, or what proportion of electricity will be used for AI by 2040—of almost 350 experts drawn from corporate AI research, academic computer science, economics, policy and "superforecasting".
Results from the first round, published on November 10th 2025, suggested AI's impacts were just beginning to be felt. The median forecast had more than 18% of American work hours being AI-assisted by 2030, up from 2% in September 2025, with AI accounting for about 7% of American electricity usage by the same year. The average expert thought there was only a one-in-five chance of developing "powerful AI" by 2030. But by 2040 they expected AI to score eight on a ten-point impact scale devised by Nate Silver—about as important to this century as electricity or the car were to the previous one—with a nearly one-in-three chance of reaching level nine, alongside technologies like the printing press that "changed the course of human history".
Silicon Valley evangelists maintain that AGI could lift annual GDP growth to 20-30% a year, or more. Daron Acemoglu of MIT estimates a more modest effect: AI lifting global GDP by no more than 1-2% in total over a decade, on the assumption that only about 5% of tasks can be performed more cheaply by AI than by workers. Epoch AI, a think-tank, has modelled scenarios in which annual GDP growth passes 20% once AI can automate about a third of tasks.
William Nordhaus, a Nobel laureate, has shown how, when labour and capital become sufficiently substitutable and capital accumulates, all income eventually accrues to the owners of capital. In his models, less-than-perfect substitutability during an AI breakout leads to an explosion in wages—yet wages still shrink as a share of the economy, since the economy grows even faster.
Tyler Cowen of George Mason University argues that change will be slower than the underlying technology permits, constrained by energy, regulation, institutional sluggishness and "human stupidity".
Geoffrey Hinton, an AI pioneer, argues there is a 10-20% chance that AI will end in human extinction. Yoshua Bengio, a former colleague, puts the risk at the high end of that range. Nate Soares and Eliezer Yudkowsky, who signed an open letter in 2023 warning of AI's perils, are publishing a book about superintelligence entitled "If Anyone Builds It, Everyone Dies".
Both Western tech firms and their Chinese counterparts are accelerating their pursuit of AGI. The logic is that even if one firm or country pauses, others will press ahead—and the benefits of attaining AGI are likely to accrue chiefly to those who make the initial breakthrough. Donald Trump vowed in mid-2025 that America would "do whatever it takes" to lead the world in AI. J.D. Vance chided a summit in Paris in February 2025: "The AI future will not be won by hand-wringing about safety."
Yes, but every time I try to see things your way, I get a headache.