FOURWEEKMBA TRACKER ยท FROM THE NEWSROOM ยท UPDATED 2 NOVEMBER 2026
Free to share with a link to this page. Each bar is the latest dated forecast we could source for that person; years away and the medians are our arithmetic from the upper end of each stated range.
The short version: the people building AI keep shortening their own clocks. Demis Hassabis of Google DeepMind now expects AGI “within two to three years”, the Wall Street Journal reported on 21 October 2026, after “five to 10 years” in March 2025. Sam Altman told TIME in August 2026 that by the end of the year OpenAI would have “an internal system he would call AGI”. Dario Amodei’s guess for a “country of geniuses in a data center” is “one to two years, maybe one to three years”. Across the 10 people here who have put a year on it, the latest forecasts sit a median 3 years from when they were made, against 5 for the forecasts on record two years ago; the median year they point to is 2029 (our arithmetic).
They are not forecasting the same thing. Hassabis means the full range of human cognition, Altman uses OpenAI’s charter (“outperform humans at most economically valuable work”), Amodei means AI “smarter than a Nobel Prize winner across most relevant fields”, Musk means “smarter than any human”. The Metaculus forecasters put general AI at 18 Sep 2030 (middle half: 22 Mar 2028 to Jun 2036), and the largest survey of AI researchers put a 50% chance on machines beating humans at every task by 2047. This page tracks 35 dated forecasts from 12 people, a lab, a forecasting community and a researcher survey, each in the speaker’s own words with its source. Data as of 2 Nov 2026.
Why the forecasts differ: what each of them means by AGI
The finish lines differ, so a shorter number does not always mean a nearer finish line. Here is each forecaster’s own definition, in their words or their outlet’s:
Demis Hassabis (AGI): โa system with the full cognitive range of the human brainโnot just some of our capabilities but all of themโ Wall Street Journal, via our newsroom
Sam Altman (AGI, per OpenAI's charter; also superintelligence): โhighly autonomous systems that outperform humans at most economically valuable workโ TIME
Yann LeCun (human-level AI): โAI systems that have humanlike and human-level intelligence, but theyโre not going to be built on LLMsโ MIT Technology Review interview
Geoffrey Hinton (general intelligence greater than a person's (2023), superintelligence (2025-2026)): โsomething that's got greater general intelligence than a personโ NPR (via WAMC)
Shane Legg (minimal AGI): โMinimal AGI means an artificial agent that can handle the cognitive tasks most humans typically performโ The Decoder, on the Google DeepMind podcast
Jensen Huang (AGI as passing human tests): โIf AGI means the ability to pass human tests, it could come soonโ Reuters, via PYMNTS
Anthropic (powerful AI): โIntellectual capabilities matching or exceeding that of Nobel Prize winners across most disciplinesโ Anthropic, recommendations to OSTP
Metaculus forecasters (general AI): โWhen will the first general AI system be devised, tested, and publicly announced?โ Metaculus question 5121
As we read it, two patterns follow. The shortest dates come with the narrowest or least defined milestones: an “internal system he would call AGI” (Altman), AI “smarter than any human” (Musk). The longest come from outside the labs: LeCun gives no year, and the surveyed researchers put the 50% mark decades out. Hassabis is the forecaster whose own estimate has moved most, in public and in one direction: from “within a decade” in 2023 to “two to three years” in 2026.
Every forecast, in their own words
35 rows, grouped by forecaster, oldest first (on a phone, swipe the table sideways). “Their words” is the forecast as stated; “Implied years” converts a relative range (“two to three years”) into calendar years from the date it was said, which is our arithmetic. “Reported” means only the article’s publication date is known. 4 rows give no year at all and say so.
Who
Said
Their words, where
Implied years
Milestone, their term
Source
Demis Hassabis Chair, Google DeepMind, and Chief Scientist, Alphabet (since Aug 2026); DeepMind co-founder and its CEO until then
3 May 2023 reported
โjust a few years, maybe within a decadeโ Wall Street Journal Future of Everything festival
โby end of this year, and no later than next year; smarter than all of humanity โprobably 2030 or 2031โโ World Economic Forum, Davos, with Larry Fink
2026โ2027 'end of this year … next year' said on 22 January 2026
Most revisions point the same way: earlier. Hassabis went from “within a decade” (May 2023) to “two to three years” (October 2026). Altman went from “during this president’s term” (January 2025) to an internal AGI by the end of 2026. Amodei’s ranges have stayed within a year or two of 2027 since 2024. The exceptions are Musk, whose “end of next year” from April 2024 became “end of this year, and no later than next year” in January 2026, and Hinton, whose milestone changed from general intelligence greater than a person’s (2023) to superintelligence (2025-2026), with a longer range. Kurzweil and Legg have not moved: 2029 since 1999, and 2028 since 2011.
The timeline of predicted years
A forecast that keeps pointing at the same year while time passes is a forecast getting shorter; one that keeps sliding forward is a forecast being missed. Musk’s line is the second kind: AI “smarter than any one human” by the end of 2025, then by the end of 2026 or 2027. Kurzweil’s 2029 and Legg’s 2028 are the first kind.
What the forecasting crowds and researchers say
Metaculus, the forecasting site, asks a stricter question that includes a robotics test: “When will the first general AI system be devised, tested, and publicly announced?” Its community forecast on 2 Nov 2026 was 18 Sep 2030 (middle half: 22 Mar 2028 to Jun 2036), from about 1.9k forecasters. The 2023 survey by AI Impacts of 2,778 researchers who publish in top AI venues put a 10% chance by 2027 and a 50% chance by 2047 that “unaided machines” outperform humans “in every possible task”; the paper says the 50% year is 13 years earlier than in the same survey a year before.
What is not in the table yet
Mark Zuckerberg has written that superintelligence is 'now in sight' (July 2025) but we found no dated year or range in his own words.
Yann LeCun gives no year for human-level AI; he says it will not come 'next year or two years from now' and not from LLMs.
Sam Altman's 2024 and 2025 essays give no year for superintelligence; his only dated AGI forecasts here come from interviews (Bloomberg, TIME).
The AI Futures Project (AI 2027) publishes its medians as charts, not text; we could not read the numbers from a primary source, so it is not in the table yet.
Metaculus's community forecast from two years ago is not on the public page; only today's reading is shown.
FAQ
When will AGI arrive, according to the people building it?
The latest dated forecasts here run from the end of 2026 (Altman, for an internal system he would call AGI, per TIME) to about a decade (Karpathy, October 2025). Hassabis says two to three years (WSJ, October 2026); Amodei one to three years for a “country of geniuses in a data center” (February 2026); Musk the end of 2026 or 2027 for AI smarter than any human (January 2026). The median of the latest forecasts of the 10 people with a year is 2029 (our arithmetic, upper end of each range).
What did Demis Hassabis say about AGI?
The Wall Street Journal reported on 21 October 2026 that he now expects AGI within two to three years, after three to five earlier in 2026; IANS reported “five to eight years” in February 2026 and CNBC “five to 10 years” in March 2025. His test is whether a system trained on human knowledge up to 1901 could produce Einstein’s 1905 work. Read our story on Hassabis’s new estimate.
Why do AI leaders give such different dates?
Because they define different milestones: the full range of human cognition (Hassabis), outperforming humans at most economically valuable work (OpenAI’s charter), Nobel-level ability across most fields (Amodei), passing every human test (Huang), or superintelligence (Hinton, Zuckerberg). LeCun says human-level AI will not come from today’s language models at all.
Has any AGI prediction already been missed?
Musk said in April 2024 that AI would be smarter than any one human “probably around the end of next year”, that is 2025; in January 2026 he gave the end of 2026, “no later than next year”. We list each forecast with its date so the record stays visible.
What do forecasters and researchers predict?
Metaculus’s community forecast reads 18 Sep 2030 (middle half: 22 Mar 2028 to Jun 2036) for a general AI system that must also pass a robotics test. The largest survey of AI researchers (2,778 people, published January 2024) put the 50% year for machines beating humans at every task at 2047.
How often is this page updated?
Weekly. New AGI and superintelligence statements logged by our newsroom are checked each week, the Metaculus forecast is re-read, and every quote is re-verified against its cached source before the page is rebuilt.
Built by FourWeekMBA from 29 sources: the forecasters’ own essays, posts and podcast transcripts first, then interviews as reported by the outlets named in each row, and our newsroom’s database for Hassabis’s 2026 estimates. Every quote is checked against a stored copy of its source before each rebuild. A forecast given in years (“two to three years”) is converted to calendar years from the date it was said, and that is labelled as our arithmetic; a range is never shown as a single year. Medians and “years away” use the upper end of each range. Where no date or year was given, the row says so.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
