William Jewell College
a proposed student research project
--:--:--local time
Sample demo — draft, not published
Countdown to AGI
How far away is AGI? Not a date — a window, built from expert surveys, live forecasting markets, and measured capability curves. Tap the timeline to see how it's built.
WJC estimate
–
to the WJC estimate of ~2032
tap for details · tap again for references
How the window is built
Live forecasting consensus — median Sep 2030
Metaculus community (1,800+ forecasters): median date for the first general AI system September 2030, central range Mar 2028–Jul 2036. "Weakly general AI" (cognitive only): August 2027 — its resolution is now under active dispute on the platform. Checked live November 4, 2026; updates continuously.
Metaculus community (1,800+ forecasters): median date for the first general AI system September 2030, central range Mar 2028–Jul 2036. "Weakly general AI" (cognitive only): August 2027 — its resolution is now under active dispute on the platform. Checked live November 4, 2026; updates continuously.
Combined market + survey aggregate — ~2031, 80% range 2027–2044
Aggregating Metaculus, Manifold, Kalshi and expert surveys (July 2026): central estimate ~2031, 80% interval 2027–2044. The band is nearly two decades wide — that width is the honest headline.
Aggregating Metaculus, Manifold, Kalshi and expert surveys (July 2026): central estimate ~2031, 80% interval 2027–2044. The band is nearly two decades wide — that width is the honest headline.
Expert survey, 2,778 AI researchers — 50% by 2047
"High-level machine intelligence": 10% by 2027, 50% by 2047 — thirteen years earlier than the 2022 round.
"High-level machine intelligence": 10% by 2027, 50% by 2047 — thirteen years earlier than the 2022 round.
Capability curve (METR) — doubling every ~7 months
Task length frontier agents complete at 50% reliability, doubling since 2019. Early 2026: ~2–5 hours. Month-long tasks 2027–2031 if the trend holds. Measures the slope, not the destination.
Task length frontier agents complete at 50% reliability, doubling since 2019. Early 2026: ~2–5 hours. Month-long tasks 2027–2031 if the trend holds. Measures the slope, not the destination.
Named forecasters — 2026 to 2047+
Kurzweil: 2029. Amodei: ~2026–27. Hassabis: ~2030. Hinton: 5–20 years. LeCun: decades, if ever. The spread is the signal.
Kurzweil: 2029. Amodei: ~2026–27. Hassabis: ~2030. Hinton: 5–20 years. LeCun: decades, if ever. The spread is the signal.
The green line — WJC estimate, ~2032
Set at the live consensus center. It's a stake in the ground, not data — it moves when the evidence moves. Definitions differ across inputs (robotics + cognition vs. cognitive-only vs. every-task), and the page never mixes them silently.
Set at the live consensus center. It's a stake in the ground, not data — it moves when the evidence moves. Definitions differ across inputs (robotics + cognition vs. cognitive-only vs. every-task), and the page never mixes them silently.
References & methodology
- Metaculus Q5121 "Date of General AI" — metaculus.com (median Sep 2030, checked live Nov 4, 2026; earlier reading Jan 2033 via aitoolsreview.co.uk, Sept 2026)
- Grace et al., "Thousands of AI Authors on the Future of AI" — arXiv:2401.02843
- METR, "Measuring AI Ability to Complete Long Tasks" — metr.org, Mar 2025; TH1.1 update Jan 2026
- Kurzweil, The Singularity Is Nearer (2024): AGI 2029, singularity 2045
- Doomsday Clock analogy: Bulletin of the Atomic Scientists, 85 seconds to midnight (Jan 2026)
- Window: 2029 (earliest credible named forecast) to 2047 (survey median); shaded band from 2027 (survey 10% mark).
- Re-cut quarterly by WJC students. Next review: January 2027. Triggers: new large survey, break in the METR trend, definitional convergence.
- Not a prediction. Not advice. Data as-of dates shown per input.