The AGI Clock — methodology

← Back to the clock · The Balance — design mockup

What it is

A public, evidence-backed "window" for AGI arrival, in the spirit of the Bulletin of the Atomic Scientists' Doomsday Clock (currently 85 seconds to midnight, set January 2026) — but transparent instead of oracular. Every input is sourced, dated, and defined. The page shows a range, never a single date.

Inputs (checked November 4, 2026)

Input	Value	Source	As of
Metaculus community median, "Date of General AI" (Q5121, 1,800+ forecasters)	Sep 2030; central range Mar 2028–Jul 2036	metaculus.com/questions/5121	Nov 4, 2026 (live page) — moved from Jan 2033 (mid-2026 reading)
Metaculus "weakly general AI" (cognitive-only basket)	Aug 2027; resolution under active dispute on the platform	metaculus.com/questions/3479	Nov 4, 2026 (live page) — moved from Jun 2028
Combined aggregate (Metaculus + Manifold + Kalshi + expert surveys)	~2031; 80% CI 2027–2044	aitoolsreview.co.uk	Jul 2026
Grace et al. expert survey (n=2,778), HLMI	10% by 2027; 50% by 2047	arXiv:2401.02843	survey fielded 2023
METR 50%-reliability time horizon	doubling ~7 months since 2019; ~2–5 hrs early 2026; month-long tasks 2027–2031 if trend holds	metr.org	early 2026
Kurzweil	AGI 2029; singularity 2045	The Singularity Is Nearer (2024)	2024
Lab leaders (Amodei, Hassabis, Musk, Altman, Hinton, LeCun)	2026 → decades	public statements	2024–2026

Window construction

• Outer window: 2029–2047. Lower bound = earliest credible named forecast (Kurzweil 2029). Upper bound = expert-survey median (2047). Shaded band on the timeline starts at 2027 (survey 10% mark).
• Consensus cluster: 2029–2033. Where the live markets and the combined aggregate concentrate (bright band).
• The page never presents a single "AGI date." The width of the band is the message.

Update cadence

• Quarterly re-cut of the window. Next scheduled review: January 2027.
• Metaculus medians refresh continuously; the page's "as of" stamps must move with them.
• Re-cut triggers: new large expert survey; sustained break in the METR doubling trend; definitional convergence in forecasting questions.

What it is not

• Not a prediction. Not investment, policy, or safety advice.
• Not a measurement of "percent to AGI" — no such metric exists. The capability curves measure slope, not distance to destination.

Refresh note — November 4, 2026

• Metaculus Q5121 median moved Jan 2033 → Sep 2030 (~2.3 years earlier) since the draft's mid-2026 reading; the live median now sits earlier than the July 2026 combined aggregate center (~2031).
• Weakly-general median moved Jun 2028 → Aug 2027, and forecasters are actively disputing whether that question has effectively resolved (unverified model claims; Metaculus investigating, not resolved).
• Open decision: the consensus cluster (2029–2033) and the WJC line (~2032) were set against the old center. Per this methodology the line "moves when the evidence moves" — re-cutting the band is a decision for the WJC partners, ahead of the scheduled January 2027 quarterly re-cut. Not changed in the Nov 4 refresh.

Publication status

• Sample demo — November 10, 2026. Draft; not published; prepared for discussion with William Jewell College.
• No analytics, tracking, or cookies on these pages.