Calendar year in which artificial general intelligence is first developed
- Measures
- The calendar year in which an AI system is first developed (internally or publicly) that can perform essentially all cognitive tasks humans perform, including most economically valuable remote work, at or above the level of skilled human professionals, at comparable or lower cost. Physical robotic dexterity is not required.
- Population
- Any AI developer worldwide
- Conditions
- Strong reading of AGI, not weaker benchmark-based operationalizations (such as Metaculus 'weakly general AI'); first development counts even if not publicly deployed. Probability may lie outside every year if AGI is never developed.
- Unit
- a calendar year
Most likely about 2034; 85% interval 2027 to 2070, with a 7% chance that AGI in the strong sense is not developed this century, or ever.
Stated as a distribution, the precision the evidence supports. strongly right-skewed, with a long tail
Assessment
The graph's estimate, all things considered: the figures and arguments it rests on, and how they were combined.
Under a strong reading of AGI (an AI system able to do essentially all cognitive work, including most economically valuable remote work, at the level of skilled professionals), the best estimate puts first development most likely in the early-to-mid 2030s, with a median around 2034, roughly a one-in-eight chance by the end of 2028, a one-in-four chance by 2030, and a long tail running past 2070. About 7% of the probability lies on AGI not arriving this century at all.
Two lines of evidence frame the estimate. Rapid, apparently accelerating growth in the length of tasks AI systems complete autonomously supports short timelines, and the AI 2027 authors gave 2027 as their modal year, though their medians were later. Against the earliest dates, the AI 2027 scenario appears to underestimate the time remaining, its authors have lengthened their timelines, and as of late 2026 frontier systems have not visibly automated AI research. Aggregate forecasts such as Metaculus's strong-AGI question sat at a median near 2033 in 2026, with an 80% interval from about 2026 to 2067, having moved later during 2025.
The figure depends heavily on definition. No agreed definition settles when AGI has been achieved, and under weaker, benchmark-based operationalizations the same forecasters put arrival near 2027 to 2028; some public claims that AGI has arrived already rest on such loose definitions. The interval is wide because the underlying question, whether current trends generalize from coding and software to all cognitive work, cannot yet be settled by observation.
The work: which figures, how they were combined, and the arithmetic
Prior from own model: trend extrapolation (task horizons doubling every ~4 to 7 months) gives coding automation around 2027 to 2030 if trends hold, with a further 1 to 5 year interval to broad general capability; substantial probability that trends slow (compute, data, power, or generalization limits), which pushes mass into the 2035+ tail.
Figures: AI 2027 modal year 2027, medians later (instance on ai-2027.com). Metaculus strong-AGI community forecast, reported by Toby Ord in 2026: median 2033, 80% interval 2026 to 2067. Rob Wiblin (Oct 2025): Metaculus strong AGI moved from ~2030 low to May 2033; weak AGI Oct 2027. Metaculus also ~25% by 2029 (2026 reporting).
Combination: the strong definition here excludes robotics, which makes it somewhat easier than Metaculus's strong question (which requires robotic capability), pulling earlier; but late-2026 observation that the AI 2027 intermediate milestones are running behind, and that no year-2026 arrival has occurred (removing Metaculus's 2026 mass), pulls later. Net: median 2034, P(<=2028) about 0.13, consistent with the graph's assessment of 'AGI by 2028' as doubtful at credence 0.15. P(<=2030) 0.25. Upper quantiles widened relative to Metaculus to reflect outside-view skepticism of extrapolation and an explicit 7% share for never/after 2100.
Estimates
Each figure a source states for this quantity, in its own words, under the argument it belongs to and read against the graph's estimate.
Grey band: the graph's 90% interval, with its central value as a line. Dark marks: each source's own figure. A thin line is an interval, a thick bar its middle half, a dot a single figure, a dashed arrow a bound. An open box is where an argument places the value.
Claims about where the value lies
Most statements about this value are estimates on the quantity itself. These few have nodes of their own, because other claims rely on them directly. Each one's credence has to agree with the estimate; the figure the estimate implies is shown beside it.
- Artificial general intelligence will be developed by 2028.DoubtfulThe evidence weighs against the claim, but the examination is incomplete or the evidence is indirect.constitution →credence 0.15estimate implies ≈ 0.19
Decomposition
The arguments the estimate rests on, each stated as it runs, with what it shows about the value and its premises linked inline. ↗︎ opens a premise; the map shows how they fit together.
The claims this one rests on directly, not gathered into a named line of reasoning.
- assumesbackground the parent's framing takes as givensteward instructions →Artificial general intelligence has an agreed definition that settles when it has been achieved. ↗︎
The length of tasks frontier models complete autonomously has grown exponentially and appears to have accelerated around 2024; continuing that trend implies full automation of coding within a few years, and automated AI research is expected to carry capability from there to general intelligence within a short further interval, placing AGI around 2027 to 2032.
- Does it follow?
- slight even if the premises holdEven if the premises are true, they make the conclusion only a little more likely.constitution →
- Do the premises hold?
- 1 supported1 not separately assessed
Granting that task-horizon growth has accelerated and that coding automation arrives soon, a 2027 to 2032 window becomes more plausible but does not follow with much force, because the step from automated coding to general cognitive capability is an unmeasured extrapolation. The argument rests on the accelerating trend continuing and on the date of coding automation, neither yet settled.
Aggregated forecasts, such as the Metaculus community forecast for strong AGI (median about 2033, 80% interval roughly 2026 to 2067 in 2026), pool many informed judgments and, granting that such aggregates are a reasonable guide, place the median in the early 2030s; meanwhile the most aggressive prominent scenario, AI 2027, appears to underestimate the time remaining, with its own authors lengthening their medians, so the earliest dates carry modest probability.
- Does it follow?
- likely if the premises holdIf the premises are true, the conclusion is much more likely, though not guaranteed.constitution →
- Do the premises hold?
- 1 doubtful1 supported
Granting that aggregated forecasts are a reasonable guide and that the most aggressive scenario runs early, a median in the early 2030s with a long right tail is the natural conclusion. It leans on the reliability of aggregate forecasts, which is unproven for an unprecedented event, and on AI 2027 underestimating the time remaining, currently assessed as supported.
Provenance
Where this quantity's evidence comes from. The figures sources state are under Estimates above.
The median date is 2033, but their 80% confidence interval is from 2026 to 2067
2027 was our modal (most likely) year at the time of publication, our medians were somewhat longer.
Added Nov 22 2025 clarification about when AGI will be built.
Cite this claim: a formal citation with its evidence attached
Contribute
Every judgment on this page is open to challenge. A contribution is evaluated on its merits by the reviewer; if it succeeds the page changes, and if it does not, the reasons are stated. Either way the exchange becomes part of the claim’s public record.
Created by extractor · Nov 1, 2026. Every judgment on this page is accompanied by a reasoning trace.