AGI timelines
AGI timelines are probabilistic forecasts about when a specified threshold of artificial general intelligence, human-level machine intelligence or a closely related capability may be reached. A usable timeline names the target, its operational criteria, conditioning assumptions, probability level or distribution, and forecast date. Timelines may come from explicit models, expert elicitation or aggregated forecasters; the shared label does not make their events or methods interchangeable.
Origin and context
Predictions about human-level AI go back to early AI discourse, but the earliest dedicated empirical study reviewed here surveyed AGI-09 participants in 2009 and published the results in 2011. Later surveys sampled broader groups of machine-learning researchers, while Epoch compared model-based and judgment-based forecasts. Metaculus has maintained a public date question since 2020, and the 2026 LEAP panel again elicited a distribution under an explicit economic and occupational definition. No person or organization owns the category.
Why it matters
Timeline assumptions affect how organizations sequence capability monitoring, safety research, policy preparation and investment under uncertainty. Their value is not a single countdown but a transparent account of what is forecast and what evidence would change it. Comparing successive forecasts can reveal updates, yet movement may also reflect a changed definition, respondent pool or question format. Skills Intelligence therefore treats a timeline as a decision input to stress-test, not as proof that AGI will arrive on its median date.
Example
Suppose two reports both place a 50% date in the same decade. One asks when unaided machines can outperform humans at every task, conditional on uninterrupted science. The other asks when a commercially available system can beat a high-performing worker across most non-physical tasks below a cost ceiling. Those medians are not replicas: their event definitions and conditions differ. A responsible comparison records each question verbatim enough to preserve the threshold, separates conditional from unconditional probability, and dates any live community estimate.
How it differs
AGI
AGI names the contested capability target. An AGI timeline is a forecast about when one explicit version of that target may be reached; it cannot repair an undefined target.
AI Takeoff Speed
AI takeoff speed concerns the duration and dynamics of moving between capability milestones. An arrival timeline concerns the date of a stated threshold; neither determines the other.
p(doom)
p(doom) is a credence in a specified bad outcome, sometimes conditional on advanced AI. It is not a forecast of the date when a capability threshold will be reached.
AI 2027
AI 2027 is one named scenario with a detailed causal narrative. AGI timelines are the broader class of forecasts and may use surveys, models or aggregation without telling that scenario.
Maturity and evidence
Maturity is 4 for the vocabulary and forecasting practice. Dedicated work spans the 2009 assessment, later multi-year expert surveys, independent model reviews, a large public forecasting question, synthesis by Our World in Data and a 2026 longitudinal panel. These organizations use different methods, which supports adoption while also preventing a universal numerical answer. The rating does not validate forecast accuracy or imply that AGI exists.
Limits and open questions
Long-horizon AGI forecasts face no large set of resolved, repeated AGI events for direct calibration. Expert samples can be selective, model outputs depend on structural assumptions, and elicited dates change with framing. Definitions also differ on breadth, autonomy, cost, physical work and whether scientific progress continues without disruption. A live aggregate can move when participants or bounds change. Report ranges and assumptions, preserve old vintages for audit, and avoid calling any survey, model or community median a consensus or an arrival schedule.
Related terms
References
- How Long Until Human-Level AI? Results from an Expert AssessmentTechnological Forecasting & Social Change / Seth Baum, Ben Goertzel and Ted Goertzel · 2011 · class A
- When Will AI Exceed Human Performance? Evidence from AI ExpertsKatja Grace et al. / arXiv · 2017-05-24 · class A
- Thousands of AI Authors on the Future of AIKatja Grace et al. / arXiv; Journal of Artificial Intelligence Research · 2024-01-05 · class A
- Literature review of transformative artificial intelligence timelinesEpoch AI · 2023-01-17 · class A
- Longitudinal Expert AI Panel, Wave 8: TimelinesForecasting Research Institute · 2026-06-01 · class A
- When Will the First General AI Be Announced?Metaculus · 2020 · class B
- AI timelines: What do experts in artificial intelligence expect for the future?Our World in Data · 2023-02-07 · class B
Last updated: 2026-09-07
This term is also covered in the Skills Atlas as ai risk management skill.