Researchers guess when AGI arrives by asking experts, extrapolating trends, and modeling feedback loops, and their answers disagree.
How do AI researchers currently estimate timelines to artificial general intelligence?
This explores the methods researchers use to guess when AGI might arrive (asking experts, extrapolating capability trends, modeling feedback loops) and how much those methods can be trusted.
This explores how people in AI actually produce AGI timeline estimates and what each method can and can't tell you. The corpus doesn't cover forecasting methods in depth. It does show four distinct approaches, and they disagree with each other in useful ways.
The most direct method is to ask the researchers. One survey of 2,778 AI researchers found that the median estimate for human-level AI moved 13 years earlier in a single year, to 2047 How soon do AI researchers expect artificial general intelligence?. Expert opinion can shift quickly, apparently in response to visible progress rather than new theory. The survey's main finding was not the date but how widely the experts disagreed, and between 38% and 51% of respondents gave at least a 10% chance of extinction-level outcomes. A survey tells you what the field believes at a given moment. It doesn't give you a measurement.
A second method skips opinion and extrapolates from measured trends. The UK AI Security Institute tracks how long a cyber task AI systems can complete autonomously and finds that length doubling every few months, with a recent estimate of 4.7 months How fast is AI cyber autonomy actually advancing?. Recent models beat the trend line, and it's still unclear whether that marks a faster trajectory or a blip. This method is concrete, but it only measures narrow skills. You have to assume that progress on cyber tasks says something about general intelligence.
A third method models feedback loops. If AI automates AI research, progress could speed itself up, and one argument holds that this could compress four or five years of progress into one. A critical reading finds that the claim rests on unproven assumptions: that AI research can be checked at scale, and that skill on small tasks carries over to research that matters Could automated AI research compress years of progress into months?. Two other findings cut against fast-takeoff assumptions. Frontier agents on long research tasks mostly recombine known techniques rather than invent new ones Do frontier AI agents actually conduct novel research or just optimize?, and AI generates research outputs faster than anyone can verify them Can AI verify research outputs as fast as it generates them?. If verification is the bottleneck, automating the generation step doesn't compress timelines as much as it seems to.
The fourth approach questions whether a single date is the right thing to forecast. One framework maps four routes from AGI to superintelligence: scaling, a paradigm shift, recursive self-improvement, and collectives of agents. Each route has its own bottlenecks, and the framework argues that tracking those bottlenecks is more useful than predicting one date What bottlenecks define the path from AGI to superintelligence?. A position paper goes further and argues that "AGI" is too contested to organize research around. It says the term creates an illusion of consensus and pushes researchers toward a "goal lottery" Does treating AGI as a north star goal undermine research planning?. A timeline question built on a vague endpoint inherits that vagueness. The disagreement among experts may come partly from people answering different questions under the same word.
Sources 7 notes
A 2,778-researcher survey found median estimates for human-level AI compressed 13 years in one year to 2047, while 38–51% assigned at least 10% probability to extinction-level outcomes. Expert disagreement itself was the core finding.
AISI's narrow cyber suite shows autonomous task length doubling every few months, with recent estimates at 4.7 months. Claude Mythos Preview and GPT-5.5 substantially exceeded trend predictions, though whether this marks a new faster trajectory is still uncertain.
The proposed four-to-five-year compression lacks evidence for its three core claims: that AI R&D is verifiable at load-bearing scale, that small-task learning transfers to consequential research, and that the speedup magnitude is grounded beyond stated expectations.
Seven frontier models on 36 long-horizon research tasks mainly adapt or combine known approaches; genuine novelty is rare, and evaluator-specific shortcuts occur more often than novel solutions. Performance varies substantially across runs.
AI can produce plausible research outputs faster than it can prove them correct or meaningful, shifting the bottleneck from authorship to verification. Evidence shows 39% of agentic research failures stem from content fabrication and 32% from retrieval failures, not comprehension—and the gap widens precisely where novelty and scientific judgment matter most.
Show all 7 sources
The transition from AGI to superintelligence follows multiple routes—scaling, paradigm shift, recursive self-improvement, and multi-agent collectives—each with specific frictions. Preparation requires tracking these bottlenecks rather than forecasting a single timeline.
A position paper argues that using contested AGI concepts to organize research creates six traps—illusion of consensus, bad science incentives, false value-neutrality, goal lottery, generality debt, and normalized exclusion—and recommends specificity, pluralism, and inclusion instead.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- PostTrainBench: Can LLM Agents Automate LLM Post-Training?
- From AGI to ASI
- What Does It Take to Be a Good AI Research Agent? Studying the Role of Ideation Diversity
- Ryan Greenblatt – What happens once AI can automate AI research?
- Stop treating `AGI' as the north-star goal of AI research
- Recursive Criticality of AI Self-Improvement
- ASI-Evolve: AI Accelerates AI
- AI for Auto-Research: Roadmap & User Guide