Skip to content

All five questions

How fast will the impact arrive?

The claim, as coded: “Transformative economic effects of AI will arrive fast (within ~a decade)”

The short answer AI-written

Will transformative economic effects arrive within about a decade? This is the least settled claim in our network. Preparedness-minded researchers treat near-term transformation as plausible enough to plan for; empiricists point out aggregate effects remain too small to detect and argue AI diffuses like a normal technology, slowly and unevenly.

A synthesis of the cited research, not a prediction or a representative consensus.

Why they disagree AI-written

Expected dates for AI matching human ability range from the next few years to several decades away.

Each row is one assumption and the published papers that make it (of 52 on this question). Dots: their estimates, hover for the paper. Bar: where their findings point.

How much more capable will AI become, and how fast?

Much more, soon8 papers · middle estimate 2027
5 for, 2 mixed
Steady progress1 paper
No estimates
1 mixed

Would settle it: Whether AI's measured ability on long, real tasks keeps improving at the recent rate.

How quickly will the economy take it up?

Quickly4 papers
No estimates
2 for, 1 mixed
Gradually1 paper
No estimates
1 against

Would settle it: Year by year counts of firms and workers using AI for core work.

What kind of evidence is the forecast built on?

Judgment of people close to AI6 papers
2 for, 1 mixed, 1 against
Surveys of many experts9 papers · middle estimate 2059
5 mixed, 1 against
Economic models13 papers
No estimates
1 for, 6 mixed, 1 against

Would settle it: Comparing past forecasts with what happened, as some of the published papers already do.

Over what period is the forecast made?

The next few years2 papers
No estimates
2 for
A decade or more13 papers · middle estimate 2043
3 for, 3 mixed, 4 against

Would settle it: Following the same forecasters over time to see how their dates move.

The case made for the claim

Brynjolfsson writes that we are approaching transformative AI and that understanding its economic reshaping is urgent, anchoring a research agenda on that premise. Korinek models explosive-growth feedback loops seriously enough to urge preparation. Bengio warns frontier capabilities are advancing rapidly; Hadfield writes that an economy of AI agents is emerging and institutions must adapt quickly; Bakker argues Europe needs an AGI preparedness plan on near horizons. Korinek and Suh model AGI transition scenarios in which full automation could be reached, treating rapid transformation as a case worth working through in detail.

Still open

  • What would count as early, measurable evidence of transformative (rather than incremental) economic change?
  • Do capability benchmarks lead economic diffusion by years, or by decades — and is the lag shrinking?
  • Can preparedness policy be designed to be worthwhile under both the fast and slow scenarios?
  • Will official statistics register a fast transition at all, given that intangible investment can make measured productivity dip before it rises?

Sources

  1. A Research Agenda for the Economics of Transformative AI
  2. When Does Automating AI Research Produce Explosive Growth? Feedback Loops in Innovation Networks
  3. Artificial Intelligence and Jobs: Evidence from Online Vacancies
  4. AI Won’t Automatically Make Legal Services Cheaper
  5. Europe and the Geopolitics of AGI: The Need for a Preparedness Plan
  6. Scenarios for the Transition to AGI
  7. The Productivity J-Curve: How Intangibles Complement General Purpose Technologies

The numbers

When AI could match humans at every task

Each point is one forecast of the year by which AI could match or outperform humans at all tasks, by the date it was published; bars show a stated range. 6 forecasts from 6 sources: surveys of AI researchers (their aggregate 50% year) and the leaders of AI companies. Definitions differ (high-level machine intelligence, AGI, “powerful AI”), and a forecast is not a finding. The dashed line marks the year each forecast was made.

AI insiders: 2026–2030AI researchers: 2047–206120102020203020402050206020702017201820192020202120222023202420252026forecast yearAI researcher survey 2016 (AI researchers): 2061. AI researchers' 50% year for AI outperforming humans at all tasks (2016 survey). “when the aggregate prediction put 50% probability at 2061, i.e. 45 years out”AI researcher survey 2016AI researcher survey 2022 (AI researchers): 2059. AI researchers' 50% year for high-level machine intelligence (2022 survey). “The aggregate forecast time to a 50% chance of HLMI was 37 years, i.e. 2059”AI researcher survey 2022AI researcher survey 2023 (AI researchers): 2047. AI researchers' 50% year for machines outperforming humans at every task (2023 survey). “the chance of unaided machines outperforming humans in every possible task was estimated at 10% by 2027, and 50% by 2047”AI researcher survey 2023Amodei (Anthropic) (AI insider): 2026. Earliest year “powerful AI” could arrive. “I think it could come as early as 2026, though there are also ways it could take much longer.”Amodei (Anthropic)AI 2027 (AI insider): 2027. Modal (most likely) year for AGI at publication; medians were later. “2027 was our modal (most likely) year at the time of publication”AI 2027Hassabis 2025 (AI insider): 2030. Year AGI could arrive (“possibly the lower end” of five to 10 years). “maybe in the next five to 10 years, possibly the lower end of that” Implied: interview of August 2025 plus five to 10 years = 2030-2035; central value at the lower endHassabis 2025

AI insidersAI researchersShaded: the middle half of each groupDashed: the year the forecast was made

All 17 numbers from 13 papers, in 5 charts, each with its quote

Who says what

Documented positions only, coded with cited evidence by an AI agent: 16 of 400 mapped researchers so far. Everyone else is unknown: no reviewed evidence, which is neither agreement nor neutrality.

7 agree2 it's complicated7 disagree16 of 400 coded

Agree (7)

It's complicated (2)

Disagree (7)

Economists and computer scientists

Documented positions on this claim of researchers on the co-authorship map, by field (from their OpenAlex topics). Researchers without a documented position are not counted.

FieldAgreeIt’s complicatedDisagree
Economists (241 on the map)5 of 122 of 125 of 12
Computer scientists (51 on the map)Too few to say (4 coded)

Each position with the researcher’s reasoning and cited evidence

What to read

Essentials on pace

Read

  • AI and Growth: Where Do We Stand?Philippe Aghion and Simon Bunel · Federal Reserve Bank of San Francisco · 2024 · Evidence for the claim

    Compares AI with past technological revolutions: 0.8 to 1.3 points more productivity growth a year over the next decade.

  • The Productivity J-Curve: How Intangibles Complement General Purpose TechnologiesErik Brynjolfsson et al. · American Economic Journal Macroeconomics · 2020 · Evidence against the claim

    Why gains from a general-purpose technology show up late: the productivity J-curve of unmeasured complementary investment.

  • Thousands of AI Authors on the Future of AIKatja Grace et al. · Journal of Artificial Intelligence Research · 2025 · Mixed evidence

    2,778 AI researchers put even odds on machines beating humans at every task by 2047, 13 years earlier than a year before.

Know

Follow

  • Erik Brynjolfsson agreesEconomist, Stanford University

    Co-author of the productivity J-curve, which explains why a technology's gains can arrive with a delay.

  • Daron Acemoglu leans disagreeEconomist, Massachusetts Institute of Technology

    Has a documented position that transformative effects will take longer than a decade.

  • Justified Posteriors (Andrey Fradkin)Newsletter · 7 posts in this guide

    Discusses forecasts of AI's economic effects as they appear.

All essentials on pace, for 10 minutes to an afternoon

Where the papers point

26 agree57 it's complicated28 disagree111 of 313 coded

How each published paper’s own finding bears on the claim, read from its title and abstract by an AI model (AI-written). 111 of the 313 published papers address this question; the rest are not counted. This codes findings, not authors.

Findings that support the claim (26)

and 23 more in the research library

Findings that it depends (57)

and 54 more in the research library

Findings that go against the claim (28)

and 25 more in the research library

All research on this questionNews on this question