AGI Timelines: What Top Researchers Actually Predict
Public AGI forecasts have compressed dramatically since 2020. Understanding why requires looking at scaling trends, expert surveys, and the structural biases that pull predictions in opposite directions.
Key facts
- 2023 AI Impacts survey: 50% probability of high-level machine intelligence by 2047.
- Frontier lab leaders have publicly stated timelines as short as 3–10 years.
- Compute used to train frontier models has grown ~4x per year since 2010.
- Historical AI forecasts have systematically been over-optimistic — and recently, under-optimistic.
What Expert Surveys Show
The largest ongoing survey — AI Impacts' biennial poll of published ML researchers — found in 2023 that respondents gave a 50% probability of 'high-level machine intelligence' by 2047, a 13-year shift earlier than their 2022 answer.
Surveys of researchers at frontier labs (OpenAI, Anthropic, DeepMind) report systematically shorter timelines, often clustering between 2027 and 2035, though selection bias is significant.
Scaling Extrapolations
Since 2020, capability has scaled predictably with compute, data, and parameters. Forecasters like Epoch AI project that GPT-4-class compute will be exceeded by 4–6 orders of magnitude before 2030 if current investment continues.
Whether scaling alone produces AGI is contested. Some researchers argue current architectures will plateau on novel reasoning and embodiment; others see no clear ceiling within the next two scaling generations.
What the Labs Are Saying
Sam Altman (OpenAI) has publicly stated AGI may arrive 'in a few thousand days.' Dario Amodei (Anthropic) has suggested 'powerful AI' could arrive as early as 2026–2027. Demis Hassabis (DeepMind) has cited 5–10 years as plausible.
These statements function as both forecast and strategy: shorter timelines justify accelerated investment, larger safety teams, and earlier policy engagement.
Why Timeline Estimates Are Uncertain
Forecasts are pulled shorter by recent capability surprises and longer by historical track records (AI has consistently been overpromised since the 1950s). Both biases are real.
The honest answer is wide uncertainty: most well-calibrated forecasters give 10–90% probability intervals spanning two to four decades.
Frequently asked
Why have AGI timelines shortened so much?
+
The 2022–2024 capability jumps from GPT-3.5 to GPT-4 to multimodal frontier models exceeded most researchers' expectations, prompting widespread forecast revisions.
Are short timelines just hype?
+
Some are; many are not. Independent forecasters at Metaculus, Epoch AI, and AI Impacts — without commercial incentive — have also shortened estimates substantially.
Sources & further reading
Continue in this series
Foundations
Defining AGI: Why the Term Resists a Single Meaning
Beyond AGI
Superintelligence: What Comes After Human-Level
Philosophy of Mind
Could AGI Be Conscious — and Would It Matter?
Safety
AI Alignment: The Core Technical Challenge
Risk Analysis
Existential Risks from Advanced AI
Economics
The Economic Impact of AGI