When does AGI arrive? The predictions compared, 2027 to never
Kurzweil holds to 2029, Amodei extrapolates to 2026 or 2027, Altman expects very capable systems by the end of the decade, and Bostrom, Yampolskiy and chip analysts give their own dates and doubts. Seeing the forecasts and their reasoning together helps readers weigh the confidence behind each.
What range do the AGI forecasts in these conversations cover? Show answer
Roughly 2026 to the early 2030s. Nobody in the conversations now says never.
Why is the narrow spread of dates misleading? Show answer
The forecasters mean different things, from passing a long Turing test to outperforming Nobel laureates in millions of copies.
What is Ray Kurzweil's forecast and its basis? Show answer
A computer passes a hard Turing test by 2029. It rests on 80 years of exponential growth in computing per dollar.
How have expert estimates moved since Kurzweil's 1999 prediction? Show answer
From an average of 100 years, with 20 percent saying never, down to about 2030.
What does Dario Amodei mean by powerful AI? Show answer
A system smarter than a Nobel Prize winner across most fields, able to work autonomously, run as millions of fast copies.
How confident is Amodei in 2026 or 2027? Show answer
He calls it an eyeballed extrapolation, not a scientific prediction, and names data and chip supply as risks.
What is Sam Altman's position on the question? Show answer
It is poorly formed because definitions vary, but he expects quite capable systems by the end of the decade.
What is Roman Yampolskiy's asymmetry? Show answer
Ten times the compute gives better performance, but ten times the money does not give ten times the safety.
Why does Nathan Lambert discount the earliest forecasts? Show answer
They come from company heads with incentives to promise early dates. He expects it after 2030.
What does Dylan Patel say will slow things down? Show answer
The cost of running advanced reasoning at scale, and errors that compound over many chained steps.
Why is arrival not the same as impact? Show answer
Physical limits, institutions, cost and trust slow adoption. Amodei expects five to ten years of change after the capability exists.
What should you ask before comparing two forecasts? Show answer
What each would count as success, and what position and incentives the forecaster has.