Demis Hassabis expects artificial general intelligence within two to three years, according to a Wall Street Journal profile published on 21 October 2026. Earlier this year, his public estimate was three to five years. The updated timeframe points broadly to 2028–2029—but it is not a promise of delivery on a specific date.
What exactly did Hassabis say?
Hassabis describes AGI as a system spanning the full cognitive range of the human brain. That distinction matters. AI may outperform humans in coding, mathematics, games or protein-structure analysis while remaining unreliable at learning new tasks, planning in unfamiliar environments, or making sound judgments outside its training distribution.
His shorter timeframe is an expert expectation, not a measured probability. The interview does not assign a 70%, 90% or other specific likelihood. Cortist will record the timeframe as stated, rather than reverse-engineering a percentage he did not give.
Why a faster timeline does not settle the question
The problem is partly scientific. A system may meet several headline benchmarks without showing human-level breadth, robustness or transfer to unknown problems. In March 2026, Google DeepMind proposed a cognitive framework for measuring progress toward AGI, acknowledging the need for better empirical tools. Independent tests—and the definition of success—are therefore as important as impressive model demos.
What should we watch next?
- Novel-task transfer: independent evidence that systems reliably solve unfamiliar problems without extensive task-specific engineering.
- Broad capability: evaluations covering different cognitive domains, not only language or coding.
- Verification: external audits, reproducible tests and access sufficient to check claims.
- Revisions: dated changes in forecasts from leading researchers or forecasting communities, recorded without retroactive edits.
How does this affect Cortist's forecast?
We have published a provisional v0.1 editorial estimate of 38% for independently verified AGI by the end of 2030. The earlier 45% launch draft is preserved in the model version history. This is Cortist's working judgment, not Hassabis's number; it is neither locked nor scored because the pass/fail definition remains under review.
Future AGI reporting will be grouped on the permanent Cortist AGI topic hub. Relevant news may strengthen or weaken the evidence, but will not silently rewrite the initial probability.
Sources
Ben Cohen, The Wall Street Journal, 21 October 2026 — interview and revised timeline.
Google DeepMind, 17 March 2026 — cognitive evaluation framework.
Metaculus, general AI forecasting question — live external comparator; its estimates may change.