Shane Legg

“I think there's a 50% chance that we have AGI by 2028. Now, it's just a 50% chance. I'm sure what's going to happen is we're going to get to 2029 and someone's going to say, 'Shane, you were wrong.' Come on, I said 50% chance.”

Shane LeggResearcher

Co-founder and Chief AGI Scientist, Google DeepMind

Deadline is end of 2028. Legg has restated this 50%-by-2028 estimate consistently since first making it in the 2000s, and as of 2026 he still frames 'minimal AGI' as roughly two years out.

“My prediction for the last 10 years has been for roughly human level AGI in the year 2025... This year I've tried to come up with something a bit more precise. In doing so what I've found is that while my mode is about 2025, my expected value is actually a bit higher at 2028. Perhaps more useful is my 90% credibility region, which from my current belief distribution comes out at 2018 to 2036.”

Shane LeggResearcher

AI researcher, Gatsby Computational Neuroscience Unit, UCL; later co-founder and Chief AGI Scientist, Google DeepMind

Legg's log-normal AGI distribution (mode 2025, mean 2028, 90% range 2018–2036) remains open; he has restated the 2028 expectation repeatedly since.

“My more serious attempt at an AGI prediction was a probability of 50% by 2028 and 90% by 2040, which I've roughly held since 1999 and have repeated in many places.”

Shane LeggResearcher

Co-founder and Chief AGI Scientist, DeepMind

Legg's long-standing distribution puts 50% on AGI by 2028 and 90% by 2040; both horizons are still in the future.

“Asked by what year he would assign a 10%/50%/90% chance of the development of human-level machine intelligence, assuming no global catastrophe halts progress: 2018, 2028, 2050.”

Shane LeggResearcher

Co-founder, DeepMind Technologies; author of the PhD thesis 'Machine Super Intelligence'

The 10% milestone year (2018) passed without human-level machine intelligence by most definitions, but the forecast's 50% (2028) and 90% (2050) points remain open.

“I'd also like to add to this prediction that I expect to see an impressive proto-AGI within the next 8 years. By this I mean a system with basic vision, basic sound processing, basic movement control, and basic language abilities, with all of these things being essentially learnt rather than preprogrammed. It will also be able to solve a range of simple problems, including novel ones.”

Shane LeggResearcher

Co-founder, DeepMind Technologies

By the end of 2019 impressive learned systems existed separately for vision, speech, control and language (e.g. GPT-2, AlphaStar), but no single system combined all four modalities with novel problem solving; the PredictionBook entry tracking this claim was judged wrong on 31 December 2019. Integrated multimodal agents such as Gato and GPT-4V only arrived in 2022–2023.

“Over the last year computer power has increased as expected, and so it looks like we're still on target to have supercomputers with 10^18 FLOPS around 2018.”

Shane LeggResearcher

Co-founder, DeepMind Technologies

In 2018 ORNL's Summit reached roughly 1.9 exaops on a mixed-precision genomics application, but sustained double-precision exascale (10^18 FLOPS on HPL) was not achieved until Frontier hit 1.1 exaflops in May 2022 — about four years later than 'around 2018'.