Text size
A
A+
A++
A+++
Somewhere between "next Tuesday" and "never," the smartest people alive cannot agree on when artificial general intelligence arrives — and the size of that disagreement is the most important, least discussed fact in technology right now. When the CEO of the world's leading AI lab says AGI could land in five years while the median researcher says forty, that's not a rounding error. That's two groups looking at the same evidence and seeing different centuries.
I'm not here to sell you a doomsday clock or a rapture date. I'm here to lay out what's actually being claimed, by whom, and — the part nobody wants to examine — why their answers are so wildly different. Because once you see the pattern in who predicts what, the whole conversation snaps into focus.
What Hassabis Actually Said — and Why It Landed Hard
Start with the most credible voice in the room. Demis Hassabis runs Google DeepMind, won a Nobel Prize for AlphaFold, and is temperamentally cautious — which is exactly why his recent acceleration got attention. He now puts roughly a 50% chance of AGI by 2030, having narrowed his own estimate from a comfortable 2030–2035 down toward the end of this decade, as Sherwood News tracked. At India's AI Impact Summit he described AGI's coming impact as "roughly ten times that of the Industrial Revolution, unfolding over a decade rather than a century."
Then he did something more interesting than predict. In a July 2026 manifesto titled A Framework for Frontier AI and the Dawning of a New Age, Hassabis called for a U.S.-based AI watchdog with authority to screen advanced models globally — industry-funded but staffed by independent technical experts, with the power to coordinate slowdowns if the danger escalates, per The420. His starkest warning: within roughly eighteen months, biological and nuclear-adjacent capabilities could be baked into open-source models and slip beyond any government's reach.
Read that twice. The man building the technology is asking to be regulated faster than governments currently want to move. When the accelerator asks for a brake, pay attention.
The Optimists: 2027 and the Coding Feedback Loop
Hassabis isn't even the most aggressive. Dario Amodei, CEO of Anthropic, has floated AGI-level systems arriving around 2027, possibly sooner, and his reasoning is specific rather than vibes: AI is now writing meaningful amounts of code, including code that improves AI, and that feedback loop compresses timelines in a way linear intuition misses. Sam Altman of OpenAI has been vaguer but directionally similar, framing it as "a few thousand days" — landing him somewhere around the early 2030s in his essay The Intelligence Age. And then there's Elon Musk, who claimed AI "smarter than the smartest human" by 2026, a prediction best filed under "motivated optimism," as aggregated by AIMultiple.
Notice what these voices have in common. They are the people raising capital, recruiting talent, and selling the future. That's not an accusation of dishonesty — belief and incentive genuinely overlap here — but it's a variable you cannot ignore when weighing the forecast.
The Skeptics: "Retire the Word Entirely"
Now the other side of the room, which gets far less airtime. Yann LeCun, one of the three "godfathers of AI" and Meta's chief AI voice, argues the term AGI should be retired altogether. His position: today's large language models, however impressive, are missing something fundamental about how intelligence works, and human intelligence is too specialized to be a clean target for "general" replication. He prefers "advanced machine intelligence" and thinks the current architecture won't get us there by itself.
He's not alone. A large contingent of working researchers see fluent text generation and mistake it for understanding — a system that can describe physics brilliantly while having no model of a falling glass. Their critique isn't that progress is fake; it's that we're measuring the wrong thing and calling the finish line by the wrong name. And they've been right before: the field has a long history of declaring human-level intelligence "a decade away," a claim that's been roughly a decade away for about seventy years now. Skepticism isn't pessimism here — it's pattern recognition from people who've watched the hype cycle rhyme.
The Number Almost Nobody Quotes
Here's the fact that should anchor this entire debate, and it rarely makes the headlines. When you survey the actual research community — not the CEOs — the picture changes completely. Across roughly ten surveys involving more than 6,000 AI researchers, the 50%-probability estimate for human-level machine intelligence lands somewhere between 2040 and 2060, per AIMultiple's synthesis. A 2023 survey of nearly 2,800 researchers pointed to around 2040. Earlier ones drifted toward 2059.
So the honest state of expert opinion isn't "AGI in five years." It's "the executives building it say five, the scientists studying it say thirty-plus, and the gap between them is a canyon." Both groups are brilliant. Both are looking at the same models. The difference is largely about incentive and definition — which brings us to the real problem.
The Reason Nobody Can Agree: We Never Defined the Word
Every prediction above quietly smuggles in a different meaning of "AGI." Some mean "can do most economically valuable work a human can." Some mean "matches a skilled expert across domains." Some mean genuine, self-directed, human-like understanding. There is no accepted scientific definition of human-level intelligence, which means these forecasts aren't even measuring the same thing. Ask "when do we reach AGI?" without agreeing on what AGI is, and of course the answers span decades.
This isn't pedantry — it's the whole ballgame. A loose definition ("AI that's economically transformative") arguably has a foot in the door already. A strict one ("a machine that thinks like a person, understands the world, and reasons across anything") might be generations away, or might require a breakthrough we haven't imagined. The word is doing enormous, unexamined work in every headline you read.
How to Think About It Without Losing Your Mind
So what do you actually do with this? A few things I'd stake my own bets on. First, stop waiting for a single "AGI arrives" moment — the transformation is a gradient, not a doorway, and it's already reshaping coding, art, medicine, and research right now regardless of what we call it. Second, weight predictions by incentive: when someone raising a $10 billion round tells you the payoff is imminent, discount accordingly — and when a cautious Nobel laureate suddenly accelerates and asks for regulation, weight that heavily. Third, watch capabilities, not calendars. The question that matters isn't "is it AGI yet?" It's "what can it do this year that it couldn't last year?" — and that answer keeps getting more unsettling.
The countdown, in other words, is real — but it's not counting down to a bell that rings. It's counting the distance between what machines could do yesterday and what they'll do tomorrow, and that distance is shrinking whether or not we ever agree on the word for the finish line. The five-year camp and the thirty-year camp are arguing about the map. The territory is changing under all of us either way.
Which leaves the only question worth asking: not when it arrives, but whether we'll have built the wisdom to handle it before the capability outruns us. On that timeline, even the optimists sound late.
