AGI Timelines
Why the Smartest People in AI Can't Agree on When It Will Change Everything
The people building AGI disagree about when it will arrive by decades — and that disagreement isn't a bug in their thinking, it's a window into how strange this moment actually is.
The Idea
AGI — artificial general intelligence, the point at which a machine can perform essentially any intellectual task a human can — has a timeline problem. Not because researchers lack data, but because they're arguing about something that has never happened before, using frameworks built for things that have. The forecasts range from 'within a few years' (Sam Altman, Demis Hassabis) to 'decades away, if ever' (Gary Marcus, Yann LeCun). This isn't the normal spread of expert disagreement you'd find in, say, economics or climate modelling, where people are working from shared priors and quibbling over parameters. Here, the experts are divided on what AGI even is, which means they're not really arguing about timing — they're arguing about definitions wearing timing's clothes. There's a useful distinction to keep in mind: current AI systems are extraordinarily powerful but narrow. They're trained on patterns, not causes. They can ace a bar exam but can't notice that the exam room is on fire. What AGI would require — genuine flexible reasoning, causal understanding, transfer of knowledge across radically different domains — may be a small increment from where we are, or it may require a fundamental conceptual breakthrough we haven't had yet. Nobody knows. And that uncertainty is honest, not a failure of intelligence. When the range of expert estimates spans 3 years to 300 years, the correct response isn't to pick the middle — it's to understand why the uncertainty is so vast.
In the World
In 2023, Geoffrey Hinton — widely called the 'godfather of deep learning', whose foundational work made modern neural networks possible — resigned from Google and announced he was worried about AI risk. His reasoning wasn't that AGI was imminent. It was subtler: he said he had changed his mind about how close we might be, and that the speed of progress had surprised even him. This matters because Hinton spent decades as one of the most sceptical voices against AI hype. He won the Turing Award, the field's highest honour, and he was not prone to apocalyptic thinking. When someone with that track record says the situation warrants genuine alarm — not panic, but serious attention — it shifts the texture of the conversation. At almost exactly the same time, Meta's chief AI scientist Yann LeCun (also a Turing Award winner, also foundational to deep learning) was publicly arguing the opposite: that current AI architectures are fundamentally incapable of producing human-like understanding, and that the discourse around AGI timelines was dangerously overheated. Two of the three people who defined modern AI, reaching opposite conclusions from the same evidence. What this illustrates isn't that one of them is wrong and one is right. It's that we are in genuinely uncharted epistemic territory — and the map-makers are disagreeing about the shape of the continent.
Why It Matters
Most of us engage with AGI timelines as spectators — half-listening to headlines, vaguely tracking whether to be excited or afraid. But the timeline question shapes decisions happening right now: how governments regulate AI, how companies invest in it, how researchers prioritise safety work, how education systems prepare the next generation. If AGI is ten years away, safety research is urgently underfunded. If it's a century away, current alarm might distort resources away from nearer problems — bias in hiring algorithms, surveillance infrastructure, the displacement of knowledge workers — that are already causing real harm today. The honest intellectual move isn't to pick a side. It's to hold the uncertainty without collapsing it. To notice that the people who are most confident tend to be the ones with something to sell — a funding round, a book, a narrative about their own importance. And to take seriously that something unprecedented is underway, even if its destination isn't yet visible. That's not fence-sitting. That's calibration — arguably the most important cognitive skill for navigating a world where the experts genuinely don't agree.
A Question to Ponder
When you hear a confident prediction about when AGI will arrive, what would it actually take for that person to be wrong — and are they the kind of thinker who has genuinely asked themselves that?
Get a new one of these every morning.
Start learning with Thinkable