In 1965, a scientist named Herbert Simon said machines would do any job a human could do within twenty years. That was sixty years ago. We’re still here, still making the exact same promise, just with a newer decade stapled to it. Something about that keeps happening and it’s worth asking why.
I used to think that was strange. Now I think it’s just an honest version of what a lot of us are doing quietly in our heads. We’re all watching for the moment a machine starts thinking the way we do. Not because we need it to sort our email faster. Because something in us wants to know if it can be done at all, and what happens to us once it is.
So why are we so obsessed with this. Not “will AI help with work,” that question got answered years ago. The bigger, weirder question is why humans want, so badly, to build something that matches or beats human intelligence itself. Is it a psychological itch, a philosophical hangover, or just business as usual dressed up as destiny. I think it’s all three, tangled together, and pulling them apart is worth doing.
The psychological itch nobody names out loud
Start with the weird part. Building something smarter than you is, on some level, an act of ego. Not always a bad kind. Parents raise kids hoping they’ll surpass them. Scientists build tools hoping the tools outlast their careers. But AGI is a strange version of that instinct because the “child” here doesn’t need to be loved, doesn’t age, and might not need us once it arrives.
There’s also a control angle that people don’t say out loud enough. A lot of the money and talent racing toward AGI comes from people who spent their careers being the smartest person in the room. Building the smartest thing in the room is, weirdly, a way of staying in control of that room even after you stop being the smartest one in it. If you’re the one who built the god, you get some say in how it behaves. At least that’s the story people tell themselves.
I’ll admit I used to buy this framing completely, ego and control and nothing else. That’s mostly true, though I’m oversimplifying it a bit. There’s a gentler version too. Some of the researchers I’ve read interviews with genuinely talk about wanting to see what’s possible, the same instinct that sends people to climb mountains that don’t need climbing. Curiosity is real. It’s just rarely the only ingredient, and it’s almost never the ingredient that shows up in the funding pitch.
And there’s the immortality angle, which people are more comfortable admitting now than they were five years ago. If you can’t live forever, maybe you build something that carries your thinking forward. It’s not a new idea. It’s an old one wearing a GPU cluster instead of a stone tablet.
The old philosophical dream, and why the goalposts keep moving
This chase is not new. Herbert Simon, one of the founding figures of AI research, said back in 1965 that machines would soon be able to do essentially any work a person could do, and he gave it a twenty-year window. That was sixty years ago. We’re still here, still writing the same kind of prediction with a different decade attached.
What’s changed isn’t the dream. It’s the definition. “General intelligence” sounds like a fixed target until you try to pin it down, and then it slides. Does it mean passing every human exam? Does it mean doing remote knowledge work as well as a competent adult? Does it need a body and hands too, or is a text window enough? The team behind Metaculus’s AGI question actually bundles in general robotic capability as one of its four conditions, which some researchers think is too strict, since robotics is lagging years behind language and reasoning work. So even the measuring stick keeps changing shape depending on who’s holding it.
There’s a philosophical reason this matters beyond pedantry. If you can’t agree on what the finish line looks like, you can’t agree on how close you are to it, and that’s exactly the mess we’re in right now. Some people are measuring “AGI” as a system that automates most cognitive labor better and cheaper than a human. Others want something closer to a machine that can genuinely reason about the world the way a person does, uncertainty and contradictions included. Those are different races with different finish lines, and most public debate doesn’t bother separating them.
The business mentality, which is less philosophical and more urgent
Here’s the part that’s less about human nature and more about quarterly incentives. Once one lab convinces investors that AGI is close and valuable, every competitor has to either match that bet or explain to their own investors why they’re not making it. That’s not a conspiracy. It’s just how capital behaves when a prize looks winner-take-most.
Dario Amodei, Anthropic’s CEO, is a good case study here because his own predictions have actually moved in both directions over the past couple of years. Researchers who track named forecasters’ public statements over time noted that through 2025 Amodei was among those pushing his AGI timeline further out, in the same direction as the Metaculus community and forecaster Peter Wildeford. Then, in the early months of 2026, several of these same people, including Amodei, pulled their estimates back in as progress from newer models sped up faster than expected. Demis Hassabis at Google DeepMind shows up in that same tracked group. It’s a small detail, but it tells you something real: even the people closest to the work aren’t in agreement with themselves year over year, let alone with each other.
Then there’s Jensen Huang, whose company sells the chips everyone else needs to even attempt this race. Back in March 2024 he predicted that within five years AI would match or beat human performance on any test, which puts his date around 2029. Notice the job. The person selling GPUs has a very direct financial interest in everyone believing the timeline is short. I’m not saying he’s wrong. I’m saying it’s worth remembering who benefits from which number.
Elon Musk has gone further still, at one point putting AGI’s arrival at 2026, this very year, while separate academic surveys of AI researchers put the median guess for the same question somewhere between 2041 and 2061. That’s not a small gap. That’s the difference between “check back next quarter” and “this is a problem for someone else’s career, maybe someone else’s grandkids.”
There’s a version of this business pressure that’s less about any one CEO and more about the whole shape of the industry. Once a lab raises money on the promise that it’s close to something world-changing, it can’t easily walk that promise back without spooking the people who wrote the checks. So the public timeline and the internal timeline start to drift apart, not always dishonestly, just because one has to sound confident to a board and the other has to survive contact with actual lab results. I’ve watched enough product launches in my own small corner of tech to recognize the shape of it. You say six weeks because eight weeks doesn’t get funded, and then you spend six weeks pretending eight weeks was always the plan.
So how close are we, actually
This is where I have to be honest and say the real answer is messier than any headline wants it to be.
As of February 2026, forecasters on Metaculus were putting a 25 percent chance on AGI arriving by 2029, and a 50 percent chance by 2033. Those numbers have shortened a lot from a median guess of fifty years away as recently as 2020. That’s a big shift in six years. But a separate, more careful survey effort tells a slightly different story. The Forecasting Research Institute runs a monthly panel called LEAP that tracks AI scientists, industry leaders, and professional forecasters. When they were asked about an AI hitting 80 percent success on software tasks needing eight or more hours of expert human effort, the median expert on that panel put a 50 percent chance on it happening by 2030 or earlier. Professional superforecasters guessed 2028. Ordinary members of the public guessed 2037.
Notice the pattern there. The people closest to building this stuff, and the people paid to forecast things for a living, both guess sooner than ordinary people do. Make of that what you want. I go back and forth on whether that means the forecasters know something the public doesn’t, or whether being close to a thing for years just makes you worse at seeing how far away it still is.
There’s a benchmark angle too that’s genuinely useful, not just noise. METR, a research group that measures how long a task an AI model can complete on its own with 80 percent reliability, has been tracking a “task horizon” number over time. Around the middle of 2026, panelists on that same LEAP survey put the median forecast for how long that horizon would stretch by year’s end somewhere between three and four hours, up from about ninety minutes when the survey period started back in April. On May 8, 2026, METR added an early preview of Anthropic’s Mythos model into its benchmark, and honestly, nobody’s fully sure yet what that preview number is going to mean for the rest of the year’s forecasts. That’s a loose thread still hanging as I write this. It might resolve into something big. It might resolve into nothing.
So where does that leave us? Somewhere between “not imminent” and “not decades away either,” which I know is an unsatisfying answer, but pretending there’s a clean number is worse than admitting there isn’t one. If you forced me to pick a lean, I’d say the business incentives and the recent benchmark jumps make the shorter end of the range, call it the early-to-mid 2030s, more believable to me than the 2050s crowd. That’s a real lean, not a shrug. I could be wrong about it, and I’d rather say that plainly than pretend otherwise.
What happens after, assuming it happens at all
Nobody agrees on this part either, which somehow gets less coverage than the timeline debate, even though it matters more.
The optimistic case says an AGI-level system speeds up scientific research, drug discovery, materials science, the slow stuff that currently takes a human researcher a career to move an inch on. The pessimistic case says most of the economic value gets captured by whoever owns the model, and the labor disruption lands faster than any policy response can catch up to. Both cases get argued by smart, credentialed people, and I don’t think either side has earned the confidence they usually speak with. The labor argument in particular deserves more skepticism than it gets; estimates of job impact from serious economists already range from a small slice of tasks automated to hundreds of millions of roles affected, and when the range is that wide, nobody actually knows.
What I keep coming back to is smaller and stranger than either scenario. If a machine really does end up matching human general intelligence, the question stops being about jobs or GDP and becomes something closer to this: what was actually special about human thinking in the first place. Was it ever the reasoning itself, or was it always the fact that a person, with a body and a mortality and a stake in the outcome, was doing the reasoning? I don’t have a tidy answer to that, and I’m suspicious of anyone who claims they do.
I think about this more than I probably should, usually late at night, usually after reading yet another survey with a hundred names in a footnote. Here’s what nags at me. Every past technology that got compared to “thinking machines,” the loom, the calculator, the search engine, ended up automating a piece of human work while leaving something untouched that we later decided was the actual point. Maybe that keeps happening. Maybe this time it doesn’t, and that’s the whole difference between AGI and everything that came before it. I genuinely don’t know which one it is, and I’d trust anyone less who told me they did.
Simon was off by decades, not years, and that should worry anyone repeating his kind of confidence today. But here’s the part I keep sitting with. He wasn’t wrong about the direction, just the distance. Every generation since him has made the same bet and lost the same way, and every generation still made the bet anyway. Maybe that’s the actual answer to why we’re chasing this thing. Not because we know when it arrives. Because something in us can’t stop guessing.