When will AI take over the world?
No date exists, and anyone giving you one is guessing. The nearest real numbers: 2,778 AI researchers put a 50% chance on machines outperforming humans at every task by 2047, and full automation of all jobs at 2116. Those are capability forecasts, not takeover dates — and they moved 48 years earlier in one year.
Why — the first-principles explanation
There's no date because "take over" isn't an event you can put on a calendar. It's not a moment when a machine announces itself. If it happens at all, it looks like a slow transfer of decisions — each one small, each one sensible, none of them a headline. The day you'd mark on the calendar would be picked by historians much later, and they'd probably disagree about it.
What does have dates is capability, because you can ask people who build the systems. In 2023, researchers surveyed 2,778 authors from the top AI venues. Median forecast for high-level machine intelligence — unaided machines outperforming humans at every possible task — was a 10% chance by 2027 and a 50% chance by 2047. Full automation of all human occupations: 10% by 2037, 50% by 2116.
That 69-year gap between the two is the most useful thing in this entire answer. The same experts think machines could beat us at everything roughly seventy years before machines actually do everything. Capability is not deployment, and deployment is not control. In between sit law, liability, physical infrastructure, capital, unions, inertia, and the plain fact that somebody has to build the robot and wire the building. Being able to do a job and having the job are different, and the experts price that difference at seven decades.
Now the honest caveat that cuts the other way: these forecasts are unstable. The full-automation median jumped from 2164 in the 2022 survey to 2116 in 2023 — a 48-year swing in twelve months, triggered by ChatGPT. That's not a measurement moving. That's a group of experts being visibly surprised and re-anchoring. It should lower your confidence in the 2047 number too. Expert timelines here are a weather forecast for a climate we've never observed — informative about current expert opinion, close to worthless as a schedule you'd plan around. If someone hands you a specific year, what they've told you is something about themselves, not about the future.
An example that makes it click
Ask when the internet took over the world. It's a real thing that happened — the world genuinely runs on it now. So what's the date? 1969, when the first packet moved? 1993, with the browser? 2007, with the iPhone? 2020, when your job moved into a video window?
Every answer is defensible, which means the question was malformed. There was no takeover day. There was a thirty-year slide during which the answer to "could we function without this?" quietly changed from yes to no, and nobody noticed the crossing. If AI takes over, that's the shape it takes — and you'll be arguing about the date afterward, not marking it on a calendar in advance.
Key facts
- In the 2023 Expert Survey on Progress in AI, 2,778 researchers from NeurIPS, ICML, ICLR, AAAI, IJCAI, and JMLR gave a 10% chance of high-level machine intelligence by 2027 and 50% by 2047 (Grace et al., arXiv:2401.02843, January 2024).
- Median forecast for all human occupations becoming fully automatable: 10% by 2037, 50% by 2116 — a 69-year gap after the 2047 capability milestone.
- The full-automation median moved from 2164 in the 2022 survey to 2116 in 2023 — a 48-year shift in a single year, following the release of ChatGPT.
- The same survey gave a median 5% (mean 16.2%) probability to AI causing human extinction or severe permanent disempowerment, with no date attached.
- A 2023 assessment by 19 researchers including Bengio and Chalmers found no current AI systems are conscious — no threshold event of machines 'waking up' is being forecast by anyone.
▶ The 60-second explainer (script)
When will AI take over the world? There's no date, and anyone who gives you one is guessing. Here's why. Take over isn't an event. It's not a moment where a machine announces itself. If it happens at all, it's a slow transfer of decisions — each one small, each one sensible, none of them a headline. What does have dates is capability, because you can ask the people building it. In 2023, two thousand seven hundred seventy-eight AI researchers gave their median forecast for machines outperforming humans at every possible task: ten percent chance by 2027, fifty percent by 2047. Then they were asked when all human occupations could be fully automated. Answer: fifty percent by 2116. Sixty-nine years later. That gap is the most useful number here. The same experts think machines could beat us at everything about seventy years before machines actually do everything. Because capability isn't deployment, and deployment isn't control. In between sit law, liability, capital, infrastructure, and someone actually having to wire the building. Now the honest catch. These forecasts are unstable. That full-automation date jumped from 2164 to 2116 in one year — a forty-eight-year swing, after ChatGPT. That's not a measurement moving. That's experts being surprised and re-anchoring. Which should lower your confidence in the 2047 number too. A specific year tells you about the person saying it. Not about the future.
What authoritative sources say
People also ask
What about people who say 2027 or 2030?
Those are usually AGI predictions from people with a stake — labs raising money, or forecasters with a public position. The 10% figure for 2027 in the researcher survey is a tail estimate, not a central one.
Why did the forecasts jump so much?
ChatGPT surprised the field's own experts. The full-automation median moved 48 years earlier in twelve months. That volatility is itself information: it means these estimates are opinions responding to demos, not models tracking a trend.
Is 2047 when AI takes over?
No. 2047 is a 50% guess for machines being able to outperform humans at every task. Being able to is not doing, and doing is not governing. The experts put full automation seventy years after that.
Should I plan around any of this?
Plan around a decade or two of rapid task-level automation, which is well supported. Don't plan around a takeover date — no one, including the experts, has one worth acting on.