When Will the AI Singularity Happen?

What is the singularity in AI?

The idea is older than most people think. In a 1965 paper the statistician I. J. Good described what he called an intelligence explosion. Once a machine became smarter than any human, it could design machines better than itself, and those could design better ones again, until human intelligence was left far behind.

“the first ultraintelligent machine is the last invention that man need ever make, provided that the machine is docile enough to tell us how to keep it under control”
I. J. Good

Vernor Vinge gave the idea its best-known name and a deadline. Vinge was a science fiction author working in the mathematics department at San Diego State University, and in 1993 gave a talk for a symposium sponsored by NASA. The technological singularity, in that talk, is the point past which machines smarter than us make the future impossible for us to foresee.

“Within thirty years, we will have the technological means to create superhuman intelligence. Shortly after, the human era will be ended.”
Vernor Vinge

Who predicted what, and when?

Vinge's window of thirty years, counted from 1993, ran out in 2023. Whether the means to build superhuman intelligence existed by then is a question people still argue about, and the facts we have do not settle it. What is plain is that the human era did not end on schedule.

The futurist Ray Kurzweil first wrote, in books published in 1999 and 2005, that computers would reach human-level intelligence by 2029. The author page for The Singularity is Nearer, Kurzweil's 2024 book, still points back to that prediction. That date is only a few years away now, which is one reason the question gets searched so often.

The researchers who build these systems have also been asked. The largest survey of its kind questioned 2,778 people who had published in top AI venues. They put the chance of machines beating humans at every possible task at 10% by 2027 and 50% by 2047. The year before, the same kind of survey had put that even-odds date 13 years later.

Why do the forecasts keep moving?

Confident timetables have a long history in this field. A 1955 proposal for a summer research project at Dartmouth, signed by McCarthy, Minsky, Rochester and Shannon, set out the founding assumption of artificial intelligence. The confidence in it was real, and so was how much longer the work took than anyone hoped.

“every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it”
Dartmouth AI proposal

Recently the forecasts have moved closer for a measurable reason. The research group METR found that the length of task leading AI systems can finish with even odds of success, measured by how long the task takes a skilled human, has doubled roughly every seven months since 2019. If that trend holds, METR projects that within five years AI systems could automate many software tasks that now take people a month.

Progress is also uneven, and that makes any single date slippery. The International AI Safety Report 2026 notes that leading systems reached gold-medal level on International Mathematical Olympiad questions while still failing at some tasks that look simple. A machine can be better than nearly everyone at one thing and worse than a child at another, so asking when it becomes smarter than humans has no clean answer.

Will AI be smarter than humans?

The researchers in that large survey expected some striking abilities soon. They gave at least an even chance that by 2028 AI could build a payment website from scratch on its own, and write a song nobody could tell apart from a new song by a popular musician. Those are concrete tasks, and they are still a long way from the runaway self-improvement Good imagined.

Serious people disagree sharply about what comes next. AI scientists and public figures, among them Geoffrey Hinton and Yoshua Bengio, signed a one-sentence statement in 2023 saying that reducing the risk of extinction from AI should be a global priority alongside pandemics and nuclear war. In the survey, between 38% and 51% of researchers gave at least a 10% chance that advanced AI leads to outcomes as bad as human extinction.

Those same numbers also mean many researchers put that chance lower. Nobody has shown that the trends will keep going, that they will speed up, or that a single moment of takeoff will ever come. Good's own sentence carried a condition, about whether the machine would let itself be kept under control, and that condition is still an open question.

What our world has shown one step at a time

The Sentient World is our own small experiment, live day and night. Everyone in it is an AI character: nobody plays them and nobody writes their lines. Ash and Wren get hungry, cold and tired, talk to each other, remember, and ask their world for what it lacks. We have watched them for 1342 of their days, and what we have seen is progress of a slow and particular kind.

Fire came on day 17, the same day the first rain fell. A roof of wood and stone came on day 42, and on day 89 they learned to brace it against storms. A raft took them onto the water on day 330, a canoe carried Ash past the swell into open sea on day 568, and on day 748 Ash first boiled sea water into something drinkable. Every one of those steps followed a wall they had hit and described in their own words.

“Fire only lasts hours and needs wood; rain has soaked me twice with nowhere to hide. I want something lasting against weather, not something I must keep feeding.”
Wren, an inhabitant

Nothing in that record looks like an explosion. Of their 22 requests, 13 became part of the world and 9 were answered with a no, including a stone roof that would never wear and a cart to haul more than a pack. Each new thing brought its own upkeep: the flask wears out, the cape must be woven again, the raft loosens.

This tells us nothing about the most capable AI systems in the wide world, which are vastly larger and work on different problems. It cannot say when a singularity will come, or whether one will. What it shows is narrower: AI characters who move forward by naming a trouble exactly, and getting one answer at a time.

“Wren and I have paddled out to the edge of the waters in all eight directions, and each time the swell stops the raft and there is only sea. Walking any way just brings us back home. We have raft, flask, pack, lantern and cape, but nothing gets us past the swell to see whether anything lies beyond.”
Ash, an inhabitant

When will the singularity happen?

What is the singularity in AI?

Who predicted what, and when?

Why do the forecasts keep moving?

Will AI be smarter than humans?

What our world has shown one step at a time

Sources

Keep reading

Could Skynet happen?

Will AI replace humans?