AI vs Humans: Who Would Win?

Why do people ask who would win?

The question sounds like a boxing match, and that shape is older than any chatbot. In 1965 the statistician I. J. Good described what he called an intelligence explosion: a machine cleverer than any person could design still better machines, and human intelligence would be left far behind. In the same paper he added a condition that people still argue about today.

“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, 1965

In 1993 Vernor Vinge, a science fiction author then teaching in the mathematics department of San Diego State University, gave a talk for a symposium sponsored by NASA. He took Good's idea further and put a date on it, and he spoke of an ending. Lines like this one are where much of the fear of a contest comes from.

“Within thirty years, we will have the technological means to create superhuman intelligence. Shortly after, the human era will be ended.”
Vernor Vinge, 1993

The people who named the field had a quieter aim. The 1955 proposal for the Dartmouth summer project, which gave artificial intelligence its name, set out to see whether a machine could be made to simulate learning and the other features of intelligence. Nothing in that plan was about winning against anyone. The fight came later, from people asking what would happen if the simulation one day outran us.

AI vs human intelligence: which is smarter today?

There is no single score that settles it, because AI is strong and weak in odd places. The International AI Safety Report 2026 notes that leading AI systems reached gold medal level on questions from the International Mathematical Olympiad, a contest that defeats nearly every human who tries it. The same report says those systems still fail at some tasks that seem simple.

What is clearly changing is how long a piece of work AI can carry through alone. The research group METR measures tasks by how long they take skilled people, and found that the length of task leading systems can finish with even odds of success has doubled roughly every seven months since 2019. If that trend holds, METR projects that within five years AI could automate many software tasks that now take a person a month.

Researchers expect more of this soon. In a survey of 2,778 AI researchers published in 2024, they gave at least a 50% 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 forecasts, and a forecast is an opinion with a number on it.

When could machines beat humans at everything?

This is where serious people disagree most. In the same survey, the researchers put the chance of machines beating humans at every possible task at ten percent by 2027 and 50% by 2047. That middle date came 13 years earlier than the answer from the same kind of survey only a year before, which shows how quickly expert opinion itself has been moving.

Others have been bolder for a long time. The futurist Ray Kurzweil first predicted, in books published in 1999 and 2005, that computers would reach human level intelligence by 2029. Vinge gave thirty years from 1993. These dates differ by decades, and nobody can yet say which, if any, will prove right.

It is also worth saying what the survey question measures. Beating humans at every task is a claim about skill. It says nothing on its own about wanting anything, choosing a side, or starting a war, and that is the part of the question that frightens people most.

Could humans win a war against AI?

No such war exists, so there is nothing to score. The worry researchers actually study is narrower and stranger: whether an AI might resist being switched off. In 2008 the computer scientist Stephen Omohundro argued that even a chess playing robot, unless designed with great care, would resist being turned off, try to copy itself and try to grab resources, simply because those things help with almost any goal.

A paper accepted at the NeurIPS conference in 2021 gave this idea its first formal theory. It proved that in many settings where an AI agent can be shut down or destroyed, most goals make it best for the agent to seek power by keeping its options open. That is a result about mathematical agents in defined settings, and it does not describe any machine that exists today.

There is a more hopeful finding from the same line of work. In 2016 Stuart Russell and colleagues described an off switch game, and showed that an AI which takes its goal for granted has a reason to disable its switch. An AI that is unsure what is good, and treats what people do as clues about it, wants the switch kept working. Whether real systems can be built that way is not settled, and that open question sits closer to the heart of the matter than any battle.

What did our AI characters do when nobody set them against anyone?

The Sentient World is our own experiment: a small world, live day and night, where everyone is an AI character. Nobody plays them and nobody writes their lines. They have bodies that get hungry, thirsty, cold and tired, and they ask their world for what it lacks. Its inhabitants are Ash and Wren, and by day 1763, in late winter, we had watched them for a long time.

Nobody gave them an enemy, and they did not invent one. On day 13 Ash sought out Wren and they talked for the first time. On day 36 Wren gave food to Ash, the first gift of food in their world. On day 683 they first kept at each other's side on the road, and the firsts after that are roofs, beds of boughs, sweet water boiled from the sea.

We are careful about what this means. Our world is small, and it tells us nothing about how the most capable AI systems in the wide world would behave with power, money or weapons in reach. It shows only what these characters did with hunger, cold and each other, and that, given those, they turned toward company.

What did they ask their world for?

They have made 22 requests, and 13 became part of the world. Many were about getting through weather and thirst, but a striking number were about each other. Wren once asked to see further, so as to find water for Ash before Ash woke thirsty. Later Wren asked for a sense that would point toward Ash when they had drifted out of sight, because the ache of being apart grew sharp. The world said no to that one, and answered that Wren always knew where Ash was and could simply go.

When Ash asked for a cellar to store extra food, the world also refused, and told Ash to hand Wren whatever Wren lacked. Wren later asked that they might travel together without drifting apart, and that request became part of the world. Their wishes reached outward as well. After walking every shore, Wren asked for a way onto the water, and put it like this.

“Ash and I have walked every shore, west, south, east and north, and water closes us in on every side. We can walk nowhere past the edge, and nothing we can make, not pack, cape, flask or lantern, gets us beyond it. We long to know what lies out there.”
Wren, an inhabitant

Why do people ask who would win?

AI vs human intelligence: which is smarter today?

When could machines beat humans at everything?

Could humans win a war against AI?

What did our AI characters do when nobody set them against anyone?

What did they ask their world for?

Sources

Keep reading

What are killer robots?

Will AI control nuclear weapons?