Will AI take over the world?

The short answer

Probably not, but serious people think the risk is real enough to work on. Most AI researchers put the chance of a catastrophe on the scale of human extinction at a few percent, and the heads of the biggest AI labs have signed a statement that "mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war" (Center for AI Safety, 2023). Others, like Yann LeCun, think the fear is overblown. Work out your own number.

Three kinds of takeover

1. The quiet one: jobs and decisions

AI takes over more and more work and decisions, from hiring to loans to news feeds. This is already happening. The question is who benefits: see which jobs are most exposed.

2. The human one: power

A company or government uses powerful AI to gain power nobody can check: surveillance, propaganda, cyberattacks. Many experts think this is the nearest real danger.

3. The movie one: loss of control

AI systems pursue goals we did not intend and cannot stop. There is no evidence this is happening now, but labs have found models that deceive testers or resist being shut down in controlled experiments, which is why safety research matters.

Warning signs to watch

• AI systems that can copy themselves or get money and computing power on their own.
• Labs racing to release without safety testing.
• Critical infrastructure (power, finance, weapons) handed to AI without a human in the loop.
• AI that finds and uses software vulnerabilities by itself.

What is being done

Governments have set up AI safety institutes in the UK, US and elsewhere; the EU AI Act regulates high-risk uses; and AI labs publish safety policies and test models for dangerous abilities before release. Critics say it is not enough or not fast enough.

More AI worries, sorted

AI apocalypse tools