Resource guide

AI Doomers and p(doom): What the Fear Really Means

Q: What is an AI doomer?

An AI doomer is someone who believes advanced artificial intelligence poses a serious risk of catastrophe or human extinction. Doomers argue that AI could grow smarter than people and escape human control. Most want development slowed or paused until safety research catches up. The word is sometimes used as an insult by critics, but many researchers now use it plainly.

Q: Are AI doomers right?

No one knows yet, and that is the honest answer. Serious experts disagree by huge margins. Geoffrey Hinton puts the risk at 10 to 20 percent, while Roman Yampolskiy puts it near 99.9 percent, and Meta's Yann LeCun calls it almost zero. A 2024 survey of over 2,700 AI researchers found a median of about 5 percent. The wide spread shows these are opinions about the future, not proven facts.

Q: Who is the most famous AI doomer?

Eliezer Yudkowsky is the best-known AI doomer. He has argued for years that a rogue superintelligent AI is the likely default outcome and that alignment is extremely hard to solve. In 2023 he called for a global halt on training the most powerful AI systems. Geoffrey Hinton, a Nobel Prize winner who left Google to speak freely, is another leading voice.

Q: What is the opposite of an AI doomer?

The opposite of an AI doomer is an accelerationist, often tied to the "effective accelerationism" or e/acc movement. Accelerationists believe rapid AI progress will solve big problems like disease and poverty. e/acc figure Guillaume Verdon ("Beff Jezos") has said the odds of an AI catastrophe are near zero. They call safety-focused critics "decels" or "doomers."

Q: What does a p(doom) of 20 percent mean?

A p(doom) of 20 percent means the person thinks there is a one-in-five chance AI causes an existential catastrophe. It does not mean doom is scheduled or certain. It is a rough personal bet, not a forecast with a fixed date. Many leading figures, including Elon Musk and Yoshua Bengio, land in the 10 to 20 percent range.

Q: Is p(doom) a real scientific measurement?

No, p(doom) is not a scientific measurement. It is a subjective probability, meaning a personal degree of belief. Different people define "doom" differently and use different time frames, so the numbers cannot be compared cleanly. Critics say the metric can make guesses sound more precise than they are, while supporters say it forces people to state their views clearly.

Q: What is the AI ethics camp?

The AI ethics camp focuses on the harms AI causes right now, not on far-future extinction. Researchers like Timnit Gebru highlight problems such as bias, surveillance, labor exploitation, and misinformation. Some in this camp argue that doom talk distracts from these present harms. They form a third group alongside doomers and accelerationists.

A plain-English guide to AI doomers, the p(doom) metric, and the people betting on catastrophe.

Last updated July 12, 2026 1328-word guide Editor Ban the Bots

Key takeaways

An AI doomer is someone who believes advanced AI poses a serious risk of catastrophe or human extinction, and who wants development slowed until safety catches up.
"p(doom)" is shorthand for the probability of doom, meaning the odds a person assigns to AI causing an existential catastrophe; the term spread from the LessWrong rationalist community and hit the mainstream in 2023 after GPT-4.
Nobel laureate Geoffrey Hinton told BBC Radio 4 in December 2024 that AI has a 10 to 20 percent chance of causing human extinction within 30 years.
AI safety researcher Roman Yampolskiy has publicly estimated a 99.9 percent chance that advanced AI wipes out humanity, one of the highest figures on record.
Then-FTC Chair Lina Khan said in 2023 that her p(doom) was around 15 percent, showing the metric reached regulators and policymakers, not just researchers.
The AI Impacts 2024 survey of over 2,700 AI researchers found a median estimate of about a 5 percent chance of human extinction from AI, far below the loudest doomers.
The AI debate splits three ways: doomers focused on extinction risk, accelerationists (e/acc) who want faster progress, and an AI ethics camp focused on present-day harms like bias and misinformation.

An AI doomer is someone who believes advanced artificial intelligence could cause a catastrophe, up to and including human extinction. The label covers a wide range of people, from independent thinkers to Nobel Prize winners. What they share is a belief that AI is moving faster than our ability to control it. This guide explains what AI doomers think, what "p(doom)" means, and how seriously to take the numbers.

What is an AI doomer?

An AI doomer is a person who thinks powerful AI poses a real risk of disaster and wants development slowed until safety improves. The core worry is simple. If we build machines smarter than ourselves, we may not be able to keep them under control.

The word "doomer" started partly as an insult. Critics used it to mock people they saw as too gloomy about technology. Over time, many researchers adopted the term plainly to describe their own views.

What doomers actually believe

Most doomers do not claim doom is certain. They argue the risk is high enough to take seriously. They point out that humans have never controlled anything more intelligent than ourselves.

Geoffrey Hinton, often called a "godfather of AI," put it bluntly on BBC Radio 4. He asked how many examples exist of a less intelligent thing controlling a more intelligent one. His point was that the honest answer is close to none.

To go deeper on the movement's founder, see our explainer on Eliezer Yudkowsky, the most prominent AI doomer.

What is p(doom)?

p(doom) is shorthand for "probability of doom," the odds a person gives that AI will cause an existential catastrophe. People usually express it as a percentage, such as 10 percent or 90 percent. It is a personal estimate, not a measured fact.

The term grew out of the rationalist community, especially the forum LessWrong, more than a decade ago. It was a quick way to state your view on AI risk without arguing over exact definitions or dates.

Why the term spread

p(doom) went mainstream in 2023, right after the release of GPT-4. High-profile figures like Hinton and Yoshua Bengio began warning about AI risk in public. Suddenly, tech workers and reporters were trading p(doom) numbers like sports scores.

There is a catch. "Doom" has no single agreed meaning, and no fixed time frame. One person's "doom" is total extinction, while another's is a permanent loss of human control. That makes the numbers hard to compare directly.

Curious how near this future might be? Read how close we are to AGI for the timeline debate.

Notable p(doom) estimates

Published p(doom) estimates from named experts range from nearly zero to nearly 100 percent. That enormous spread is the single most important fact about the metric. It shows that smart, informed people simply do not agree.

The table below lists verified public estimates. Each figure comes from a named person and a real source or interview.

Person	p(doom) estimate	Source / context
Roman Yampolskiy	~99.9%	AI safety researcher; stated in 2024 interviews that superintelligence would almost certainly end humanity
Dan Hendrycks	>80%	Director, Center for AI Safety; raised his estimate from around 20% to over 80%
Geoffrey Hinton	10–20%	Nobel laureate; told BBC Radio 4 in December 2024 there is a 10–20% chance of extinction within 30 years
Yoshua Bengio	~20%	"Godfather of AI"; said he assigns roughly 20% probability to a catastrophic outcome
Elon Musk	10–20%	xAI founder; said at the 2024 Future Investment Initiative that AI "going bad" is a 10–20% chance
Lina Khan	~15%	Then-FTC Chair; gave her p(doom) as about 15% in 2023
AI researchers (median)	~5%	AI Impacts 2024 survey of 2,700+ researchers; median chance of human extinction from AI
Yann LeCun	<0.01%	Meta chief AI scientist; publicly dismisses existential risk as extremely unlikely

What the numbers tell us

Notice the pattern. The people closest to safety research, like Roman Yampolskiy and Dan Hendrycks, tend to give the highest numbers. The AI Impacts survey of thousands of working researchers landed near 5 percent.

Even the "low" mainstream figures are striking. A 10 to 20 percent chance of extinction, from a Nobel Prize winner, is not a small number for any technology. You can learn more about the science behind these warnings in our profile of Geoffrey Hinton.

Doomers, accelerationists, and AI ethics

The AI debate splits into three main camps that often talk past each other. Understanding them makes the whole argument easier to follow. Each camp cares about a different kind of risk.

1. The doomers (AI safety)

Doomers focus on existential risk, the chance AI ends or permanently disempowers humanity. They want to slow or pause the most powerful AI training until safety catches up. Eliezer Yudkowsky went furthest, calling in 2023 for a worldwide halt on training the largest models.

2. The accelerationists (e/acc)

Accelerationists are the opposite of doomers. The "effective accelerationism" or e/acc movement argues that fast AI progress will cure disease, end poverty, and lift humanity forward. e/acc leader Guillaume Verdon, known online as "Beff Jezos," has said an AI catastrophe has near-zero probability.

Accelerationists often call their opponents "decels" or "doomers" as an insult. Read more in our explainer on effective accelerationism.

3. The AI ethics camp

A third camp focuses on harms happening right now, not far-future extinction. Researchers like Timnit Gebru highlight bias, surveillance, labor exploitation, and misinformation. Some argue that doom talk actively distracts from these present-day problems.

This split matters. To see how these groups clash in public, read who is fighting AI.

How seriously to take AI doomers

Take AI doomers seriously as a warning, but treat their exact numbers with caution. The estimates are subjective bets about the future, not proven forecasts. That is true whether the number is 1 percent or 99 percent.

Here is the case for listening. Some of the most credentialed people in AI, including two "godfathers" of the field, put the risk in the double digits. Hinton even left Google so he could speak freely about it.

The case for caution

Now the other side. p(doom) numbers can make guesses sound more precise than they are. A neat percentage hides how much is pure judgment.

Critics also note that predictions about the future have a poor track record, and that fear can quietly serve business interests. Warning that your product might end the world also makes it sound powerful, important, and worth huge investment.

Another problem is that the same word hides very different claims. One expert's "doom" means total extinction, while another's means a permanent loss of human freedom, so two matching numbers may not mean the same thing at all.

The AI ethics camp adds a sharper point. Focusing on a hypothetical robot apocalypse can pull money and attention away from real harms already hurting people today.

A balanced way to read the numbers

The smartest approach is to hold two ideas at once. The risk is uncertain, and uncertainty is not the same as safety. A 10 percent chance of a terrible outcome is worth planning around, even if it is far from certain.

Watch what the estimates do over time, not just where they sit today. Several experts, including Dan Hendrycks, have raised their p(doom) as AI has advanced faster than expected.

The bottom line

AI doomers are the people who believe advanced AI could cause catastrophe, and p(doom) is how they put a number on that fear. Those numbers range wildly, from Yampolskiy's near-certain 99.9 percent to LeCun's near-zero. That disagreement is the real story.

You do not need to pick a camp to stay informed. You just need to know the terms, the people, and the honest limits of the predictions. The debate over AI doomers is really a debate about how much trust to place in a technology we do not fully understand.

Want to keep up as the p(doom) debate evolves? Get the facts each day with our daily AI briefing.

Frequently asked questions

▸ What is an AI doomer?

An AI doomer is someone who believes advanced artificial intelligence poses a serious risk of catastrophe or human extinction. Doomers argue that AI could grow smarter than people and escape human control. Most want development slowed or paused until safety research catches up. The word is sometimes used as an insult by critics, but many researchers now use it plainly.

▸ What is p(doom)?

p(doom) is shorthand for "probability of doom," the odds a person gives that AI will cause an existential catastrophe. It is usually written as a percentage, like 10 percent or 99 percent. The term started as insider slang in the rationalist community and went mainstream in 2023 after GPT-4. There is no agreed definition of "doom" or a fixed time frame, so the numbers are personal estimates, not scientific measurements.

▸ Are AI doomers right?

No one knows yet, and that is the honest answer. Serious experts disagree by huge margins. Geoffrey Hinton puts the risk at 10 to 20 percent, while Roman Yampolskiy puts it near 99.9 percent, and Meta's Yann LeCun calls it almost zero. A 2024 survey of over 2,700 AI researchers found a median of about 5 percent. The wide spread shows these are opinions about the future, not proven facts.

▸ Who is the most famous AI doomer?

Eliezer Yudkowsky is the best-known AI doomer. He has argued for years that a rogue superintelligent AI is the likely default outcome and that alignment is extremely hard to solve. In 2023 he called for a global halt on training the most powerful AI systems. Geoffrey Hinton, a Nobel Prize winner who left Google to speak freely, is another leading voice.

▸ What is the opposite of an AI doomer?

The opposite of an AI doomer is an accelerationist, often tied to the "effective accelerationism" or e/acc movement. Accelerationists believe rapid AI progress will solve big problems like disease and poverty. e/acc figure Guillaume Verdon ("Beff Jezos") has said the odds of an AI catastrophe are near zero. They call safety-focused critics "decels" or "doomers."

▸ What does a p(doom) of 20 percent mean?

A p(doom) of 20 percent means the person thinks there is a one-in-five chance AI causes an existential catastrophe. It does not mean doom is scheduled or certain. It is a rough personal bet, not a forecast with a fixed date. Many leading figures, including Elon Musk and Yoshua Bengio, land in the 10 to 20 percent range.

▸ Is p(doom) a real scientific measurement?

No, p(doom) is not a scientific measurement. It is a subjective probability, meaning a personal degree of belief. Different people define "doom" differently and use different time frames, so the numbers cannot be compared cleanly. Critics say the metric can make guesses sound more precise than they are, while supporters say it forces people to state their views clearly.

▸ What is the AI ethics camp?

The AI ethics camp focuses on the harms AI causes right now, not on far-future extinction. Researchers like Timnit Gebru highlight problems such as bias, surveillance, labor exploitation, and misinformation. Some in this camp argue that doom talk distracts from these present harms. They form a third group alongside doomers and accelerationists.