p(doom)

basicoutlook

p(doom) is a personal estimate of the likelihood that artificial intelligence will cause a fatal, irreversible catastrophe for humanity.

In plain terms

p(doom) is a gut feeling expressed as a percentage: how likely someone thinks it is that the development of artificial intelligence will turn out very badly for humanity. The 'p' is the mathematical symbol for probability, and 'doom' means fate or destruction.

It is like when your friends guess how likely it is that Saturday's garden party will fall apart due to rain. One says 10 percent, another 70, even though they are looking at the same sky. The number is not a measurement; rather, it shows how worried each person is and why.

The concept is useful for expressing a hard-to-capture fear briefly with a single number. A researcher's or CEO's p(doom) roughly reveals where they stand in the debate on AI risks.

It differs from weather forecasting or insurance risk calculation in that there is no historical data series behind it. Such an event has never happened before, so the estimate is based on arguments and assumptions, not statistics.

How it works

First, you must decide what you mean by 'doom': human extinction, the permanent loss of control, or any irreparable, civilization-scale harm.
Afterwards, you consider the chain of events: will superhuman AI emerge, will its goals be aligned with humans', and will there be time and tools to fix errors?
Some assign a separate probability to every step and calculate from these, but most simply give a single number based on intuition, often with a timeframe, for example, 'by 2100'.
You say the number, and others compare it with their own. The real debate is about which step each person sees differently.
The estimate is not final: new research findings, regulations, or an unexpected technological leap may cause many to adjust their own valuations.

An everyday example

In a podcast, the host asks the AI researcher guest: 'And what is your p(doom)?' The guest replies that it is roughly 10 percent, and the rest of the conversation is about why it is not 1 and why not 50.

Who is affected

everyoneYou hear such numbers in the news increasingly often, and it is good to know that these are personal estimates, not measurement results.

lawyer, citizenIn debates on AI regulation, these estimates serve as arguments for how strict the rules need to be. → Law

developerFor anyone building increasingly autonomous systems, such as agents, security questions are part of everyday design decisions. → Code

parent, student, teacherAlarming figures can cause anxiety among young people, so it is important to discuss them in context. → Education & Family

A common misconception

Mythp(doom) is a scientifically calculated probability, so if an expert states 20 percent, that is precisely the level of risk.

In realityp(doom) is a subjective estimate with no unified methodology or definition; values from prominent experts range from near zero to nearly one hundred percent.

Limits and risks

Everyone means something different by 'doom' and estimates a different timeframe, so two figures are rarely directly comparable.
A single number obscures the reasoning behind it, yet the essence is why someone considers the risk large or small.
The estimate can be distorted by interest and attention-seeking: a dramatic figure makes a good headline, while a reassuring figure may serve business interests.
Debates about distant catastrophes can distract from AI harms that already exist today, such as misleading content or biased decisions.

Similar, but not the same

existential risk – Existential risk is the danger itself that threatens humanity's survival, while p(doom) is a personal estimate of the probability of it.
AI safety – AI safety is a research and engineering field working to reduce risks; p(doom) is just a number expressing the perceived magnitude of that risk.
Technological singularity – The singularity is a hypothetical turning point at which the development of machine intelligence accelerates beyond anyone's ability to follow, and its outcome is not necessarily bad.

Related concepts

Broader

AI safety, existential risk, risk assessment

Narrower

–

On the edge

superintelligence, general AI, goal alignment, Technological singularity, AI regulation

A brief history

The term began as semi-joking jargon among AI researchers and internet forums. In 2023, with the rapid spread of large language models, it entered wider media coverage when prominent researchers and CEOs publicly shared their own estimates.

Where things stand

By the end of 2026, p(doom) is a common phrase in interviews and debates about AI risks, but it still lacks a unified definition and calculation method. Expert estimates continue to show enormous variation.

Status: 2026 November

Also known as

p(doom) · P(doom) · pdoom · p-doom · probability of doom · probability of doom · chance of AI doom

Sources

Grace et al.: Thousands of AI Authors on the Future of AI (arXiv, 2024)

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.