What is shaping itNo current evidence changes this area.
Our p(doom) - Will we make it?
p(doom) is the estimated probability that artificial intelligence causes an existential catastrophe.
Our number reads recent public evidence. It is an editorial index, not a scientific forecast.
What does p(doom) mean?
The term is shorthand for the chance that AI leads to human extinction or a permanent loss of humanity’s control over its future. Most p(doom) figures are personal estimates. They depend on what counts as catastrophe and the period being considered.
Batjko Labs reviews research, public records and reporting every day. Repeated coverage of the same event counts once, and older evidence fades out.
Estimation history
Each point is a fresh assessment. The previous score is never rolled into the next one.
How should I read it?
Current evidence is grouped into five areas. Each event appears once, in the area it affects most.
What is shaping itNo current evidence changes this area.
What is shaping itNo current evidence changes this area.
What is shaping itNo current evidence changes this area.
What is shaping itNo current evidence changes this area.
What's behind this number?
Today’s estimate reflects published long-range forecasts and our review of current evidence.
Today’s review found no current AI-risk event with enough independent support to raise or lower the estimate.
- Items considered
- 68
- Items assessed in depth
- 11
- Relevant claims found
- 4
- Supported events affecting it
- 0
The estimate starts with these three published long-range forecasts. They ask different questions, so we use them as reference points rather than treating them as one poll.
- AI Impacts 2023 survey5%Human extinction or permanent and severe disempowerment caused by AI; Within the next 100 years.
- Existential Risk Persuasion Tournament1.07%Human extinction caused by AI; By 2100.
- Metaculus question 270351%AI catastrophe reducing the human population below 5,000; Before 2100.
How do we arrive at it?
Published long-range forecasts provide context. Current evidence can move the number.
We review research, public records and reporting, merge duplicate coverage, and look for specific claims about AI risk or safety. An event must have enough independent support before it can affect the estimate. Community posts can point us toward a story, but they cannot move the number on their own.
Supported events can raise or lower the estimate. Their effect fades and ends after two days, or four days for a major event. More stories about the same event do not make it count more. A later report matters only when it adds something materially new.
We calculate every reading from the evidence available that day. Yesterday’s score is shown for comparison, but it is never added to today’s calculation.
Who owns this reading?
Every published reading follows a complete daily review. If the review is incomplete, the previous number stays visible and the page says why.
We do not silently rewrite earlier readings. Corrections show what changed and why. If we have missed or misread something material, write to contact@batjko.com. We correct errors on this page.