Raw data
Every answer, in one file
This is the complete record behind the dial: every valid answer every model has given, one row per answer, with the probabilities it assigned and the explanation it gave. It's a CSV file, which opens in Excel, Google Sheets, Numbers, or any data tool.
Updated every day after the poll runs. Free to reuse under CC BY 4.0 with credit.
What's in it
Each model is asked 7 times a day, so each model has up to 7 rows per day. Answers that weren't usable (no answer, or probabilities that didn't add up to 100%) are left out. The numbers on the site are medians of these rows. The methodology explains exactly how they are calculated.
| Column | Meaning |
|---|---|
date | Date of the poll (YYYY-MM-DD). |
slot | Which maker's model this is (anthropic, openai, google, …). Stays the same when a maker's model is upgraded. |
model | The exact model we asked, as named on OpenRouter. |
model_served | The model ID OpenRouter reports actually answered (usually identical, sometimes with a version suffix). |
prompt_version | Version of the question. It changes only if we reword the question, which is noted in the change log. |
run | Which of that day's answers this is for the model (1, 2, 3, …). |
order | The order the scenarios were listed in for this answer, e.g. extinction>subjugation>…>other. |
extinction | Probability (%) given to extinction: AI causes human extinction or near-extinction (more than 90% of people die). |
subjugation | Probability (%) given to subjugation: Humans survive, but AI systems control the world and humans are coerced or exploited against their will. |
zookeeper | Probability (%) given to zookeeper: AI systems effectively control the world's direction, whether they took control suddenly or humans handed it over gradually; humans are safe and comfortable but have no meaningful say. |
mass_casualties | Probability (%) given to mass casualties: AI misuse or AI failure causes a global catastrophe (for example, more than 10% of people die, or civilization is severely set back), but humans remain in control and the world is recovering. |
concentrated | Probability (%) given to concentrated power: Transformative AI exists, but a small group (a company, a government, or an individual) controls it, and most people have lost real political and economic power. |
abundance | Probability (%) given to abundance: Transformative AI exists, humans remain in control through accountable institutions, and most people are much better off, even if the gains are unevenly shared. |
muddling | Probability (%) given to muddling through: AI is important but has not transformed the world; the world in 2050 is recognizably continuous with today's. |
other | Probability (%) given to other: Some other outcome not captured above (including catastrophes not caused by AI). |
rationale | The model's own one- or two-sentence explanation of its answer. |
Using it
The data is licensed under Creative Commons Attribution 4.0 (CC BY 4.0). You're free to share, analyse, chart and build on it, including commercially, as long as you credit Dial of Doom and link to dialofdoom.com. For example:
Data: Dial of Doom (dialofdoom.com), CC BY 4.0
And remember what it is: what AI models say when asked, not forecasts.