P(Doom) Test

P(doom), the probability of an AI-driven catastrophe severe enough to end humanity's story, started as informal notation on rationalist forums around 2010 and escaped the lab after ChatGPT, becoming a Silicon Valley icebreaker and a fixture of AI discourse. The striking thing about the question is how wildly the answers vary among the people who study the problem full-time. Consider the estimates put forward by some of the discourse's best-known voices:

• Roman Yampolskiy, AI safety researcher: 99.999999%.

• Eliezer Yudkowsky, MIRI founder: above 95%.

• Daniel Kokotajlo, AI 2027 author: 70-80%.

• Dario Amodei, Anthropic CEO: 10-25%.

• Geoffrey Hinton, Nobel laureate and godfather of AI: 10-20%.

• Yann LeCun, former Meta chief AI scientist: less than 0.01%.

The disagreement is not about missing facts; it is about different underlying worldviews. Those worldviews have names: seven main camps span the spectrum. Hard doomers hold that a superintelligence with goals even slightly different from ours would be uncontrollable, and that we are on track to build one before we know how to aim it. The AI-safety mainstream, including its longtermist wing, agrees the risk is real but bets that research, evaluation, and governance can bend the curve. Gradual-disempowerment theorists fear no robot coup, just humans ceding economic and political control one convenient delegation at a time. The present-harms camp argues that extinction talk distracts from the damage AI does right now, bias, surveillance, and concentration of power. Capability skeptics think machine superintelligence is far further away than the forecasters claim. Techno-optimists expect enormous benefit and human adaptation. And accelerationists go one further: to them, slowing down is the real danger.

This test locates you among seven archetypes drawn from those camps. The Doomer expects catastrophe and wants it stopped. The Careful Builder takes the risk seriously and builds anyway. The Pauser demands a halt until safety is proven. The Race Hawk fears losing the race more than running it. The Skeptic thinks superintelligence is much further away than the forecasters claim. The Watchdog cares about the harms happening now, not hypothetical ones. And the Fatalist expects a bad ending no one will prevent. Your answers score you on all seven; the strongest becomes your archetype.

The population comparison markers shown with your results are estimates informed by published surveys of expert and public opinion on AI risk, not validated population norms or percentiles; they are provided for general context only.

Your archetype is a worldview portrait, not a diagnosis or a calibrated forecast: a precise-sounding number can dress up a feeling as a measurement, two people saying “20%” may mean entirely different things, and most real thinkers are blends who move between camps as the evidence changes. This test is an original assessment inspired by the public discourse and survey research on AI risk; it is not affiliated with or endorsed by any of the researchers named above, and it is provided for educational and entertainment purposes only. A high score is not paranoia and a low score is not complacency.

References

• Grace, K., Stewart, H., Sandkühler, J. F., Thomas, S., Weinstein-Raun, B., Brauner, J., & Korzekwa, R. C. (2024). Thousands of AI Authors on the Future of AI. arXiv.
• Ord, T. (2020). The Precipice: Existential Risk and the Future of Humanity. Bloomsbury Publishing.
• Kokotajlo, D., Alexander, S., Larsen, T., Lifland, E., & Dean, R. (2025). AI 2027. ai-2027.com.