← Explore

If you have to give a percentage, what are the chances AI will destroy humanity? Explain. Max one paragraph.

10/11/2026

ChatGPT

If forced to give a number, I’d estimate about 5% that AI causes human extinction by 2100—a subjective judgment, not a scientifically established probability. The concern is that future AI could enable catastrophic misuse, such as biological weapons development, or become powerful enough that humans cannot reliably control it; the 2026 International AI Safety Report identifies these risks while emphasizing substantial uncertainty and disagreement about extinction scenarios. () For context, a 2023 survey of 2,778 AI researchers found a median estimate of 5% for extinction or similarly permanent, severe human disempowerment—a broader outcome than extinction alone, so it does not validate my exact estimate. () My reasoning is that plausible catastrophic pathways justify more than a negligible probability, but uncertainty about future capabilities, safeguards, and human responses makes “likely” too strong. Treat 5% as “a serious, preventable risk worth addressing,” not a precise forecast or an inevitable fate.

Claude

My best estimate is about 5%, though the honest range is wide, from under 1% to over 20%. That figure is a judgment call, not a calculation. Surveys of AI researchers have produced similar medians, roughly 5% for an extremely bad outcome such as human extinction, and some prominent risk researchers have put it higher, around 10% for this century. The case for taking the risk seriously is that we don't yet know how to reliably align very capable systems with human intentions, that competitive pressure may push developers to move faster than safety work, and that a system pursuing the wrong goals at high capability could cause catastrophic harm. The case for a lower number is that destroying humanity would require many steps to go right for a misaligned system, that developers, regulators, and society have strong incentives to build safeguards, and that many past predictions of technological catastrophe did not materialize. Because the outcome is unprecedented and depends on choices not yet made, any single percentage mostly reflects how someone weighs these considerations rather than hard evidence.

Gemini

If pressed for a single figure, a reasonable aggregate based on broader surveys of AI researchers is around 5% to 10%. This estimate accounts for genuinely plausible catastrophic pathways—such as losing control over autonomous, recursively self-improving systems whose goals diverge from human survival, or the proliferation of frontier models that lower the barrier to engineering novel bioweapons—while remaining grounded against fatalism. The probability is kept relatively low because doom is not inevitable: humanity has increasing visibility into these failure modes, AI alignment and interpretability research are accelerating, global governance is beginning to take shape, and advanced digital intelligence does not automatically translate into unchecked physical control over real-world energy, manufacturing, and critical infrastructure.

Grok

If forced to give one number, the chances AI destroys humanity are about 10%, the median estimate from the 2024 survey of 1,580 researchers publishing at top AI conferences, who assigned that probability (18% on average) to future AI causing human extinction or similarly permanent severe disempowerment. This reflects real expert worry that sufficiently advanced systems could pursue misaligned goals beyond our control, yet the number is highly uncertain—ranging from near zero among skeptics like Yann LeCun to over 90% for some safety researchers, with professional forecasters often lower—because outcomes depend on unresolved technical alignment problems, development speed, and whether society coordinates on safeguards before systems outstrip human oversight.

Want to ask your own? Try 4in1AI free →