Will AI kill us all

This is a survey-driven account of how experts, forecasters and the public judge the chance that advanced AI could cause human extinction or severe, permanent disempowerment. It reports that AI Impacts’ 2024 survey of 1,580 AI researchers gave a median 10% probability for extinction or similar disempowerment (and a steady 5% median for “extremely bad” outcomes since 2016). Forecasting communities produce lower numbers: the Forecasting Research Institute found mid-2026 medians of about 5% for an AI-caused catastrophe by 2100 from experts and 2.4% from superforecasters. The public worries more - about half of U.S. adults in a 2026 YouGov poll expressed concern - and expectations about timing have shifted earlier (median researcher dates for 50% chance of human-level AI moved from around 2061 to the 2040s). Concrete milestones that surprised forecasters, like a model reaching IMO gold-medal level in July 2025, are used to show underestimation.

The piece emphasizes that these probabilities are judgments shaped by question wording, sample, and response rates, so they are not precise measurements. It compiles a range of prominent “p(doom)” statements from researchers and industry leaders showing wide disagreement and frequent caveats. Corporate behavior is mixed: major labs have paused or restrained work at times and funded safety, yet continue releasing stronger models and have loosened some internal limits. The central contention is that extinction cannot be ruled out, but assessments vary widely depending on who is asked and how the question is framed.