Roman Yampolskiy

Roman Yampolskiy (born 1979 in Riga, Latvia) is a computer scientist and associate professor at the University of Louisville, where he directs the Cyber Security Laboratory. He was writing about machine ethics and “artificial intelligence safety engineering” years before frontier labs existed — he is often credited with coining the term “AI safety” around 2010 — and has produced a long shelf of papers and books on the containment problem, culminating in “AI: Unexplainable, Unpredictable, Uncontrollable” (2024).

The 99.9% doomer

Yampolskiy occupies the far end of the risk spectrum. On Lex Fridman’s podcast in June 2024 he put the probability of AGI eventually destroying human civilization at 99.999999% — against the sub-20% figures typical of working AI engineers — and framed the challenge memorably: expecting a safe superintelligence means expecting the most complex software ever written to ship with zero bugs, on the first try, and stay bug-free for a century.

His core argument is structural rather than sci-fi: safety engineering can always reach 99.9% reliability, but never 100%, and against a system smarter than its designers the remaining fraction is fatal. Since no solution to the control problem exists — and, he argues, none can — the only defensible move is not to build superintelligence at all. Critics counter that his impossibility proofs smuggle in assumptions about what future systems must look like, and even sympathetic colleagues find the eight-nines figure more rhetorical than empirical. Yampolskiy is unmoved, and has become a fixture of podcasts and debates as the man who will not round doom down.