PERSON

Roman Yampolskiy

The computer security researcher who coined “AI safety” in 2011—then spent more than a decade arguing, with formal proof rather than intuition, that a sufficiently advanced AI may be unexplainable, unpredictable, and uncontrollable in principle, not merely in practice.

Roman Yampolskiy is the thinker who arrives at the AI safety debate carrying the wrong toolkit—and turns out to be the only one with the right one. Trained as a computer security researcher, a professional whose discipline begins with the assumption that every system can be broken and that the burden of proof lies with whoever claims it is safe, he applied that mindset to the question of superintelligence and arrived somewhere more radical than the field was prepared to go. His argument is not that advanced AI will be dangerous in the way that nuclear power or aviation are dangerous—hazards manageable through engineering and regulation. His argument is that a sufficiently advanced artificial intelligence may be uncontrollable in a deep and possibly permanent sense, a claim he reaches not through alarm but through the same impossibility results from theoretical computer science that established the limits of computation itself. He distinguished three failures that interlock: the

Keep reading with YOU ON AI