AI researcher Russ Salakhutdinov posed a question about โP(doom),โ which he defined as the probability that AI completely wipes out humanity. His own answer was 1e-5, or 0.001%, followed by: โHey, I am an optimist.โ
Alex Polozov replied with 3e-4, or 0.03%, calling it โobviously the best P(doom), hands down.โ Salakhutdinov answered: โObviously :)โ The reply kept the exchange firmly in conversational territory.
Neither post offered a model, time horizon or evidence for the numbers. They are personal estimates, not a study or a consensus forecast. The mathematical gap is still striking: Polozovโs figure is 30 times Salakhutdinovโs, even though both are well below 1%.
Later, researcher Seth Karten proposed a different question: โForget P(doom). What is your P(success)?โ He defined that as the chance that AI and humanity succeed beyond peopleโs โwildest dreams,โ but did not attach a number to it.