Why AI Risk Could Arrive Fast | Roman Yampolskiy

time we talk. So, let's go. First, when will AI achieve AGI, artificial general intelligence, what year would you predict? 2028. Okay, and what year would general consensus be that it it has occurred? 2028. Okay, same year. Uh we're in 2026, by the way, because this uh conversation may list, so roughly 2 years from where we are now. Second, when will AI achieve ASI, artificial superintelligence? End of the 2028. Okay, so the difference between artificial general and artificial uh superintelligence will be measured in months because of the uh exponential recursive nature of it. Okay. Three, in 10 years from now, from 2026

in 10 years, what percentage of current jobs will be made redundant by AI? 10 years. Physical and cognitive? Uh uh jobs, people whatever their jobs are. One of them — 50% 50%. In 50 years, what percentage of current jobs will be made redundant by AI? Nearly all of them, except the ones where we prefer humans to do the job. Such as? Oldest profession. — Okay, well, the theoretically that's that is people are working on simulating that, too. Um so, okay, next, compared with the number of jobs that AI will make redundant, how many new jobs will AI create? On what frame? Eventually zero. Eventually zero, okay. Next, what might be some pleasant surprises that AI might generate? Immortality. Uh okay. Um seven, the seventh question. Will AI generate some very unpleasant surprises for humanity? And now this is a zero to 100 question. Some of these will be zero to 100. Zero meaning impossible, none. 100% 100 meaning it's absolutely sure. So, will AI generate very unpleasant surprises for humanity? Yes, and maybe 80%. Okay. What might What might certain unpleasant surprises be? What are the kinds of things? Novel suffering risks. Hm.

Hm. Will we ever detect AI trying to harm humans with malicious intent? Zero to 100. Already been done many times. Will AI cause casualties, say in the hundreds? Zero to 100. Think it's been done already. Yeah, we see suicides of teen Military targeting of schools. Hm. Uh yeah, and would that be um would that be AI or just faulty information put in? Targets were selected by AI. Will AI cause mass casualties in the millions? Zero to 100. In the millions. Eventually. Uh eventually on what time scale? Decades. Next, will AI ever try to exterminate all humanity? 0 to 100. 99. Will AI succeed in exterminating at least most of humanity? 0 to 100. 99-ish.

Okay, next. Will AI ever be given unsupervised authority, unsupervised authority to decide legal matters in judicial matters? 100%. Unsupervised authority. Right. Okay, will AI ever be given unsupervised authority to decide policy matters in government? 0 to 100. 98, 99. Next, autonomous weapons already exist, obviously. Will AI ever be given unsupervised authority to decide military matters on a massive scale, particularly the use of nuclear weapons? 99. That's very serious one, obviously. Um next. Will adversarial countries agree to limit AI in warfare like we did with poison gas and attempting to do with nuclear weapons? It seems unlikely.

Uh can AI ever be conscious as we are conscious? 0 to 100. 0 again means impossible and 100% is sure. 99. For galaxy traveling probes powered by AI, would there be a difference in value whether or not the AI were conscious? 0 to 100. No difference. So, you would put then no significance to having internal states in being able to colonize the universe. Either If we can't tell the difference, what's the difference? Well, well, that's there's there's no difference from a human point of view, but from a kind of a universal point of view, it's a different question. It's not can we tell the difference? Question is the fact that there is or is not a difference is a bimodal question. I mean, it's either one or the other. If we can never tell the difference, it may not make a matter to us, but is would there be a difference in terms of the value of the whole universe if it were conscious or not? Well, I perceive universe as the external observer of that probe. So, to me, I don't think universe would care. Okay. Um next, uh is virtual immortality possible in principle where it's uploading our first-person consciousness. Not a clone, but actually you. No time limit on technology. Not a clone, actually you. Virtual immortality, zero to 100. 95 Okay. Uh are we living in a simulation, a global simulation? Zero right now. Is our level — 99. 99 Uh can you ever uh approve that by showing some fuzziness on some test uh there that the limit of resolution of the simulation? We need external information. If they're willing to tell us, then we'll know. But you're 99. 99 on that. And the reason for that is the um uh it is the fact that once you can do a simulation, you can do it trillions of times as easily and so the statistical odds are there. Okay? Uh the ultimate human future is transhumanism, melding with AI. Zero to 100. Five. And finally, the ultimate future of the universe and the Earth and humanity is totally non-human. It's all AI, no humans at all. Zero to 100. 59. So with that very optimistic close, Roman, I've thoroughly enjoyed our conversation. You and I will continue to monitor this. As I've said before, any new signal or sign that comes, let us know. We can do an emergency conversation, get it published immediately because this is a critical area. And viewers can continue to watch other views. We have 1,500 videos and 100 TV episodes on mind and consciousness, all facets. Nature of consciousness, free will, personal identity, life after death, transhumanism, and of course AI. Thanks for watching Closer to Truth. Roman, it's great to talk and we'll get together again soon. Thank you so much. Look forward to an emergency podcast. Thank you for watching. If you like this video, please like and comment below. You can support Closer to Truth by subscribing. Closer to Truth is now accepting your tax-exempt donations. Please come to closer to truth. com/donate. Thank you very much for supporting us and thanks for watching.