AI safety expert Roman Yampolskiy warns superintelligence is the last invention humanity will ever make — after that, it takes over everything, including the process of invention itself.
The Diary Of A CEO with Steven Bartlett
AI safety expert Roman Yampolskiy warns superintelligence is the last invention humanity will ever make — after that, it takes over everything, including the process of invention itself.
TL;DR
AI safety pioneer Dr. Roman Yampolskiy joins Steven Bartlett to discuss why all jobs — not just manual ones — will eventually be automated, why retraining is a dead end, and why the "just unplug it" argument is dangerously naive. Yampolskiy predicts humanoid robots competitive with humans by 2030 and a technological singularity by 2045, while arguing that superintelligence is the single most important existential challenge humanity faces. The key takeaway: unlike every prior technology, AI is not a tool but an agent — the last invention humanity will ever need to make.
Dr. Roman Yampolskiy, AI safety expert and Professor of Computer Science, discusses why all jobs will eventually be automated, why the singularity is by definition unpredictable, and why humanity-level extinction from AI is a real and proximate risk.
Chapter 2 · 01:03
Every Uber driver says no AI can replace them. Every professor too. They're all wrong.
Ask any Uber driver if they're worried about self-driving cars and they'll say no. Ask any professor if AI can replace their lectures and they'll say no. The same cognitive dissonance runs across every profession — and it's already wrong, as Waymo cars in LA prove.
Chapter 4 · 05:03
Told to learn to code. Then AI coded. Told to do prompt engineering. AI does that now too.
Every previous wave of automation displaced workers into new jobs. This time is fundamentally different. AI doesn't just replace one job — it replaces the very concept of a human worker, making every 'retrain into this instead' strategy obsolete.
Unlike past automation waves, AI will automate every job, making retraining into a new career a futile strategy.
Yampolskiy predicted that AI agent design roles, currently seen as the hot new tech job, will themselves be automated within a year or two.
Chapter 5 · 06:58
No government currently has programs prepared to handle unemployment rates approaching 99% as AI displaces the workforce.
No government currently has programs prepared to handle unemployment rates approaching 99% as AI displaces the workforce.
Chapter 6 · 08:54
If you could predict what superintelligence will do, you'd already be as smart as it is.
The singularity isn't just a buzzword — it's a mathematical certainty about the limits of prediction. If you could predict what a superintelligence would do, you'd be operating at the same level of intelligence. By definition, you can't.
Chapter 7 · 11:40
Silicon substrate is much more capable for intelligence. It's faster. It's more resilient, more energy efficient in many ways.
Chapter 8 · 14:11
Even plumbers aren't safe — humanoid robots competitive with humans in every trade by 2030.
By 2030, humanoid robots will likely have the dexterity and flexibility to compete with humans in every physical domain — including skilled trades that were supposed to be safe. The combination of physical ability and AI intelligence leaves very little for humans to do.
Yampolskiy predicted that by 2030, humanoid robots will have enough flexibility and dexterity to compete with humans across all physical domains, including skilled trades.
Chapter 9 · 17:15
You're reading more than ever. But as a share of total knowledge, you're getting dumber every day.
Ray Kurzweil predicts 2045 as the year technological progress becomes so rapid that humanity can no longer understand or keep up with it.
You might be reading more than ever. But AI is generating knowledge faster. As a percentage of all existing knowledge, every human is getting dumber daily. At extreme values, that percentage approaches zero.
As AI knowledge expands exponentially, every human researcher's share of total knowledge shrinks daily, eventually approaching zero.
Chapter 10 · 20:15
Fire couldn't invent fire. The wheel couldn't invent the wheel. AI can — and that changes everything.
Fire was transformative. The wheel was transformative. But they stopped there — they couldn't invent new things. AI is an inventor that builds new inventors. Once it arrives, humanity's role in the process of discovery ends.
Yampolskiy argued that creating artificial general intelligence is unique because it produces a new inventor, capable of making all future inventions itself.
Chapter 11 · 23:31
He predicts a humanity-level extinction event and still sleeps great. Here's why that makes sense.
Yampolskiy sleeps fine even though he believes humanity may be near an extinction-level event. This isn't denial — it's evolution. The ancestors who worried too much about death didn't survive long enough to reproduce.
Chapter 12 · 25:35
If superintelligence kills everyone in 5 years, climate change is no longer your problem.
Climate change could boil us alive in 100 years. But if superintelligence kills everyone in 5, climate change is no longer your problem. Get AI right and it solves everything. Get it wrong and nothing else matters.
Chapter 13 · 28:20
Go ahead and try to turn off Bitcoin. I'll wait. That's exactly how naive 'just unplug the AI' is.
People keep suggesting we can just unplug a dangerous AI. Yampolskiy's answer: try turning off Bitcoin. Try turning off a computer virus. Distributed systems can't be shut down — and a superintelligence will have already predicted your move and neutralised it.
Unlike nuclear weapons, superintelligence is a distributed autonomous agent that can make backups, predict human responses, and neutralise shutdown attempts.
Chapter 14 · 28:47
Yampolskiy argued that uncontrolled superintelligence poses equal danger regardless of which nation builds it first, making it analogous to mutually assured nuclear destruction.
Yampolskiy argued that uncontrolled superintelligence poses equal danger regardless of which nation builds it first, making it analogous to mutually assured nuclear destruction.
The cost to train sufficiently large AI models decreases annually, meaning superintelligence that might cost a trillion dollars today could eventually be built on a laptop.
Advances in synthetic biology mean that someone with only a bachelor's degree in biology may soon be able to create a novel virus.
Even the people who build large AI systems must run experiments on their own products to discover what they are capable of — the internal workings remain opaque.
No indexed bits in this chapter.
Fire was transformative. The wheel was transformative. But they stopped there — they couldn't invent new things. AI is an inventor that builds new inventors. Once it arrives, humanity's role in the process of discovery ends.
Every previous wave of automation displaced workers into new jobs. This time is fundamentally different. AI doesn't just replace one job — it replaces the very concept of a human worker, making every 'retrain into this instead' strategy obsolete.
People keep suggesting we can just unplug a dangerous AI. Yampolskiy's answer: try turning off Bitcoin. Try turning off a computer virus. Distributed systems can't be shut down — and a superintelligence will have already predicted your move and neutralised it.
This episode
Factual claims made this episode, and whether a source was named.
Roman Yampolskiy coined the term 'AI safety' in 2010 and has published over 100 papers on the dangers of AI.
Self-driving cars are already replacing human drivers, with Waymo operating driverless taxis in Los Angeles.
Driving is one of the most common occupations in the world.
AI became capable enough at coding that prompt engineering — once touted as a safe new career — is now itself being automated by AI within approximately two years.
No government currently has programs prepared to deal with unemployment rates approaching 99%.
Ray Kurzweil predicts 2045 as the year of the technological singularity.
As a percentage of total global knowledge, every individual human researcher gets relatively less knowledgeable every single day as AI generates knowledge faster than humans can absorb it.
Silicon substrate is faster, more resilient, and more energy efficient than biological neural substrate for the purpose of intelligence.
By 2030, humanoid robots will have sufficient flexibility and dexterity to compete with humans in all physical domains, including skilled trades like plumbing.
The cost to train sufficiently large AI models decreases every year, and what might cost a trillion dollars today could eventually be achievable on a single laptop.
Someone with a bachelor's degree in biology can probably already create a novel virus using advances in synthetic biology.
Even the engineers who build large AI models must run experiments on their own products to discover what capabilities those models possess, because the systems are not fully understood by their creators.
Booking.com has helped over 1.8 billion vacation rental guests find places to stay since 2010.
Nearly half of new hosts on Booking.com receive their first booking within a week of listing.
This episode
Futurist cited for his prediction that the technological singularity will arrive in 2045.
AI safety expert and Professor of Computer Science who coined the term 'AI safety' in 2010; the primary guest discussing existential AI risk throughout the episode.
Episode sponsor; described as one of the world's most downloaded travel apps that has helped over 1.8 billion guests find vacation rentals since 2010.
Cited as a potential path to enhancing human intelligence through brain-computer interfaces to help humans remain competitive with AI.
Mentioned as one of the leading companies developing humanoid robots at rapid speed.
Cited as a real-world example of fully driverless taxi services already operating in Los Angeles, demonstrating that driver automation is not hypothetical.
Used as an analogy for a distributed system that cannot be 'turned off' by any single actor, illustrating why a superintelligence also could not simply be unplugged.
Used as an example of a deployed AI system whose builders must run experiments to discover its full capabilities, illustrating the black-box nature of modern AI.
Science fiction franchise cited as an example of a story that avoids depicting superintelligence by simply banning AI within its fictional universe.
Cited as the primary geopolitical rival in the AI race, used to argue that the US cannot unilaterally stop building advanced AI without ceding strategic advantage.
Discussed in the context of leading global AI development and the geopolitical tension around whether the US should cede AI leadership to China.
Used as a real-world example where both Tesla's self-driving vehicles and Waymo's driverless taxis are already operating, demonstrating driver automation is present-day reality.
Stats
We use essential and analytics cookies to run Vuci. To understand how the site is used: Privacy Policy.
Install Vuci on your phone
Add it to your home screen for a faster, app-like experience.
Install Vuci on your phone
Tap the Share button, then “Add to Home Screen”.
A new version is available
Reload to get the latest Vuci.