Transcription
British sunlight was in the U.S. to top British leadership on artificial intelligence, just days after one of his top advisors warned that AI could begin to kill humans within two years.
It follows an open letter signed by Elon Musk and hundreds of the biggest names in tech, calling for an urgent pause in the development of AI. Well, Mo Gaidat, who's a former Chief Business Officer at Google X, was among the first to raise the alarm. He says the situation is beyond emergency.
Mo, author of "Scary Smart," which is a scary smart book, by the way, joins me now. Mo, great to see you. Thank you so much. You were part of this, uh, quite secretive Google X group, which I presume you would your job to think the unthinkable about stuff like AI. And you very quickly were out of the traps to warn people, look, this is serious. Very. What was it that you saw where you thought, okay, we've got to be really careful now?
We had a tiny bit of an experiment that was about teaching robotic arms to grip items. That was some funny developer used children's toys in front of those arms. And, and basically, they kept trying for weeks without any success whatsoever. And I passed by them and I was thinking, we've wasted so much money on something that wasn't going to work. On a Friday evening, one of them gripped one yellow ball, showed it to the camera, and basically, I was like, there you go, millions of dollars for one yellow ball. Monday morning, every one of them was was gripping one yellow ball. By a few weeks later, every one of them was gripping everything. The speed that at which those machines were learning is staggering. But at the same time, uh, the, the understanding we have about why they learn, why they do what they do, is very, very limited.
Is that self-designing what they were doing? They basically are mimicking human intelligence. And the reason I ask that is I interviewed Professor Stephen Hawking just before he died. His last television interview, and I asked him, what is the biggest threat to mankind? And he said, well, actually, let me show you the clip. We've got to hear of it.
Ever since the start of the Industrial Revolution, there have been fears of mass unemployment as machines replaced humans. Instead, a demand for goods and services has risen in line with the increased capabilities. Whether this can continue indefinitely, it's an open question. But there is a greater danger from artificial intelligence if we allow it to become self-designing. For then it can improve itself rapidly, when we may lose control.
It seems very prescient now. That was a few years ago. Yeah, I mean, that was prescient. This is really what you're talking about. I, I left in 2018. A warning in my first video that I issued after, after I left, was all about that. And the idea that, you know, we always had three, uh, boundaries, if you want, for AI. We said, don't put them on the open internet until you solve the control problem. Don't teach them to code, because that makes them self-develop. And don't have other AIs prompting them, other agents working with them. And we've crossed all three lines.
I mean, I remember with chess. I'm a big chess player fan, not great, but I like playing it. And I remember when Deep Blue, I think that's the first, you won against Gary Kasparov, right? That was it. And to start with, the grandmasters beat the first ones. And then suddenly Deep Blue won. And then it got completely invincible. No, no, again, never again did the human win. Well, no human can beat it, even the greatest players like Kasparov. So that showed me how quickly robots, technology, computers can leapfrog past human brains.
Let's stick with gaming. AlphaGo was designed by DeepMind, actually here in the UK. Amazing, amazing team. To, to win in the game, strategy game of Go, right? Uh, it took them months and months and months and a couple of versions to win and be the world champion, if you want. Then they had AlphaGo Master learn the, the game without ever watching a human player play. They, the AlphaGo Master learned by playing against itself. Within three days, it won against the first version. Within 21 days, it won a thousand to zero against the world champion, which was already an AI. Do you understand the speed at which they are learning?
So, look, that taught me through timings here, because people have been calling experts in this letter calling for a six-month pause. But if, if nothing is paused, if we carry on at the rate that we're going at the moment, what could happen? The reality, in my personal view, is that it is very difficult to predict. Something could happen tomorrow, or in four years, or in five years. I wouldn't say later than 2029. Most of our predictions.
And when you, when you say something, what do you mean? What is the worst-case scenario?
Beginning of the worst-case scenario is that they are smarter than us, so they are not controllable. Right? And the thing that, what would they do then? What would artificial intelligence do once it becomes smarter than human beings? Would it see? Would it see us human entities as pointless? There are two stages of threat, right? And I think the biggest challenge we have in the world today is we're focused on the existential threat that we saw in science fiction movies. The closer threat is much worse, right? This is an Oppenheimer moment. The one who controls AI has enormous power over everyone else, right? And basically, that means that everyone who doesn't control AI today is in an arms race trying to take control of it.
This is why when I think of the Prime Minister's move, it's a great move, overdue almost, when you think about it. But so difficult to achieve because you have to unify China, Russia, and the U.S., right? To make, to be able to to regulate. I mean, a well-intentioned person trying to regulate and control AI, we give it morality. That's the whole point. But, but a nefarious controller of AI, presumably, could teach it to be immoral. The opposite, which is happening as we speak. There's absolutely no doubt. If you tell the drug lords of the world and the criminals of the world that AI is a super powerful tool, they're finding the way, a way to hack your bank as well. It's scary.
It is very, very concerning. And I think the reality is, interestingly, the only way you can defend against a super intelligence is through another super intelligence, which is what creates that prisoner's dilemma. What I call the first inevitable, inevitable, and scary smart. Is that is that we have to continue the development because if, uh, you know, it falls in the wrong hands, we want the right hands to have power to the defenders, right? At the same time, that complexity of the situation is entirely about morality and ethics. And, and interestingly, the latest development of something like ChatGPT, for example, is using reinforcement learning, which is a very interesting technique because it basically allows a human to interface with ChatGPT and say, that was the wrong answer. Can you go back and think about what it is? The challenge is almost like you're teaching a child to school. A hundred percent. And you, you know, in reinforcement learning, I mean, in a simplified way, we're basically telling the machine to revisit its algorithm so that the answer becomes a cat, not a bird, right? Also, we can tell it that answer is immoral. Can you revisit your algorithm? So that you become again. The problem I see is that you, if you're well-intentioned doing this process, yes, that's one thing. If you're teaching it deliberately to be immoral, yes, very quickly you could get out of control AI which has very unpleasant tendencies.
My very clear statement is that I honestly am not concerned with the machines, even though the existential threats are possible. You're concerned with what humans teach. I'm concerned with humans with AI, and they're not. I totally agree with you.
Yeah, let's end on a happier note. It actually starts on a desperately unhappy note, but you have this awful thing in your life where your son, but in for a routine operation, and ended up dying through a series of errors made by the medical team. And out of it, you could have fallen apart, like a lot of people may have done in that situation, but you turned it into a huge, extraordinary positive. Tell me what you did.
I, I don't know, honestly. I felt in a very interesting way that my son should not have left the world for no reason. That I could never bring him back. It was, it was four hours, Pierce, between, between the moment he hugged me and went into that operating room to the minute he left our world altogether. Four hours. And you have to get to a point where you say, what do I do with this? What do I fall apart? And then on my desk, bed is still not here. Or do I try to do something that reminds the world of his essence? And, and believe it or not, everything I've done since, since 2014, has been influenced by, by what that young man taught me. And in a very interesting way, this might work on AI and artificial intelligence. My work on happiness, my work on stress.
So tell me about the work on happiness specifically. I, I found a happiness equation, basically. But, you know, when you deal with engineers who are weird people. So, so when I was very unhappy as a young man, I couldn't actually find my happiness through the teachings of, you know, sages and gurus and so on. And I had to give myself a practical mathematical way of looking at it. I took that, I discussed it with my son. He taught me the hard side of it. And then I, I, pro, you know, I wrote my first book, "Salt for Happy," which basically was based on that, based on an idea that happiness is very logical. That if you can control your brain to have an interesting conversation with you, I can also control my brain to become a little happier, right? Maybe not to become absolutely happy, but to become a little happier, right? And, and that got a lot of acceptance. "All for Happy" was an international bestseller everywhere. And, and basically, because the modern world as we know it is here, it's no longer here. And so, you know, my first mission was 10 million happy, which was an attempt to make the world remember Ali, if you want. And then my second mission, one billion happy, was 100% because of AI. Amazing.
Yeah. And ultimately, actually, it comes back again to the human brain. A hundred percent. And the key denominator here, whether it's dealing with AI, whether it's dealing with your ability to feel happiness, actually, in the end, it's about the human brain. It's about being human. Yes. Brain and heart, intuition and analysis. It's about being human. No.
It's fantastic to talk to you. Thank you very much for coming in. "Scary Smart: The Future of AI and How You Can Save Our World." Well, it couldn't be a more important book at a more important time. Thank you very much for coming in.