Mo Gawdat, former CBO of Google X, argues AI development cannot be stopped because of economic incentives and a prisoner's dilemma between nations. He traces AI's evolution from early programming through deep learning to transformers and GPTs, which build their own internal representations beyond what humans teach them. He sees three inevitables: AI will happen, it will become smarter than most humans, and alarming outcomes may follow.
Chapters
And the one thing I want to chat to you about is AI and to me it's the opposite.
Exponential, crazy, speeding up. So the three inevitable. So I want to start there. The first was AI will happen. Well, I think that's a big tick. It's here. Two, it will become smarter than us, potentially it's already there or it's close to it.
And then smarter than most of us.
There we go. And then the third was the most alarming bad things will happen. So the first was, is a big tick. Second, let's just hover on the second that's going to become smarter than us. What do we mean by it's becoming smarter or is smarter than most of us?
So the first is a big tick. But it's also important to understand that there is no stopping it. So AI will happen. We know that AI has happened already when I, when I spoke in, in Scary Smart. So I attempted in Scary Smart, believe it or not, I wanted to write scary smart in 2018. And my very first video, the day I left Google was about that. And it's, you know, public. It's been seen 20 million times or so. And, and, and the idea is that people don't understand because people that you tell about AI and the possible threat of AI will tell you, okay, you know what? If it's threatening us, we'll, we'll switch it off. We'll just cut the electricity supply to it. No, that's not going to happen. Right. You know, if we, if we feel threatened by it, we're going to, to just get together and stop developing it. No, that's not going to happen. I think the reality of the matter is that between a prisoner's dilemma, that that was created by, by, by capitalism and our economic dependence on the Internet, it's almost impossible. It's not almost. It is impossible to stop AI So, so when I say in the first inevitable of Scary Smart that, that AI will happen, I, by that, I mean that's it. The genie is out of the bottle. There is no putting the genie back in the bottle. Right. The second inevitable is, is very important to understand. So let's, let's very quickly tell our listeners what is AI and what, what AI Is. You know, I coded computers since I was eight. And you know, we had, at the time I had a Sinclair and then a Commodore 64 and a Commodore 128. And if, if you don't know what I'm talking about. Lucky you, you young person.
I remember the Commodore.
Yeah. And, and, and those for For a, for a geek like myself, where heaven. But if you compare them to today's computers, they, they're insulting really. I mean, the way we programmed those was so level, so low level, so geeky that you know that, that you, you really couldn't get it if you were not a geek. Right. And, and I, you know, and I even then dreamt of building an AI, right. Every one of us wanted to build an AI, to, to transfer your, your smarts into a machine. Right. The, the challenge was technology with computers in general and programming that most people who have not programmed, don't understand is that until the turn of the century, computers were not intelligent in terms of, they were just using my intelligence to perform the task. So I solved the problem first and I outlined a way to solve it. And then I told the computer, when you get this number as A and that number as B, add them up and then divide them by, by this number as C and you will get a result. Tell that to the customer. And the customer presses, yes, do this. If the customer presses no, do that. I solved the flow chart and then I told the computer to do it. By the turn of the century with deep learning, mainly we, we, we, we deviated from that. We basically said to the computer, look, we want you to find the number eight in all of those pictures that you find in front of you. We're not going to tell you how to find it. We're not going to tell you it's made up of two circles or, you know, because it could sometimes be made of us, you know, a scribble that looks like an, you know, upside down infinity symbol and so on and so forth. We're just going to tell you guess, right? Like a little child, you learn what is eight and what is six? And the way to do it is you, you, we show you a picture and you tell us if it's a, if it's an 8 or, or whatever else. And if you're right, we will keep your code and improve it. If you're wrong, we will kill your code and use the better code. Okay, which sounds really vicious, but that's the way we did deep learning until, you know, 2018 or so, when Jeffrey Hinton and others were advocating reinforcement learning for a very long time. And, and we then started to tell the computer, no, no, no, hold on, it's not a six. This actually is an eight. What can you change about your algorithm so that you see it as an eight? Okay. Which is a very, very strong milestone that led us to the GPTs and the transformers and so on. Right? So, so, so with that, AI Is not actually a piece of code. It's not human intelligence. It's learning from human intelligence, but it's building its own intelligence.
