Transcription
I'm here to tell you this is it. The world as you know it is over. Completely done. Okay. Uh, it's not about to be over. It's over. Right.
I'm a very serious geek. I started coding at age 8, which feels like a zillion years ago. And when I coded, believe it or not, at age eight, uh, first time I coded, I was using BASIC, which is, as the name implies, a very basic language. And I wanted to create an AI. That's that was my dream, right? Every one of us geeks, we just wanted to create an AI. It was everyone's dream. We never managed to, but we did something that appeared intelligent.
So, you know, let me let me start by defining what is AI. Every piece of code I've ever written before the year 2000 was a piece of code where I solved the problem first with my intelligence and then told the computer how to solve it. Right? So it's almost like if I call one of you and give you a puzzle and then tell you, "This piece you put in the top left corner, and then that piece under it, and that piece under it, and then turn this one," and so on. You can finish the puzzle. And if someone doesn't hear me giving you the instructions, they would think you're very smart, but you're not, right?
By the turn of the century, this ended. So the idea of us telling computers what to do. By the way, they did it very quickly, very accurately, at a massive scale, and so they appeared smart, okay? But it was me who solved the problem. It was my intelligence multiplied. By the turn of the century, we discovered something called deep learning. And deep learning was truly teaching computers the way I taught my children how to be intelligent.
The way I taught my children how to be intelligent was to give them little puzzle pieces when they were young, you know, those cylinders and a board with different holes in it. And no one ever told their children, "No, no, hold on, my son. Take the cylinder, turn it upside down, look at the cross-section, it will look like a circle, compare it to those holes. It will match the one that is a circle, put it there." Nobody ever tells their children that. What we do is we give them a cylinder and a board, and they keep trying, trying, trying until one time it goes through, and then suddenly, light bulb, and they learn something, right? That's exactly what we did with computers with deep learning.
By the way, we started to be able to do deep learning because by the year 2020, but sorry, by the year 2000, the internet was big enough for us to have those number of trials for the computers to learn on their own. Now, please understand, when we did that, I think my favorite example of that was the year 2009. We did it earlier, but at the time I was at Google, and we published a paper called the cat paper. And the cat paper was, uh, basically, we asked the computers to watch YouTube and tell us what they find. We had a lot of compute capacity. So they would take YouTube videos, cut them into, uh, 10 frames per second, and compare the patterns on the 10 frames. Basically trying. And then one of them said, "We, I found something." We had to write a bit of code to find out what it found. And it had found a cat. Of course, it's YouTube, right? And it didn't only find one cat. It found what makes something a cat. It found it could literally find every cat on YouTube. We never taught it how. We never understood how it did it. Understand that, right?
And the code that was written for a computer to find every cat on YouTube, if I had written it the traditional way, would have been probably 200 million lines of code because I would have to find every possible interpretation of how a cat looks on YouTube and get the computer to see that, right? That code was this big, right? But there was a lot of numbers and mathematics that we don't understand. Just like I can ask you a question, and you'd give me a very intelligent answer, and I have no clue what happened inside your brain to get to that answer. This is where we are. Computers find their own intelligence. We don't teach them how. We just give them the data to learn, and they learn like humans.
And every task we've ever given them, since they've become better than humans at. Right? So they are the world champion in chess. They are the world champion in Go. They are the world champion in everything we've given them. They are the best manipulator of humanity on the planet in terms of social media engines. They are the best writers, the best artists, the best musicians, the best anything we've ever given them. Right?
And between the turn of the century and the time when we recognize that, which is what I normally refer to as the Netscape moment, right? So 2023, every one of you started to hear about AI. Uh, that's not because AI started in 2023, right? AI started in 2000, and by 2016, all of us in the lab, we knew we got it, right? By 2016, we had created code that would blow you away, right? Uh, we started to fold proteins. If you understand, protein folding is one of the most complex problems ever faced by biologists in history. You know, it would take a group of PhD students around 8 months to fold one protein. We created AlphaFold in 2016, 2017, and it folded 200 million proteins, okay, in a day.
Now, when ChatGPT came out in 2023, people said, "Oh, there's something called AI." It's not because AI started. It's because we had a browser, just like Netscape. When, you know, the internet came out in 1995, the internet had existed for almost 15 years before. It's just that for the first time, we had a browser, we could see it. Now, since 2023, uh, until today, it's been mind-blowing. Okay?
Just to give you a few statistical pointers to understand. First of all, intelligence is a lot deeper than ChatGPT. Let's be very clear. Okay? The task given to ChatGPT and Gemini and others is linguistic intelligence. It's the ability to understand knowledge and communicate and so on and so forth. Okay? Of course, there are other forms of intelligence, emotional intelligence, for example, they haven't learned yet, right? Uh, intuition, uh, um, you know, complex mathematics, um, deep reasoning, all of those forms of intelligence. We haven't seen AI perform that way yet, but we will very soon.
Uh, when the rules of the game change to the point where your customer is not necessarily making their choices themselves anymore, but that the machine will recommend to them, the rules of the game change. When you know, I've seen beautiful marketing advertising campaigns here, when there will be a moment in the near future when machines will be marketing to machines, and we will be out of the, uh, of the picture altogether, the rules of the game change, right?
And I want to say that as the rules of the game change, we can end up in a magnificent utopia of abundance, right? And we can also end up in a dystopia of a very difficult time. Let me talk about the, the, the utopia of abundance first, so that you understand. I'm not here to scare you. I'm here to tell you that the water is boiling, right? My objective is for you to jump out of the pan, right? Uh, uh, but, but I, but today, to do that, I have to tell you it's boiling, and that hurts.