Loading summary
Transcript36 lines
- [00:00]
Matt Clifford
Thank you so much for listening to the Rest Is Politics. Here's a thought for Christmas, you can gift somebody membership to the Rest Is Politics. Plus ad free listening, bonus episodes, early access to Q and A book discounts. So spread a little political peace and goodwill, head to therestagepolitics.com and click Gifts.
- [00:20]
Rory Stewart
This is the single most important opportunity challenge in our lifetimes. I'm afraid it's bigger, more transformative than climate change, the Trump administration, what Russia.
- [00:32]
Matt Clifford
Doing in Ukraine, you might be underselling it. Far from being in a bubble, AI is underhyped.
- [00:38]
Rory Stewart
These companies aren't trying to build toys, they're trying to build something that can do everything that a human can do.
- [00:43]
Matt Clifford
Whether it's in coding, whether it's in radiology, whether it's legal work. We're already seeing AI models being able to outperform humans in certain domains, and.
- [00:51]
Rory Stewart
Yet the basic debate is, is it going to be able to write my essay for school for me? Most of these companies are relatively new and the people at the top of them are unbelievably personally wealthy.
- [01:03]
Matt Clifford
People made AI videos and it's pretty coherent.
- [01:06]
Rory Stewart
It looks good. The problem, these large language models exist only at the moment in China and the US AI, which is taking over the world.
- [01:15]
Matt Clifford
My worry is that we say let's be cautious, let's not adopt, let's try and preserve things as they are. But our competitors internationally don't do that. We need a solution to stagnation, and it's very possible that AI is our way out of the mess, or our.
- [01:30]
Rory Stewart
Route into mass unemployment, increased inequality, and massive threats to the future of humanity.
- [01:36]
Matt Clifford
I suppose the thing that is most urgent is that no one knows, and anyone who tells you otherwise is lying.
- [01:44]
Rory Stewart
Hi, it's Rory here and I've just recorded a very exciting series on what I believe is the single most important issue facing us. And that isn't actually China, Russia, Ukraine, maybe even not climate. It is artificial intelligence. I've been sitting down with Matt Clifford, who was the government's AI advisor, who put together the AI Safety Summit, who's an entrepreneur, and we've been really getting into the details of this, what it means for geopolitics, what it could mean for our economy, what it could mean for jobs, what it could mean actually to the existential future of humanity. Not to be too pretentious about it, and the way in which politicians are simply not talking about it, the public isn't engaged and we haven't really started this discussion properly. So here's a taster for you and if you'd like to hear more, I'm really proud of this series. Please do sign up and become a member. Go to therestispolitics.com Sign up to trip and come with us on the series through artificial intelligence. This is something which can change all our lives for good or else. And yet it's something about which we know so little. I've set it up in very grand terms. Do you agree with me that it's a bigger issue than climate change in Russia?
- [03:05]
Matt Clifford
I actually think if anything, where we are and where we're going might be underhyped. So like three facts I often use with politicians to try and bring this home why this might be real. One it's already here in the GDP numbers. So if you look at US economic growth in the first half of this year, without AI, the US economy would have grown by about 0.1%. With AI, that's about 1.1%. So in other words, like AI is already accounting. This is mainly Capex build out the build out these huge data centers. But AI already accounting for 90% of US economic growth in 2025. So like we cannot ignore it if even if all you care about is today. Second thing, you know, you mentioned the business that I built, entrepreneurs first. I've been investing in AI for, for about 12 years. Over that period we have seen an increase in the investment in AI, the inputs of about a billion fold. I don't think that's ever been true in the history of technology before. And by the way, at the time people thought, was there a bubble? I remember people saying, oh, is there a bubble? We now invest a billion fold more each year as a species in this technology. So that's the second big thing. The third big thing is we're seeing this extraordinary competition for talent and for the other raw inputs of AI, energy, chips, et cetera. The stat that I like to use is Mark Zuckerberg is so worried about the impact that AI might have on his business Meta, that he is now paying top AI researchers $250 million a year to come and work at Meta rather than somewhere else. Now, maybe he's crazy, but something is happening and it's something that we shouldn't ignore.
- [04:36]
Rory Stewart
Single human individual. Single human individual make $250 million a year because of their knack of being able to work, their ability to squeeze.
- [04:45]
Matt Clifford
Intelligence, I guess you would say, out of this infrastructure.
- [04:49]
Rory Stewart
Okay, well let's try to think about what the big issues are around AI. And I think the first one is the question of power. So if these machines work remotely in the way that the people funding them and dreaming about them hope, they are going to be very, very capable indeed. They're going to be able to do many things at least as well as humans, and some things better than humans. And I think that's the beginning of the whole conversation, because that is huge potential to do good, for example, creating medicines, huge potential to do bad, for example, creating deadly viruses, fantastic capacity for improving defence and security, but also, on the other hand, a threat that could be posed to us.
- [05:37]
Matt Clifford
Absolutely.
- [05:38]
Rory Stewart
So that's one thing I think we need to get into. What are the capabilities of these machines? What can they actually do? Now, the second thing, of course, that goes along with power. Who has that power?
- [05:48]
Matt Clifford
Right.
- [05:48]
Rory Stewart
And we can think about that in different ways. Is the power concentrated in a very small handful of companies around the world, companies which already account for something like 30% of the entire US stock exchange, which is 70% of the global stock exchange. So let's say 20% of the entire global equity markets are bound up in these companies. Or can you think about it in terms of countries? Because these large language models exist only at the moment in China and the us. So is there a sovereignty problem? Does that mean the US and China now has its hands effectively on our throats if they wish to assert that influence? What happens if these things somehow get out of control? What would happen if these things are not safe, that we can't predict, explain or control them? And there the problem is not that we're being controlled by Elon Musk or Mark Zuckerberg, or we're being controlled by Xi Jinping, but potentially these things are no longer under anyone's control. So huge issues to get into. So I'm going to pose as a new, I don't know, minister appointed in the Department of AI. How would you begin by trying to explain to me what AI is and isn't.
- [06:56]
Matt Clifford
I think I'd start by saying something like, the reason humans are special is our ability to reason, you know, our ability to solve problems in the world in order to achieve our goals. And for a very long time, that has been the only way to do thinking, to do reasoning, has been in the human brain. And we have over the last, well, depending on how you measure it, you know, say, let's say 100 years, figured out lots of ways to get machines to help us do that. But until now, that's very much been about tools. You know, a calculator is way better than you or me at arithmetic. But no one would say a calculator is thinking. But the explicit goal of the companies building AI artificial intelligence is to build what they call AGI, artificial general intelligence. In other words, not tools, but brains, if you like machines that are capable of doing everything that a human could do. Now, they're not there yet, but that's the goal.
- [07:47]
Rory Stewart
So the goal is that this thing would be much more flexible, much more general in the kind of tasks it could take on. Unlike a calculator, which can't write a Shakespeare poem, can't help me fry an egg, these machines ultimately would be able to, if they became super intelligent, be able to do everything a human could do better.
- [08:13]
Matt Clifford
That's the goal of the companies building them. And, you know, there's lots of different definitions of AGI, artificial general intelligence. And then you've already mentioned super intelligence, which is almost a stage beyond where the machine can do more than any human can do. The definition I like of AGI, or at least I think a helpful starting point, is imagine an artificial intelligence that can do anything a remote worker can do. Now, that's important because, of course, part of the way that you think is actually embodied. You have a body and you use it to do all sorts of things. But there are lots of tasks today, particularly after the pandemic, that the only connection to the real world you need is a mouse and a keyboard and a screen. So one way of thinking about what AGI is is it's an artificial intelligence system that is a plug and play replacement for anything a remote worker can do.
- [08:58]
Rory Stewart
Okay, so let's start with what for most of us was the big wake up moment, which was the release of a ChatGPT3. And it was the moment where suddenly, on the market, we woke up to something that we could talk to that was no longer a kind of clunky chatbot giving us weird answers. We'd had those sort of things for 10, 15 years. When you're trying to, I don't know, deal with what's going wrong with your telephone, your bank, suddenly you've got something that was able almost immediately to do extraordinary things. And to give you an example, on my way into work yesterday, I asked it to analyze all multilateral institutions in the world this year and bring out common themes from all the different multilateral conferences. And here I saw both its strengths and its weaknesses. It managed to produce this incredibly beautifully organized, precise, tailored answer, going right up to what had happened the previous day. It wasn't doing it through finding an article on Google, nobody had written about this subject up to yesterday, before it thought of dimensions I would never have thought of, historical, economic. It personalized it to me in a way that was extraordinary. But it also got stuff wrong. For example, it complimented me on my amazing job administering Mosul in northern Iraq. And I never administered Mosul in northern Iraq. And indeed what I did in Iraq wasn't that amazing. So it was a sort of wonderful kind of example of incredible power fluency and getting stuff a bit wrong. I mean, is that your sense too, that this is really when the public woke up was the release of these things that we call large language models? We'll get on to what they are in a second.
- [10:41]
Matt Clifford
Absolutely. And you know, I think for people who'd been in the field for a while, in a way it was quite a surprise because you could look at this discontinuous moment. We go from AI basically being very hard to get on anyone's agenda to suddenly everyone wants to talk about AI all the time. The other way of looking is actually a smooth curve. It's just that it finally crossed a threshold where it was useful in the ways that you describe. One thing I would say which I find really interesting is when you talk privately to the people at OpenAI who released that particular product, ChatGPT, they were also taken by surprise. You know, they did not expect it to become the fastest growing consumer application of all time. I mean, it's worth just giving the stats on that. So within two months of its release, 100 million people were using it. That's never happened before, ever. You know, the fastest growing consumer app before that was TikTok, and that took nine months to get to 100 million. So there was something very special, but they didn't know it. I think the team size at the time within OpenAI, working on ChatGPT specifically at the time was like five or six people. So it was, it was a huge moment.
- [11:42]
Rory Stewart
What are things that people don't understand about what these things can do? I mean, I often meet people who will say, well, it's not that awesome. It's not that very different from a Google search. It's not very useful for me. I don't think it's really going to make any difference to my company. I don't think it's really making any difference to the world. Are there a few ways of really making, I don't being unfair to my mother, but making my mother wake up and realize that this thing really is extraordinary. How would you illustrate how amazing this New technology is and why it could make huge differences.
- [12:13]
Matt Clifford
Yeah, I think it's really instructive to look at the history of chess. So there was a long period of time where humans were just obviously better than machines. Then there's the phase.
- [12:23]
Rory Stewart
And we were very proud of it.
- [12:24]
Matt Clifford
Extremely proud of it. In fact, we thought chess was so hard and complicated that it would be like one of the last things to go. I think, you know, like this was like the, you know, we would hold up these gardens of chess as like the paragons of what human intelligence was.
- [12:36]
Rory Stewart
Chess grandmasters, the Kasparos and Kasparos were considered. The Bobby Fishers were considered genius.
- [12:41]
Matt Clifford
They are in that domain. And then of course, in the late 90s, you have this very famous match between Kasparov and Deep Blue, which, by the way, is not a modern AI system at all. It's a different type of system, Completely different type system. But, you know, to cut a long story short, Deep Blue wins. The machine is better than the human.
- [12:57]
Rory Stewart
Right? And then there's an extraordinary moment which I remember, and Gary Marcus, I think, writes about this at nyu. He's a bit of an AI skeptic and he's saying, well, yeah, okay, Deep Blue has managed to beat Kasparov, but actually, and this is now going back 10 years, we've discovered that a human and a computer, this is going to get to you better than any computer in the world. And so there were these extraordinary teams of humans and computers. And if you were a chess grandmaster, you were quite cheered up by that because you still seem to be very necessary to the system. You could make the computer better. And then.
- [13:28]
Matt Clifford
Yeah, so some people call this like the Centaur era on the basis that the centaur is the hybrid. And human machine teams were the best, sadly, or not, depending on how you look at it today, the human has nothing to contribute. You really get nothing from adding the human into the mix.
- [13:44]
Rory Stewart
It makes it worse.
- [13:45]
Matt Clifford
It makes it worse, yeah. And so, sadly, I'm 40 years old, but I employ some young people and they have this great phrase of that's cope, meaning you're telling yourself that to feel better about the world. I think the idea that there's this long term stable equilibrium of the Centaur era of human machine interaction being the best is probably cope.
- [14:06]
Rory Stewart
Thank you for listening. If you enjoyed that and you'd like to hear more from this series on AI from Matt Clifford and me, getting into the real issues, sign up at thererestlesspolitics.