Is AI Coming for Your Job? (Your Radical Questions with Matt Clifford)

0:00This BBC podcast is supported by ads outside the UK.

0:05Amol Rajan:Hello, I'm Dr Joyce Esther, President of Governor State University. At GovState, we're more than a campus. We're a community that supports you every step of the way. Whether you're returning to finish your degree, advancing your career or starting a new chapter, you have a home here. Let us help you reach your goals with flexible scheduling, caring faculty and programs that meet today's workforce demands. Learn more at govst.edu. GovState, affordable, exceptional, achievable. Step into a world reborn. Explore the fascinating story of life after the extinction of the dinosaurs, when enormous snakes, tiny horses, towering penguins, and strange primates rule the tropical earth.

0:49After the Age of Dinosaurs reveals a chapter in Earth's history that you never knew you were missing. Now on view at the Field Museum.

0:58Amol Rajan:Buy tickets now at fieldmuseum.org. That's fieldmuseum.org.

1:08Amol Rajan:Hello and welcome to Your Radical Questions. This is where I put your questions to one of our radical guests. This is your chance to engage very, very directly with the very smart, very interesting and very influential people that we have on this podcast and ask them about their ideas for the future. And if you heard Thursday's show, we had the magnificent Matt Clifford on He is the co-founder of Entrepreneurs First. He just stepped back from that after 15 years. Former AI advisor to Prime Ministers Sunak and Starmer. And who knows what he's going to do next. But he's got some very radical ideas to make this country rich again by embracing innovation, science, technology and AI and perhaps refashioning our state a little bit to make that happen.

1:47Amol Rajan:We talked about all of that in our last episode. Please do go back and listen to that. Matt Clifford is here and we have been bombarded with questions. You haven't just got your mates in the AI community to send us these questions, have you? They're far too hard, I definitely would have got them to send easier ones. They're hard and they're a lot. You think, is anyone going to send in questions for an AI super wonk? And here we go. Lots of them have come in. Let's crack on with it. This is the first one. It's a voice note from Flora. Hi, Amal. My name is Flora Nutchins. I've got a question for Matt Clifford.

2:15Amol Rajan:I'm an arts person, but I developed a love of science through my training to be a conservative of historic textiles. I'm interested in the relationship between creativity and science. In his book At Home, Bill Bryson talked about how many great scientific discoveries were made by vicars who had a crucial set of circumstances in their favour. Excellent, ironically mostly arts-based education and an intellectual network of other clerics and university fellows. A comfortable home and income. A job that before the Evangelical Revolution was relatively undemanding. Servants and a wife or sister to run their households for them.

2:53Amol Rajan:I'm concerned that modern education society doesn't provide time space and comfort for deep thinking or creativity Capitalism does not tolerate unproductive people or spaces or allow anyone to make mistakes or go up blind alleys How can we foster the intellectual creativity and plasticity that great scientific leaps forward require? Wow Flora I feel like I've just been sort of on a tour through centuries of history and education At the core of that is a concern to make sure that we have the time and the freedom to pursue our minds and to pursue kind of those kind of imaginative wormholes where great things happen.

3:38Amol Rajan:I'm going to just send it to one bit. I'm concerned that modern education in society doesn't provide time, space and comfort for deep thinking or creativity. How, Matt Clifford, can we create more of that time, space and comfort? I mean, I agree with the general thrust that actually that time and space is key to breakthroughs, intellectual, scientific, etc. I actually am a bit more optimistic than four on this. I think it's very easy to romanticise the past, but how many people were actually vicars with that time and space and a servant and a wife and a sister? There were a fair few. There's obviously a parish church in every village, but they were obviously also of a particular demographic.

4:17It wasn't some great meritocratic moment. And actually, I would argue that, you know, not to constantly be the defender of capitalism, but modern capitalism plus philanthropy has actually created a lot of these spaces. I'm super fortunate that a lot of my worldview was shaped by a year that I spent studying in the US at MIT, paid for by the Kennedy Memorial Trust in the UK, which was funded by philanthropy and still is. But actually, I think that we were actually pretty good at parts of this and it's much more meritocratic and open than it was before. I also think, though, that this is one of the main reasons that I started Entrepreneurs First with my co-founder Alice was that I do think it's true for a very particular kind of pursuit, which is entrepreneurship.

5:03So one of the things that I think is a big problem is that entrepreneurship is actually, in theory, one of the great levelers, one of the great ways to allow people from any background of talent to kind of have this uncapped upside. But actually, can you afford to start a company is a really difficult question. Most people can't. And so one of the things that we pioneered at EF was paying people to quit their jobs and paying their delivery expenses for three months to explore. And when we started doing that in 2011, people thought it was really weird and niche and fortunately worked out. And actually, there's a lot of things like this today.

5:37So I don't want to claim that we live in this perfect world where everyone has time and space to think. But actually, I think relative to 19th century Church of England, there's a lot more meritocratic space for this to thrive today. Would I like there to be more? Absolutely. And that's why we need to be a richer country that can afford to pay for it.

5:51Amol Rajan:The one other sort of subplot to Flora's magnificent question, for which, thank you, Flora, is this concern that we're sort of, I think, so overstimulated. We've got so much sort of incoming that we don't create the space for boredom. Some of the investors and Silicon Valley tech bros and sisters that I follow have this kind of refrain where they talk about how the great scientific leap forward or leaps forward that Flora refers to often arise from boredom. So Isaac Newton, Under the Apple Tree, Archimedes, this idea that basically you have this kind of incredible insight when your mind is free to wander.

6:24Amol Rajan:Just away from your knowledge of AI and tech and entrepreneurs, as a parent of two kids, do you worry a bit about social media and smartphones and this constant incoming? Do you feel kind of this avalanche of info coming at you and want things to slow down sometimes? I do, but I think it's really important not to take away our own agency in this. So I think it's very tempting to always blame the system. It was so hard in capitalism, in modern society. Go for a walk. Go for a walk. I do this all the time because I know that I'm actually an information addict. I think that's served me very well in some ways, but it absolutely means that if I don't consciously take time to not have a flow of information, then thoughts of my own don't have a chance to bubble up.

7:08So I take a walk. Capitalism isn't stopping me from taking a walk. That's so interesting, sir,

7:13Amol Rajan:because we did an episode with the pollster James Kanagasurium who is very, very influential and he was sort of arguing, this is just a few episodes ago, if people want to go back to see it in the feed, he was saying that one of the big ideas in British politics is agency. In other words, people feel a lack of agency, an inability to affect their circumstances. And for some people, people who have got some material comfort, maybe a bit of disposable income, you sort of want to say to them, you be the master of your destiny. Take that agency back. You take back control. yes it's the people who really don't have agency because they're struggling to feed their kids they're the ones who've got a legitimate complaint i think both things are true at once so i think simultaneously yes the system our system could do a lot more to encourage and enable agency and we should absolutely do that all the ways that you described and i also think we need to take more responsibility and you know you mentioned parenting i actually think that this is an area where you know we should we should ask for intervention a bit less and take a bit more responsibility Flora, thank you so much for your question.

8:11Amol Rajan:Thank you very much, D-Matt, for your answer. This is from James. Hi, I'm James. I'm sending this from a boat in the Solent. So my question is in relation to how the government can make policy effectively when there is so much uncertainty around the development of AI, the ways in which threat and opportunity can manifest, which we're seeing happening at an almost weekly level, and certainly potential landing zones and kind of areas that are going to be emergent. I think within that context, there's a considerable challenge to be able to know which areas to focus on, to which to prioritise, and how to marshal interest and engagement from government departments.

8:52So how would you, if you were in government, approach this challenge and really which areas would you prioritise and how would you go about really crystallising that narrative?

9:03Amol Rajan:James thank you just to confirm Matt James isn't one of your three siblings pretending to be in the Solent when actually he's living at home with your parents because James we set up this bonus episode on Mondays to try and bring you guys closer to people who can actually answer your questions and I pursue those answers on your behalf no man in the history of the universe is better qualified to answer this question than Matt Clifford especially when James refers to potential landing zones which is a subject that Matt mentioned on Thursday but Matt go on take it away I think this is a great question.

9:33It's not from one of my siblings. I think there are three things that we have to get right if we want to do this well. And James is absolutely right to say that this is decision making uncertainty, under uncertainty. But of course, most of what government does is that in various ways. So what do you need? Well, you need good information about the state of the world. You need proper scenario planning. And then you need to be able to be willing to make some bets to allocate your chips, as it were. again um uh figuratively so how we're doing on that in the uk well actually on the first on do we have good information actually i think this is probably one of the areas that the uk is world leading i think the the ai security institute in the uk is i was going to say probably i think it is the the best state capacity in the world for understanding what's what's happening in ai so when anthropic uh announced their mythos model um a couple of weeks ago the one government agency in the world that had tested it in advance and knew exactly how good it was at what was the UK's AI Security Institute.

10:33So in a way, our political leaders have no excuse for not knowing what's going on. They do have that situational awareness. The second is scenario planning. Here, I think we could do better. I think exactly as James said, the range of outcomes is pretty large. Now, obviously, you can't make all the investments that you'd like to make for every outcome, but you at least need to think through what it would look like. So on Thursday on the show, We talked about, for example, what is going to be the labour share of income in a world of AI transformation? Do we actually have a plan for the world where that's a lot in the lower end of scenarios?

11:06Do we have a plan where, again, back to one of our other conversations, the City of London loses out significantly relative to faster adopting financial centres? I worry today we don't have enough serious thinking about some of these scenarios. And that's where I think we could do more. And then the third, like, where do you want to place your bets? So, like, some things like writing scenarios are relatively resource unconstrained. Obviously, you have infinite civil service to work with, or you do have AI, and, you know, it's more infinite. But some things like investment, you know, we obviously have relatively limited resources.

11:39And that's where I think you have to make some, a mixture of no regrets moves, which are useful in every scenario, and then some just big bets. And, you know, what I really like, I was going to say, what I really like about the AI opportunities action plan, which I wrote. is that I think the no regrets moves are we have to have domestic compute. We have to be building. We have to be making it easy for the private sector to build data centers in the UK. By the way, that's not a lot of government money. I don't want to turn just too much down a sidetrack, but something I think about a lot when people say, oh, you know, it's a drop in the ocean what the UK government can spend.

12:13I think in the spending review, we got$2 billion for AI, which is 0.07 % of GDP. you compare this to in the peak year of the railway revolution in 1847, we spent 7 % of GDP on the railway build-out. So if we did that, it would be£210 billion a year. However, that was not public set to money. In fact, the UK had the most privately funded railways build-out in the world. I think that's the solution for us in the AI revolution. Government's not going to spend£210 billion on AI infrastructure, and it shouldn't. Actually, international investors are lining up. they would love to invest more in the UK.

12:53So it's back to the conversation we had on Thursday about how do we make that possible? How do we make the planning system, the energy system, fit for purpose so that private capital could pour in? That's one of the ways that governments can make it, I think, a no-regrets move.

13:05Amol Rajan:Well, that's an incredibly comprehensive answer to James' question. Thank you. I've got two things that come to mind. First of all, 1847, that's a bit early for the peak of the railroad revolution, isn't it? It's the peak year for the amount that was spent in this country on railway build-up. I'm obsessed with railways, so I was perturbed by that, but I'll take your word for it. I'm now going to look it up. I'm actually going to go and look it up because I thought I knew about the mid-19th century. Anyway, the other thing was just briefly, you mentioned Mythos Preview. I'm just really interested.

13:31Amol Rajan:So for those who are listening to this, a few weeks after this was announced, Anthropic came out with an update to its clawed chatbot, which basically was called Mythos Preview, and they kind of put it back in a box because they were really worried that it exposed vulnerabilities in all the world's web browsers, potentially compromising systems to do with transport and finance and so on. someone steeped in AI I'm just really interested in your point of view as to whether or not you're kind of at all surprised by that you know if you're kind of someone who's been watching this over the last 15 years do you see what Anthropics say about Mythos Preview and go yep that was inevitable which isn't saying you're relaxed about it it's such a great question it's totally predictable and the surprise is how well following a straight line on a graph has held up as the best way to predict it's like a shocking moment but not surprising given the trend or something let me take one minute on this because I do think I often say when I'm speaking to corporate leaders, political leaders, the most important fact to know about AI is not where we are at any given moment.

14:27It's what the shape of the curve is.

14:30Amol Rajan:Welcome to a podcast called Radical, all about the underlying trends. And I think for very technical listeners, I acknowledge there's a ton of caveats to what I'm about to say. But I think the sort of bare, bold version of it is really important. So I'm going to give the un-caveated version. There is a little organization in California called METR, M-E-T-R. and they produce what others, not they, have called the most important graph in the world. I think it is the most important graph in the world. What it shows is over time, what is the longest, most complex task that you can delegate to an AI and have it at least half the time perform successfully?

15:05And they've been tracking this for a long time. Now it's in software engineering for now, which is super relevant to Mythos because that's obviously a cybersecurity-focused product. What's interesting is that not where we are at any one time. Today, I think we're at 10 hours. So roughly, if it takes a human software engineer 10 hours, the best AI today has about a 50 % chance of doing it successfully in a fraction of that time, of course. That's not the interesting thing. The interesting thing is the doubling time on that, which is currently, depending on whether you believe there's been a recent acceleration, which me to do, somewhere between four and a half months and seven months.

15:38Now, this I think is an astonishing fact about the world. So I used to say this to the prime minister when I was working there, the times moved on, but I said, today we're at two hours. But there's still seven more doublings before the next election. A world where you can delegate two hours of work to an AI is very different from one where you can delegate a working month. And actually, if you double two, seven times, you roughly get to a working month. And I think this is the point about Mythos. It's not surprising because if you just follow that line, it's exactly where you'd expect us to be.

16:07The surprising thing is the line has held up really well and doesn't seem to be slowing down.

16:12Amol Rajan:And yet you can still look at what it does and be absolutely flabbergasted by this Promethean technology. Matt, thank you so much. And James, thank you. Hope you're having a good time on the Solent if you're still here. In general, we want more. We want more. We want more voice notes from boats in wonderful parts of Britain. Thank you so much. I'm going to do the next two questions together because they're on a very similar subject. These are from Isabella, who's in Bristol, and Matthew. Thank you both. Isabella says, with AI being adopted by some companies as a way to replace part of their workforce, how can this coexist with expanding the wealth of the common person as it replaces their jobs and thus ultimately serving to protect the existing wealth of companies?

16:51Amol Rajan:Isabella, let me just say that we did talk about that, that we'll talk about again in a second. We talked about that on the main show on Thursday, and you'd be interested in what Matt had to say there. Matthew says, similarly, there isn't a lack of wealth in the UK. There is a lack of wealth equality. how can an AI revolution be guided to benefit all in society, not just to make the richest even richer? Yeah, I'm going to respectfully disagree with two of the premises here that I think is actually really important. So I don't think the best way to think about AI is about job replacement. I really do think it's about productivity increases.

17:25So when I think about this as an entrepreneur, which is what I spent most of my life doing, most of my career doing, Usually when you end up being able to do more with less, what you do is not cut, but grow. So if I employ 10 people and then I get this incredible productivity enhancing technology that means that one person can now do the work of 10, the natural response of an entrepreneur isn't to say, great, let get rid of nine people. It's do 10 times as much. Now, that might not be true in every part of the economy. And that's why we do need a safety net to cushion the blow from the people that lose out.

17:56But in general, most entrepreneurs, when given a chance to do more with less, will do more. And so actually, that's why historically wage growth has followed productivity growth.

18:06Amol Rajan:But will they do more and will they hire more if the people that in the past they were likely to hire were doing the kind of cognitive tasks that AI can replace? Who knows? I think it's very hard to say, but it goes back to something you said on Thursday, which is that jobs are actually bundles of tasks. So for sure, if your job is literally all tasks that can be done, you know, kind of one for one by an AI, then that job will disappear eventually. And that's terrible for you now, which is why we need the safety net. But it's really good for society. Like none of us want our children to be chimney sweeps.

18:38Amol Rajan:As Richard Thaler, the Nobel Prize winning economist, was sat in that very chair a couple of months ago. And he said about AI, he said, whoever wanted to be a toll booth operator. Right. And if AI could be a toll booth operator, then toll booth operators could go and do something else and hopefully get paid for it. Yeah, and we mustn't be callous about the impact on people that it affects right now. And again, that's why we need AI to succeed in this country so that we have the resources in order to be able to cushion that blow. How does Isabella, and Isabella, thank you for the question, and Matthew as well, if they're concerned to sort of upskill, retrain, but they want to make sure they've got the skills to, if not win in the AI economy, to give themselves the best chance of not being replaced because their tasks are automatable.

19:23Amol Rajan:What do they need to do? How do you upskill? Well, I think a lot of it is actually about being a skilled user of AI. You know, what I'm certainly seeing in our business is that, you know, among our younger members of staff, the people that are really curious about, you know, how can they use AI? They're just getting loads done. And obviously, they're the people that are thriving. And I do think that in general, thinking about replacement is just the wrong metaphor. Like, if you can... Think about enhancement instead. Think about enhancement. Now, again, I'm not trying to pretend that there will be no job replacement.

19:58Jobs are going to change a huge amount because of AI. But one thing I would say, and it's a thing I've said a lot to a lot of leaders, is there are no AI-shaped holes in the world. We've co-evolved the economy with human minds. And so the holes we have in the world are human-shaped. Now, over time, that will change. We'll start to, in the same way that, to begin with, the point of the steam revolution wasn't like plugging a steam engine into a hand loom. We had to create new types of organizations, new kinds of work to take advantage of steam. And we'll do that with AI. But there is something incredibly special about human mind and the whole economy has co-evolved in it.