Cautionary Tales with Tim Harford - LIVE: We Are Not Machines - with Sarah O'Connor

Episode Date: July 3, 2026

When the Luddites smashed factory frames in a bid to defend their craft and livelihoods, the machines came out on top. Today, AI and automation threaten to wipe out skilled jobs and flood us with inferior products at a fraction of the price: is history going to repeat itself? FT journalist and author of We Are Not Machines Sarah O'Connor joins Tim to discuss her findings from the front lines about the future of work. Which professions are most at risk, and is there anything workers can do about it?   See the show notes at TimHarford.com Sarah's book, "We Are Not Machines: The Fight for the Future of Work", is available here (UK) and for pre-order here (US). See omnystudio.com/listener for privacy information.

Pushkin. Tim here with an exciting announcement about the Cautionary Club, our Patreon community for Cautionary Tales listeners. So many of you joined our live table read earlier this year that we're doing it again. Join me and my production team for a live reading of an unreleased episode about the people who almost invented the iPhone 15 years early and the surprising reasons they failed. This will be a chance for you to see how the stories we tell are developed in real time

and ask your burning questions about cautionary tales. It's on the 8th of July at 5pm UK or noon eastern. If you join the Patreon before then or if you're already a member, you'll get your exclusive invitation. Head to patreon.com slash cautionary club. That's Patreon, P-A-T-R-E-O-N-com, slash cautionary club. This is an IHeart podcast.

Guaranteed human. Run a business and not thinking about podcasting. Think again. More Americans listen to podcasts than add supported streaming music from Spotify and Pandora. And as the number one podcaster, IHearts twice as large as the next two combined. Learn how podcasting can help your business.

Call 844-844-I-Hart. Mrs. Goodair is enjoying a leisurely breakfast while she reads the morning paper. It's April 18. and the headlines are dominated by Wellington's military successes in Spain. A strange noise jolts her from her reading. It sounds like the iron gates to her home rattling. She rushes upstairs to get a better view and sees the mob, for it is a mob, more clearly.

They're throwing stones, making lewd gestures and shouting for her to open the gates. Two men appear to be wearing dresses, proclaiming loudly to be General Ludd's wives. She knows exactly what this is about. It's about what's in her husband's mill next door. So why are they at her house? Her husband, John Goodair, is out of town. She hovers at the window, unsure of what to do. The mob leaders grow impatient.

They shout to their comrades that it's time to move on and begin marching through the streets of Stockport. As they weave through the town, they smash windows and break down factory doors. By noon, the mob has swelled over 2,000 men. Soon they run out of targets, so they turn back to where they began. John Gooder's enormous mill.

The place many of them have spent their working lives. Once a symbol of pride in their craft, now a symbol of everything they feel is wrong with the industrialized world. They break down the doors, smash through the windows, and destroy all they can inside, every last frame, every last spindle. The mob has grown so large, but even the iron gates to the good air home can't hold them back. Mrs. Goodair has already gone. A wise move. The Luddites burn the house to the ground.

I'm Tim Harford, and you're listening to cautionary tales live at the Bristol Festival of Economics. Loyal listeners will remember this is not the first time the Luddites have featured on this podcast. Two years ago, I told you about the failure of the Luddite Revolution. the weavers who were incensed that automation was devaluing their craft, in general Ludd's rage against the machines. Today, the Luddites seem more relevant than ever, as the so-called AI revolution threatens to overturn the world of work. So should we be smashing up large language models? Or is resisting as futile as it was for the Luddites? Who's going to be most affected?

And have economists been asking the wrong questions all along? I'm joined by a very special guest, Sarah O'Connor. She's a Financial Times journalist with her specialism in technology and work, and she's written a book called We Are Not Machines, the fight for the future of work. She is the perfect person to guide us through what's coming next, and she's got plenty of cautionary tales to tell us. Welcome, Sarah.

Hello, Tim. Thank you for having me. Oh, it's a pleasure. We've got to get something out of the way first. So Sarah O'Connor, have you seen The Terminator movies yet? No. You're going to have to watch them. It's a point of principle. Yeah, it now feels like stubbornness.

Yes, so you might have noticed that my name bears a faint resemblance to the heroine of the Terminator films. Growing up occasionally people made Terminator jokes to me, and as a result, I never watched them on principle because it annoyed me. And so about a decade ago, I went accidentally, sort of globally viral when I very simply tweeted a news story that I'd seen come out, which was that a robot had killed a worker in a Volkswagen factory and didn't realize that this would cause the internet to break. So, yeah, so people just found it very hilarious that someone called Sarah O'Connor had written about robots. And for about a decade, people have occasionally sent me, you know, terminated gifts. posters of the Terminator films with my face photoshopped in. And as a result, no, I have not watched those films.

Okay. So the book is fantastic. We are not machines. What's the headline? What are you basically driving out with this book? I think what I'm trying to say in the book is that, you know, we're all sort of drowning in information, predictions about AI and what it might do to the world of work. And a lot of those are coming from either economists or.

or the big tech executives who have created large language models and are marketing them. And both these sources are unreliable for different reasons. Yeah, the tech guys have something to sell, right? And so they have their own motivations. And then, you know, economists, I think, are trying their best to figure out what's going on. But the way in which they're doing that, I think, feels very abstract to me. As someone, you know, I'm a reporter at heart. So my favorite thing to do is to, like, get my note.

book, put my boots on and go and meet people and stand outside the factory gates and talk to people about what's actually happening. And so over the past few years, I've just felt very frustrated that everything that I'm hearing about this is coming from people who are either looking at spreadsheets and models or who have a product to sell. And I just wanted to hear less about what these men say is going to happen and more about what's already actually happening from the people that it's actually happening to. So in the book, I basically went to find people and places and where, workplaces that are on the front lines of what's already happening. So people who are encountering autonomous robots in their workplaces

or are working with or four or around large luggage models and they're starting to change the way they work. And when I did that, what I realised was that, you know, I wasn't really hearing this very utopian story that this is liberating me, this is freeing me from dull, dirty, dangerous work and making everything fantastic. But nor was I really hearing a lot of stories. that were like my job has been completely wiped out.

What I was hearing instead were stories about quite profound changes to the nature of people's jobs and how they felt about their jobs and whether they enjoyed their jobs or not. And some of those were quite cautionary tales. You know, there were a lot of people who were saying, actually what's happening to me is almost the opposite of this idea that technology is going to give us space to be more human. People were finding themselves sort of crunched into systems

that were paced by machines. run by machines in which judgment was being overtaken by machines, in which creativity was being sort of compressed or contained by machines. And I thought actually if that's what's beginning to happen to people, then everyone should know about that and we should try and talk about why that's happening and whether there is a way to avoid that, because fundamentally I don't think that's what anybody wants. Before we sort of take a trip to the cutting edge, I just wanted to go back a couple of hundred years. We began with the Luddites.

Are they still relevant to people? people misunderstand the story of the Luddites? Yeah, I think they are still relevant. I mean, Luddite is just now used as a sort of a flippant term for somebody who's sort of reflexively anti-technology and just doesn't like any new tech. But actually the Luddites, they weren't really fighting against technology full stop. They were fighting against the way in which machines were being put to a new use, which was to cut them out of the equation. These sort of knitting and wide-frame machines, they weren't displacing work entirely. what they did was they allowed people who had no skill and no experience to make a less good quality product.

And in that sense, as we might come on to, it actually has an awful lot of similarities with some of the technologies that we're seeing today. Instead of knitting a stocking, you knit a massive bit of cloth that you can make a punch stockings out of and you just cut them into strips and then you make the stockings. The stockings are terrible. They fall apart. So you've suddenly got interesting, well-paid jobs being replaced by boring, badly paid jobs. you've got expensive, high-quality work being replaced by cheap, crappy work. Even if they were lower quality, they were also a much lower price. And I think that was a deal that a lot of consumers were willing to make.

So let's talk about warehouses. The picture I have in my mind, or the picture I had in my mind before I read the book, was in fact, I realized a picture painted for me by you about 10 years ago. one of the things that some workers in these had was they had this earpiece, the Jennifer unit, basically just a voice in their ear, telling them where to go in the warehouse, what to pull off the shelves.

You don't need them to think about the best way to get around the warehouse or to remember anything, just to bathe the voice in your head. So this sort of creation of these flesh robots. So I was reading this article by you about 10 years ago and thinking, that sounds bad. But of course, that's 10 years ago.

So how accurate is that image of what is now happening? What's changed? Yeah, so quite a lot has changed. And actually, I think that article from 10 years ago and that sort of reporting that I've done and just to clarify, the Jennifer unit isn't something that's used in Amazon warehouses, but it is used in some other warehouses.

Really informed the way I used to think about technology. So, yeah, I had met and interviewed a lot of people over the years who worked in jobs in which they were basically expected to work like robots. I used to think, well, you know, bring on the damn robots then. You know, like let's get real robots in to do these jobs. Like this is a waste of kind of human potential. And this is a great example of where automation would be brilliant. So I used to be quite a techno-optimist in the sense that I thought there were huge numbers of roles that actually could and should be automated.

So what's happened now in Amazon warehouses, at least in the one that I went to visit, is that the robots have arrived, but not really in the way that I'm, had imagined. So I'd basically imagined like a one-for-one swap, right? So instead of a person walking around, you'd have like a kind of human, humanoid robot doing it. Yeah, but it's never actually like that, is it? No, it's never, ever like that. Like a robot accountant is, is basically Microsoft Excel. A robot accountant is not C3PO sitting in your chair. Tumping away your laptop, no, exactly. And it's not how it works in Amazon either. So I went to visit a new Amazon warehouse, which has their sort of latest automation technology. And then, The way it works now is that rather than you as a worker walking around for like maybe 10 miles a day,

just weaving between all of these shelves and picking things off, now on each floor, there's a big fence. And around the perimeter of that fence, the workers stand stationary in one place. And inside the fence are the robots, but the robots are basically like giant rhumbas. And what they do is they drive around inside this perimeter fence and they pick up the shelves and they bring the shelves to the workers. And so you now, as a worker, you stand in one place and a robot, brings you a shelf and then a light illuminates which part of the shelf you need to look at and a screen gives you a picture of what the thing is you need to pick off. And I tried this out.

I think I had to pick off like a mobile phone case with a dolphin on it. So you know, you pick it out and then another light illuminates which box that you need to put it into and you push a big button. And then that robot takes that shelf away and the next robot's already queuing up for you with the next shelf and you do the same thing. How long did you do it for? Two minutes. Okay. How long do they do it for? 10 hours a day, 40 hours a week with two 30 minute breaks. And so this is not what I had imagined when I sort of cheered on the robots and thought this would be great. You know, in some ways, I think that job is better. It is less physically demanding than walking around for 10 hours a day.

Amazon says that it's safer, so they have, you know, fewer injuries and accidents. And I think that's probably true. But if anything, I think it's also become more boring and more monotonous. And also, it's still not that easy on the human body to stand up for 10 hours a day. And because they're now much more productive, right, because you're not, you're not walking around from one place to the next. And so the products are coming at you. So you're bending, you're lifting, you're twisting over and over and over again. So I think it's still quite a physically demanding job and arguably like a bit more monotonous

and a bit lonelier. Presumably there are people who did the old job and now do the new job. So what do they think? Well, I interviewed one worker who had volunteered. to transfer from a manual old-fashioned warehouse to a new one. And he wanted to go back. He said, actually, I don't like working with the robots. It's too intense. It's too exhausting. It's really lonely. I don't get to talk to anyone anymore. And he, yeah, he's now transferred back to his old warehouse. By way of contrast, you also visited a mine in Sweden. So how is automation

taking off there? Yeah, so this mine in Sweden was right up near the sort of fringe of the Arctic Circle and they mine copper there and various other minerals. And it is one of the most sort of technologically advanced mines in the world, or so they say, they say. So it used to be that miners would drive vehicles around below grass, very deep down. It's horrible down there. It's like really dark. It's really humid and claustrophobic. And now the vehicles are autonomous, so they drive themselves. And the miners sit in a control room. They can take over the vehicles if they need to, but a lot of the time they're just, they're sitting there, they're watching what's happening on big screens, they're in a comfy chair, and they're listening to Spotify, they're listening to the ice hockey

and the ones that I spoke to were quite satisfied with this change in their working conditions. Is the fact that the Swedish mine workers are happy and the Amazon workers at best ambivalent about the way their job works? What explains the difference? So I think there's a few differences. One difference is that, you know, the reason that in Amazon they don't have humanoid's walking, around doing the whole thing is that that technology is not ready yet. And so in some ways, humans are still kind of plugged into a system which is like part robot, part human, but is increasingly being paced by the capabilities of the robots. Whereas I guess for the miners, they do not need to be in the cab anymore. And so that means that there's the potential

for a much kind of greater improvement in their health and safety. But as well as that, I think there is a governance thing. So I don't know how much everyone knows about Sweden, but they have a very distinctive kind of labour market whereby trade unions are very powerful. And there is a rule that any new technology has to be bargained collectively about before it happens. And so before they introduced this new technology, they had to sit down with the workers and they had to talk about it and how it would go and what the workers' concerns were and what the company wanted to get out of it. And, you know, more broadly in Scandinavian countries, what you find is that when people answer surveys about how they feel about technology, people are much more positive than they are

in the UK or in the US. And I think that's because they have a sense that they will have a seat at the table when those decisions get made. Thank you, Sarah. You're listening to a special cautionary conversation recorded live at the Bristol Festival of Economics. And after the break, my guest Sarah O'Connor will be telling a cautionary tale about language translators and artificial intelligence. Stay with us. Listen. And you're there for heart-wrenching knockouts. The world's biggest stage.

And breathtaking triumph. In 2026 FIFA World Cup, the knockout stage. Every match, every moment. Listen on TSN. Radio. Join the globe on the road to the July 19th final. 2026 FIFA World Cup. Stream it all live on TSN Radio. Available on IHeart Radio.

Run a business and not thinking about podcasting, think again. More Americans listen to podcasts than ad-supported streaming music from Spotify and Pandora. And as the number one podcaster, IHearts twice as large as the next two combined. So whatever your customers listen to, they'll hear your message. Plus, only I-Heart can extend your message to audiences. across broadcast radio. Think podcasting can help your business. Think IHeart. Streaming, radio, and podcasting. Call 844-844-I-Hart to get started. That's 844-844-I-Hart. Tim here with an exciting announcement about the Cautionary Club, our Patreon community for Cautionary Tales

listeners. So many of you joined our live table read earlier this year that we're doing it again. Join me and my production team for a live reading of an unreasonable. released episode about the people who almost invented the iPhone 15 years early and the surprising reasons they failed. This will be a chance for you to see how the stories we tell are developed in real time and ask your burning questions about cautionary tales. It's on the 8th of July at 5pm UK or noon Eastern. If you join the Patreon before then or if you're already a member, you'll get your exclusive invitation. Head to patreon.com slash cautionary club.

That's Patreon, P-A-T-R-E-O-N dot com slash cautionary club. We're back. This is the Bristol Festival of Economics. I am Tim Harford and my special guest is Financial Times journalist and author, Sarah O'Connor. So, Sarah, perhaps the canaries in the coal mine. are the translators. When I thought about translators, I just thought, oh, you guys are just all, you've lost all your jobs because of Google Translate. It's all gone. But actually, the story

you tell in the book is a lot more interesting and more complex than that. Yeah. So I interviewed some translators for my book. One of them is a guy called Peter. He lives in the Czech Republic, and he translates subtitles for TV shows, which is a great job. I mean, he told me lots of things about why it's more difficult than you might imagine to do that. So for example, you know, in English we just have one way of saying you, but in lots of languages, including Czech, there's like a formal tense of you and an informal tense of you. And so when you're writing the translated subtitles, you have to think like, how well do these characters know each other? Like, what tense would they be speaking in? And that might change, you know, through the course of an episode or

through the course of a series. And so there are lots of kind of interesting judgment calls that you have to make. And then, you know, translate. jokes is another thing that's like incredibly difficult because so many jokes are like culture-specific or they're like plays on language. We're in a bookshop. We are the guests of Waterstones in Bristol. There must be some asterix and obelix somewhere in the bookshop.

Yeah, they're the perfect example, right? So Obelix's dog in the original French is called Idaefeeks, which means obsession. And the brilliant English translation is dogmatics. You know, so like there's so much sort of creativity and humour involved in being a really good translator and so like this is a great job for people that enjoy doing that sort of thing. But what's happened to him and to people like him is that he has not lost his job entirely. Like it's actually quite hard to get a machine to translate things with that level

of cultural knowledge and understanding. So just to explain slightly the way a lot of translation works is a lot of translators are freelancers and there are agencies that will take work from, you know, a TV studio and then parcel it out to freelance translators. And what those agencies have started to do is to take something, get it translated by a large language model or a different kind of machine translation service, and then give it to the translator and say, hey, can you just check that this is right and maybe like just finesse it a little bit, polish it, make it sound a bit more human. And this has become the kind of bet noir of lots of translators because what they say is that it's, A, you know, you're expected to do this much faster.

and B, you expect to do it for much less money. But in fact, if you care about quality, it's very difficult to actually check a translation is accurate. And then even things like trying to make it sound a bit better. Like, that's actually what the translator's told me was it's actually quite hard to do when there's already like an answer in front of you. And it feels much less creative and more kind of cumbersome and less enjoyable. And so a job that was sort of creative and challenging and interesting has become

faster, harder, more monotonous, feels more mechanical and fundamentally kind of less satisfying. And I think that is a real cautionary tale. And when I was speaking to them, it made me think a lot of the Luddites, because what the translators will tell you is that the end product is less good quality. Then there were even some studies that have checked this. So there was a big study that looked at a bunch of subtitles from TV shows before and after the introduction of machine translation post-editing. and the quality has kind of deteriorated. It's the stockings again. It's the stockings again, exactly.

And I guess partly, obviously we care about the quality of jobs. That's important. But I guess one of the interesting questions from the point of view of the consumer, because the whole idea here is even if individuals get worse jobs, the consumer gets more choice, more quality, lower prices. I always do sort of wonder, is, is the product actually better and will the market deliver kind of the right trade-offs or are we actually just going to make all kinds of mistakes and we end up with crappy products that

we don't want and I'm just I'm just wondering whether I know I don't sound like like an economist here but I'm wondering whether we're all just going to get stockings that fall apart and translations that we that are joyless and but some manager somewhere was convinced by some AI salesman somewhere that it would be, that it would be fine. Yeah, I mean, I think this is one of the things that I realize working on the book is that, you know, economists, when they're sort of doing their modelling about which jobs might be displaced or degraded or changed by technology, they look at like how good is the technology and compare that to how good is the human. But actually, like, can a robot do my job as well as me? It's not always a particularly useful question. Like, can someone persuade my

boss that a robot can do my job as well as me is a more relevant question or would my customers or consumers be content to have someone do my job less well than me but for a much cheaper price with the the case of the translators is this fundamentally about the technology this just happens to be a thing that computers can do at a certain speed with a certain quality that kind of is is maximally disruptive to the job of being a translator or is it actually you? more to do with the fact, for example, that all these people are freelancers? Is it actually to do with economic power and not to do with technology at all? I think it's to do with both.

But I think you make a good point, which is that often this question about, like, what will AI do to the world of work as seen as a technical question? But it's not just a question of the technical side. It's also a question of, like, consumer choice and institutions and culture and bargaining power and economic. power. So yeah, I mean, partly the reason it's been very disruptive to translators is that the translating is like basically all they do, like that one task of like translating something. If you're a freelance translator, that is pretty much your whole job. Whereas for a journalist,

writing an article is like actually quite a small part of my job. My job also involves doing a lot of reading, going out to interview people, doing stuff like this. And so it's more, it's more disruptive in some particular professions, depending on how much of your tasks are automatable. And then also, yeah, like your market power, right? If you're a freelancer, then that's a very different thing to being a well-protected miner in a unionized mine company in Sweden. And can we talk about coders?

Because superficially programmers, software coders, seems quite like translators. it turns out that for some reason large language models seem to be pretty good at coding at least that's what I'm told I wouldn't know but people say it's pretty good so do code is in the same position as translators suddenly I've got all this terrible code and kind of they want me to do twice as much work for half the salary do they feel the same way as the translator most of the ones I interviewed do not feel like that um I think coding is an example of a job that in many ways actually has probably been enriched and made more fun and more fun and more

productive by this new technology. And I think the reason that it is playing out different is partly, you know, this thing about bargaining power and where you sit in the sort of economic chain. But also, if you're a computer programmer, like you're, what you really care about and love is not literally sitting there and writing out lines of code. What you love doing is like solving problems. And so what the tools do, they might automate some of the kind of the actual literal coding, but they're not automating that kind of really fun, creative problem solving bit. It just allows you to like try things out or do stuff faster. This is the promise. This is what they keep telling us it's going to do for all the jobs.

It's going to do the boring stuff so you can do the fun stuff. And this is why they're so enthusiastic and think it's brilliant because for their jobs, it is pretty brilliant. What is it then that determines whether your job is like translating or whether your job like coding and is it under our control or not? Yeah, so I think one of the things that I've realized, and I now really flinch when I hear it, and I flinch when I think that I've probably said it sometimes as well in the past,

is that I think we're really using some of the wrong metaphors when we talk about this. So I don't know if you've noticed this, but I find that technology executives in particular, but also sometimes economists will talk about, like, there's a tsunami of change coming, or there is a new wave of technological change, I think it's misleading because it makes it sound like it's akin to a natural phenomenon.

Like it's something that's just happening that just came out of nowhere and it's going to happen to us. And it invites us to think that at best all we can do is, A, prepare for it and B, get ready to mop up after it. And, you know, these metaphors, I think, are really useful to technology executives in their conversations with policy makers because, you know, nobody wants to look like the fool who thinks you can hold back the tide, right? And so it invites you to think of this as something that is happening anyway. And, you know, there's very little that you can fundamentally do about it. But that's not true at all. That's never been true.

Technology is stuff that is made by people and implemented by people. And there is a huge variety, as we've just discussed, in the ways in which it can play out in different occupations and in different countries and in different parts of the labour market. and a lot of that is dictated by the choices that people make and the things they choose to value. So you can do something about this because you can write columns in the Financial Times telling Sam Altman that he's being a naughty boy.

What can ordinary people who do not have newspaper columns and who are not running AI companies, just regular people, what can they do to help direct this change in a more productive, empowering way? I mean, I think people do and are and will direct and determine this next progression of technological change, because that's what's always happened. I mean, some people obviously have more power than others, particularly in terms of what kind of working conditions they may or may not have to accept. But, you know, everyone that I interviewed in my book is responding to what's happening. And, you know, whether that is that they are going on strike or forming a union or whether.

it's that they are changing their profession or figuring out some new way of doing what they do in a different way. Also as consumers, you know, a lot of the questions we've been talking about is like, well, what will the market accept? We are the market, you know, so it's kind of up to us to decide whether we like AI music or real music, you know? I feel so flat that no one's ever said I'm the market before. You are the market. It's the nicest thing anyone's ever said to me. That's great. How to compliment an economist. Absolutely. So you heard it. Fight, fight back. Stand up. Show the man why you're better than the machine. Sarah and I are going to talk about large language models.

After the break. Listen. And you're there. For heart-wrenching knockouts. The world's biggest stage. And breathtaking triumph. In 2026, FIFA World Cup. The knockout stage.

Every match, every moment. Listen on TSN. Radio. Join the globe on the road to the July 19th final. 2026 FIFA World Cup. Stream it all live on TSN Radio. Available on IHeart Radio. Run a business and not thinking about podcasting, think again. More Americans listen to podcasts than ad-supported streaming music from Spotify and Pandora.

And as the number one podcaster, IHearts twice as large as the next two combined. So whatever your customers listen to, they'll hear your message. Plus, only IHeart can extend your message to audiences. across broadcast radio. Think podcasting can help your business. Think IHeart. Streaming, radio and podcasting. Call 844-844-I-Hart to get started. That's 844-844-I-Hart. Tim here with an exciting announcement about the Cautionary Club, our Patreon community for Cautionary Tales listeners. So many of you joined our live table read earlier this year that we're doing it again. Join me and my production team for a live reading of an unreleased episode about the people who

almost invented the iPhone 15 years early and the surprising reasons they failed. This will be a chance for you to see how the stories we tell are developed in real time and ask your burning questions about cautionary tales. It's on the 8th of July at 5pm UK or noon Eastern. If you join the Patreon before then, or if you're already a member, you'll get your exclusive invitation. Head to patreon.com slash cautionary club. That's Patreon, p-a-t-r-e-on.com slash cautionary club. We're back. I'm Tim Harford. We are live at the Bristol Festival of Economics, and this is a cautionary conversation

with expert in artificial intelligence, expert in workplace, is author of We Are Not Machines, Sarah O'Connor, who is also my colleague at the Financial Times. The title of the book is We Are Not Machines. You've given us some really striking examples of demands of made of workers to be more machine-like, being under constant surveillance, having to move in a very kind of weird and unnatural way. I wanted to bring it back to the experience a lot of white collar workers have of sort of sitting at a computer every day

and suddenly discovering there's this thing called chat GPT. Do you think that chat GPT, is this making us more machine-like and we don't know it? So this is something that a lot of the translators that I interviewed brought up. I think because they're linguists, they have like an ear for language

and how it's changing. but a number of them said to me independently that they had noticed that the sorts of language that we are using is becoming narrower and flatter and more homogenous. And I wonder if that is partly because we are using chat GPT more to write our LinkedIn posts and our emails and maybe that even seeps into the way that we communicate with one another. So I think there are certain risks there. But also, you know, I think large language models do clearly have the capacity to do that thing that we talked about at the start, which is like take away some of the

boring stuff. And I think there are plenty of examples of people doing that, whether it's like, oh, now I don't have to take any minutes for this meeting, or now I don't have to spend two hours Googling three relevant research articles. But I do think it's also, it's very easy to start also using it for things that maybe in the past we would have put a bit of human care and attention into, like, you know, I think a lot of people are using chat GPT now to write wedding speeches or speeches at funerals and that sort of thing, which, you know, you might once have thought were among the more human tasks that we did. Yeah. Although I have heard a few wedding speeches, chat GPT would have done better than some of them. So, you know, yeah, rising tide lifts

some boats. Are you frightened or are you hopeful for the future? I'm a lot more hopeful now than I was when I started writing the book, actually. And I think that's probably because of the people that I met along the way. So like another of these sort of metaphors or like cliches that we often hear is, you know, people will say, well, you know, inevitably there's going to be winners and losers. You know, there's dangers and opportunities. And that sort of invites us to think like, okay, well, I better just wait and find out if I'm going to be a winner or a loser and like, you know, fingers crossed, I'll be a winner.

But you just kind of got your lottery ticket and then you wait and see. But like, that's not actually what people are doing. like people actually are maneuvering themselves and trying to figure out what they can do to make sure that things work out well for them. Like everyone is shaping and determining this. So in that sense, I ended up more optimistic because I think that we have a lot more agency

to figure out how this goes and to make sure it goes in a way that we want than we're sometimes led to believe. Thank you so much, Sarah. Let's take some questions. So we've got a question from a cautionary club member Emily. By the way, anybody wants to join the cautionary club. It's very wonderful. You'd get the podcast ad-free, you get bonus episodes, all very good. And you get your question read out live at the

Bristol Festival of Economics. So Emily says there's a research institute called METR, which found that both the expected and the perceived efficiency improvements from using AI in software R&D were substantially overstated. She asks, do you think businesses are likely to scale back AI investment until the efficiency gains promised are demonstrated? Or have they sunk so much money in at this point that they feel they have to justify the expense? I think it remains to be seen. I think that METR study was really interesting because what it sort of highlights to me is that, you know, we were talking about this distinction between tasks and jobs and that actually your whole job might disappear but some tasks within it might change but when you think about something at the level of an entire organisation

you've also got to think about jobs and systems and so you know a computer programmer on his or her own might have become much more productive if they can use a coding AI a system but actually if there are still bottlenecks in other parts of that system you might not really see the benefit of it you You know, you might just create lots and lots and lots of code, but then you still don't have enough people to review it, for example, or, you know, whatever it might be. And so I think what a lot of companies are finding is that suddenly being able to do certain tasks within jobs much more or much faster doesn't necessarily mean that you're going to see the same kind of productivity outcome filter through to the whole organization. The other thing I would say is that we're still in the very early stages of companies trying to figure out how to use these new tools effectively. And what I think we might start to see is companies kind of redesigning workflows around what the machines can and can't do. I mean, in a way, that's exactly what Amazon has done, right?

I mean, it's completely changed its workflow in order to make the most of the machines that it has available to it at the moment and plugged humans in. to do the bits that the machines can't do. But I think it's still a really open question. You know, particularly with large language models, I mean, the kind of the hallucination problem, I think, could be quite existential for a lot of sort of high-risk professions. You know, certainly, you know, we're not going to be using large language models at the FT to write our articles because we just can't possibly trust them not to hallucinate.

Yeah. I mean, I was fine when I'm using large language models. The first question is, will anybody else ever see any of this? and if the answer is yes, then it's unusable. Like, if it's just for me, then sometimes it comes up with useful ideas, but I absolutely cannot trust it at all to produce anything that anybody else will ever see. So, anyway. More questions.

There were loads of questions. Can we get the mic? Somebody looking quite young and fresh face there. What a question from... Hi. I have a question. Which industry do you think is most at risk of... AI replacing the job or having the biggest amount of the job being taken away.

So Sarah, which industry is most at risk of just having these jobs completely removed? I mean, I hate to say it, but I'm... It's the translators? I know I was going to say it's, I think, sort of creative industries more generally. I mean, not that I think that, you know, people will stop wanting to pay for, human created stuff but I think that there are lots of people who work in you know they're not necessarily originating the ideas but they work in in the kind of in that long process between someone coming up with an idea and the thing being made whether that's like working in special

effects um in Hollywood or music production or um storyboarding in an advertising agency or whatever it might be or copywriting, I think there are lots and lots of jobs there which, you know, are good jobs and people like them. And so I think it's a real shame if they become disrupted. But I can imagine that those are the sorts of roles that could be quite vulnerable. But I might be wrong and I hope I am. So you heard it here first. Human creativity is dead. It's definitely what I said, thanks Tim. This is from the woman who calls herself an optimist. So thank you for that. We've got time for one more question. You said that AI is not advanced enough to translate the tax accurately enough, but it's really been around for two years.

Like woven by 2030 is going to be enough advanced to replace some of the jobs. Surely it will lead to like structural unemployment and you reckon government will let it happen. Thank you very much. So basically I think two ideas nested in that question. So One is this point that we can point to what the AIs can't do yet, but they change fast. So maybe a lot of the things that we're saying, oh, AI can't, the AI vision doesn't work that well, AI translation doesn't work that well. Maybe the answer is just wait. And then the second question is, if this really is incredibly disruptive to labour markets, our government's actually going to just do something drastic to outlaw it.

So the first part of your question, I mean, we've already seen how much better these things have got in the last few years. I mean, when things like Mid Journey first came out, the image generation, like every person's hand had like six or seven fingers, right? And we thought it was hilarious. And now they're not making those mistakes anymore. So yes, definitely things are getting better. And in a way, that's what's a bit difficult about writing about this topic is it's such a kind of moving target. So you're like you're trying to write about something that just keeps, keeps changing. But as I was saying, like, I don't actually think the capability of what they

can do is the only thing that matters, because it's also about, like, what kind of trade-offs are people willing to accept? And then in terms of if we started to see really big labor market dislocations, would government step in and do something drastic? Maybe. I mean, I don't think there's any precedent for that that I can think of. I mean, in the example of the Luddites, the government stepped in on behalf of the owners of the machines and sent the Luddites to Australia. So that was like the opposite. Well, the precedent surely is tariffs. Because, I mean, effectively, this is the point that economists like to make, but trade

with China, for example, it's effectively a technology. I mean, yeah, it's China over there, but it could be just some robots in a factory in Los Angeles next to the docks. and whether you're shipping grain into the robot factory and they're turning it into television sets or whether you're shipping the grain across the Pacific and the Chinese buy the grain and then they send you television sets.

I mean, it kind of doesn't matter. So you could imagine taxing robot-produced goods. I just don't know if people hate robots as much as they hate foreigners. Let's find out. That is sadly all we have time for. Thank you so much for coming. It's been an absolute pleasure to be here in Bristol. Thank you so much to my wonderful guest, Sarah O'Connor. Her book, We Are Not Machines, The Fight for the Future of Work, is out now. Do please pick up a copy. But before then,

please join me in thanking my guest, Sarah O'Connor. Portionary Tales is written by me, Tim Harford, with Andrew Wright, Ryan Dilley and Alice Fines. This live episode was produced by Georgia Mills with help from Mary. and Rust. The sound design of original music is the work of Pascal Wise. The sound engineer for tonight's show was Tom Dunn with Eduardo Galli. Thank you to Zania Levantis and the Bristol Festival of Economics. This is an I-Heart podcast. Guaranteed Human.