Sarah O'Connor: We Are Not Machines: The Fight for the Future of Work

Ben Zweig: Most coverage of AI and work is about what might happen to the quantity of jobs. My guest today went and found out what is happening to the quality of jobs inside Amazon warehouses, Swedish mines, newsrooms, and law firms. And she just wrote an excellent book about it. Sarah O'Connor is a columnist at the Financial Times and author of "We Are Not Machines, The Fight for the Future of Work." Here's our talk. All right, welcome to the Economics of Work. I'm excited to welcome Sarah O'Connor, journalist at the Financial Times and one of my favorites, one of my favorite journalists working on the future of work and author of the excellent new book, "We Are Not Machines." Sarah, welcome to the show.

Sarah O'Connor: Thank you for having me, Ben.

00:35 Who is fighting whom over the future of work?

Ben Zweig: Yeah, for sure. It's exciting. It was an excellent book. And the subtitle is "The Fight for the Future of Work." I really like it. I like the energy behind it. So like, it's like spicy, but tell us like, who's fighting whom and what are they fighting about?

Sarah O'Connor: I think there are a couple of ways to answer that question. I mean, the first thing I think is that one of the things that I realized while reporting the book and to give a kind of quick overview of what I tried to do in the book, it's basically my attempt to like inject some proper detailed rigorous on the ground reporting into this debate about AI, which has been dominated I think by really big picture predictions from, often like technology CEOs, also economists, like I'm a big fan of economists. I'm married to an economist. I have nothing against them. I write about economics.

Ben Zweig: Oh, wow. Oh, so you really get the inside scoop.

Sarah O'Connor: Yeah, I'm cool with economists, but I think that their approach to how AI is disrupting the labor market inevitably is very sort of quantitative. And yet what I started to realize through my work and reporting on how people's jobs are really changing is that a lot of the changes are actually in the quality of work at the moment, rather than necessarily the quantity of work. And that it's really, a lot of what's happening is about the way in which people's jobs are changing and how that's changing the way they feel about their work for better or worse, for how much power they have to change their situation for better or worse. And that all of these things I thought were being kind of neglected by this narrative that was either about predictions of what will happen five or 10 years in the future, or just really about kind of sifting through labor market data and looking at kind of AI exposure scores and breaking everyone's job down into kind of a list of tasks and trying to figure out who's exposed. It just misses so much of the really rich detail about how AI is really changing, how we feel about work itself. So that's what I tried to do in the book by kind of going and meeting real people. I basically tried to go and talk to people who are, I think on the kind of front lines of this. So the first to really feel the full force of AI and indeed robotics physical automation as well. So I went inside an Amazon warehouse that is full of robots now. I went down a mine in Sweden where they have sort of self-driving trucks driving around and the miners are kind of sitting in control rooms. I spoke to translators, coders, lawyers. Yeah, to try and get a much richer picture of what's really happening. And the reason that I kind of ended up with that subtitle, The Fight for the Future of Work, was that in the course of doing that, what I realized was that often the narratives that we get told about AI invite us to kind of see it as this thing that is just happening to people. That we're all gonna be sort of swept with this wave of technological change and we will either sink or float. There'll be some winners, there'll be some losers. Let's hope you're in the right end of that. And actually, what does happen when AI enters someone's workplace and starts to change things is that people don't just kind of lie back and say, oh, okay, well I guess my job has been stripped of meaning now, or I guess I am surplus to requirements now, or whatever it might be. Like people respond and react and they either like pivot away from risks, or they try and figure out something new that will make them less exposed, or they try and figure out ways to kind of use this technology to their advantage, but nobody just kind of like lies back and takes it. And so actually, everyone individually is fighting for their own place in the future of work. And sometimes they're doing that individually and sometimes they're doing it collectively. And so I wanted to kind of reflect the fact that we're not just kind of passive recipients of technology, which is being kind of pushed down upon us, although that is often the way it feels that we are protagonists in this story rather than the technology being the sole protagonist.

04:30 Exit, voice, and loyalty

Ben Zweig: Yeah, it's a very optimistic story in a way where, in some ways, of course, like technology can be a wave. We don't, we individually don't have control over what it's good at, what it's bad at, what the economic returns are to this or that. But the fact that we all do have this ability to sort of adapt at the micro level is very encouraging. It kind of reminds me of, do you know this book, "Exit, Voice, and Loyalty" by Albert Hirschman? Yeah, so just this idea that when you're faced with a bad situation, you can leave and there's some power in the choice of exiting something, then that's one choice. You could exercise voice, you can try to change something, or you could just suck it up and take it and just be loyal. And what do you think are, I mean, it sounds like, in your book, if I had to put words in your mouth, it sounds like voice is the hero in the book. That we are engaging in this fight. We are trying to communicate what we want as workers, and that can be a winning strategy. What about exit? Do you think that if jobs transform to a point where they're just like worse, and AI automates the good stuff and leaves you the bad stuff, do you think exit will play a bigger role? Or do you think there's just some market mechanism that will cause people to not want to do meaningless stuff in the same way that there's some wage premium for doing dangerous work?

Sarah O'Connor: Yeah, I mean, talking about exit is interesting right now, because that's really the one thing that is not happening in the labor market right now. What we're seeing is there aren't a lot of people getting fired and there aren't a lot of people quitting their jobs either, and I don't think that's because everyone's super happy with their work and the way things are going, but one thing that happens when economic uncertainty increases and fear increases, and I think right now everyone is really worried about the future, and that's not just about AI, but it is partly. And so you're not getting a lot of exit, you're having a lot of people just sort of, I wrote a column recently calling it the great hunkering down. People are just, people feel like there's a storm coming and they're battening down the hatches, and they're kind of hoping that they won't get swept away. And so it kind of doesn't feel like a great time to be exercising the exit option.

07:00 The great hunkering down and the gig economy

Ben Zweig: And sorry, so your read on that is that is, so this lack of transition, that is mostly due to uncertainty and just being scared of other options, not as much of like a holdover from the great resignation or job matching frictions, like you think it's primarily due to like fear of AI?

Sarah O'Connor: I do, that is my like over-riding sense right now, is that we are in a kind of holding pattern. And it's interesting because one read on the fact that the labor market is actually pretty static right now is maybe everyone's just worried about nothing, like maybe nothing much is actually happening except at the margins. I mean, I wrote this week about how I think a lot of the sort of sharpest effects of like genuine AI disruption and people finding that their jobs are being kind of automated away is happening in the gig economy. So it's not really happening in payrolled, direct employment jobs. And so the economists are kind of poring through the payroll data, looking for every kind of sign that they can glean of people losing jobs or not being hired in such great numbers. But actually if you look at some of those like white collar gig platforms, for example, Upwork, Fiverr, the work is just disappearing off those platforms because that is exactly the kind of job which is, it's already been kind of stripped of the need for institutional knowledge. It's been stripped of the need for interpersonal relationships, it's been turned into something which is like perfect for AI to take over because it's a series of kind of discrete measurable tasks. And a lot of that stuff is like it's copywriting or it's simple bits of coding or it's marketing. And so there are people that are losing work because of AI, like that is real. And I think that we will start to see that more and more with like driving, with delivery, riders, people like that. But that's not the part of the labor market where we have the best amount of data. And it's also like the way that tends to play out is often just that people, there's less work to go around. There aren't gonna be job losses because nobody employs these people. So no one's gonna fire them either. So the way that shows up is like very quietly in more and more people chasing fewer and fewer tasks and their incomes going down or their hours going up. We're starting to hear that anecdotally from like drivers in cities, there was a good study of what happened to taxi drivers in Wuhan when robo taxis were introduced. And basically like their income went down and their hours went up and their stress went up. And so I think a lot of that kind of disruption is beginning to happen, but in the part of the labor market that we have the least good data on. But to come back to your sort of general point about like, exit, I mean, my view really is not that, I don't want people to think that I'm kind of like naively optimistic that like, hey everyone's got agency. So it's all gonna be fine. Like some people have a lot more power than others. And some workplaces facilitate worker voice better than others. Some countries have institutional setups in which it is much easier to raise your voice within the workplace and to have some kind of power or seat at the table when it comes to like negotiating over how technology changes work. But if I was a policymaker, I think right now what I would be trying to do would be to maximize both voice and exit for people, right? Because I don't think you can control exactly how AI gets implemented in a given workplace. It's like, it's so heterogeneous. There are so many different ways that this can go. Each sector is gonna have its own issues, but what you can do is give people as much power as possible to either negotiate over how that goes for them within their workplaces, or if things start to go wrong. And we are seeing there are some chapters in my book about people's jobs kind of losing the bits that they really love rather than the kind of boring mechanical bits that they wanted to have automated away. So yeah, like making it easier for people to walk away. And like from a policymaker's point of view, how do you do that? Well, you need like a robust safety net so that your income isn't gonna fall through the floor the moment you walk out of a job. And the US is not that, neither is the UK where I live, like replacement incomes over here are terrible when you lose your job. Making it easier to access education and training later in life, like all of those things to basically equip people to find better jobs, to walk away from bad jobs, or to improve the nature of work where they are. I think that's what I would be trying to do if I was a policymaker right now.

11:40 Messy jobs and narrow tasks

Ben Zweig: Yeah, it's ironic that there's like better protection for formal work where they do have this ability to kind of reconfigure and adapt and no protection for informal gig work. I think this example of like the gig economy vulnerability is such a useful example because in some way, it's very not surprising. This is an economy where people transact in tasks, not in bundles of tasks. And so when if a task gets automated, there's no ability to reconfigure. There's just, they're gone. Whereas in a job there's people can fight, they can exercise their voice, they can sort of say I can do more of this or I can do less of that, or even the company tries to utilize them and introduces new things for them to do, which makes sense. That is like the adaptation we need. And maybe it's because of the, I mean, the messy jobs that we spoke about, Luis Garicano's work, just before this conversation. And he has this case that when a job is messy or we can think about it like having a diverse set of things to do, then if some part of that gets automated, no worries, you can reconfigure. But if it's super narrow, like drivers or maybe translators, I mean, I don't know, like on the narrower end of the spectrum, that's where people can be more vulnerable. So I guess I wonder if there's a, I don't know if there's a policy for this, but it sounds like maybe having jobs or like work arrangements that are diverse is a good thing. And maybe is there anything that can be done to try to push people toward having jobs that are like complete and diverse rather than stripped into a narrow thing where someone's like a cog in an assembly line?

Sarah O'Connor: I mean, what's interesting, I guess, is that the, one of the kind of directions of travel and it's it's a very long road, which began with the Industrial Revolution really has been in the opposite direction, is to make jobs less messy, it's to disentangle all of the different stuff that people do and to turn them into cogs in a machine, right, to kind of, to make a formal process and split humans into doing like one task over and over and over again, because that is narrowly more efficient and we saw that obviously with manufacturing, Henry Ford's production line, like Frederick Winslow Taylor's scientific management was all about this, it was all about like actually having a bunch of human professionals who have their own kind of craft knowledge and tacit understanding of how to do their job is much less efficient than management knowing everything and just telling workers exactly what to do and which order to do it in, basically.

14:38 Knowledge work vs. the production line

Ben Zweig: Do you think that's really like the trajectory we've been on? I mean, I'm a little nervous to draw that conclusion because we've also at the same, I mean, not at the same time, but later on, has sort of seen the rise of the knowledge worker and like I mean, white collar work at least just thinking about white collar work, I feel like that has exploded in its like, social elements being a connective, being some sort of connective tissue in a complex organization, like the diversity of those jobs seems to have expanded and I wonder if it's really the right read just to think that the long arc of history has led to more sort of winnowing of job responsibilities or like, if there's maybe some divergence, I don't know, I mean, what's your read on like the history in the context of like, there's also been some knowledge workers and that's kind of a new thing.

Sarah O'Connor: No, absolutely. I mean, both of those things have been happening, right? So there has been particularly in manufacturing in the making of physical stuff, there has been a production-line-ization of all kinds of work and you've seen it in jobs that were previously much more messy, like for example, warehouse work, like over time that has become much more structured and ordered and if you go to, in the book I go and look at an Amazon warehouse, like Amazon warehouse jobs 10 years ago were not like massively kind of fulfilling or enriching, but they did involve a mixture of walking around, some talking to people, picking things off shelves, moving them about. Now with the introduction of robots, like the human job has been reduced to standing in one place for 10 hours a day and the robot bringing the shelf to you and now your only job is to look where the light shines on a shelf, pick out the item, which is illuminated and put it in a box, which also illuminates and push the button and then the robot takes that shelf away and the next robot brings the next shelf in, right? So this is an example of a human job becoming sort of like narrower and more repetitive over time. Now obviously simultaneously there has been massive growth in knowledge work, which is the opposite, which is like super messy, which includes loads of different skills that are all bundled up together.

16:52 How AI changed translation work

Sarah O'Connor: One of the sort of trends that I sort of noticed in the book, which I thought was troubling was that, there are some kind of creative professional jobs, which are, if anything, becoming a bit more like production line jobs because of AI. So the example that I write about is translators. So I interviewed this guy who translates the subtitles for TV shows like Netflix shows, Disney shows, from English into his native language, which is Czech. That used to be a job where you were given the English version and you did the subtitles yourself. And actually that is quite a creative job. Like you have to make lots of human judgements about like, how should I translate that joke in a way that is still funny in a completely different language and culture? Or what tense would these people be talking in? Because in English, we just have one way of saying you, but in Czech and like lots of European languages, there's like a formal and an informal version of you. And so as the translator, you need to make that judgment call. How well do these two people know each other? At what point might they flip from the formal to the informal? So it's like, it's quite a creative, fun job. And he used to really love it. What's happened to those jobs is that the agencies that intermediate that work have brought in something called machine translation post-editing. So now those agencies take the English translation, get an AI translation, and then give that version to the translators. And they're told that to kind of check it for accuracy, tidy it up, make it seem a bit more human, but they have to do that at like twice the pace and for half the price. And actually it becomes more intense and more boring. You're not really making as many creative decisions anymore. The solution is already in front of you. So it's harder to kind of think of your own creative solution. And also you're kind of on the clock because you're getting paid less for each task. You feel like you need to sort of rattle through them really to maintain your income. And the agencies have said like, actually people can make more this way because it shouldn't take you very long to post-edit something. And so if you can rattle through loads of these, you might end up with even more money. But the translators that I interviewed said that was not really true because if you genuinely care about quality and accuracy, to kind of go through a translation and improve it and check it for accuracy is actually quite time consuming. But the main thing that they said was that it had been stripped of this, the creative part of it, the kind of the joyful part of it. And it had become something that felt quite mechanical. Yeah. And I think there will be areas where that is what happens. That the ways in which employers redesign jobs in order to make the most of the new technology will be basically to reduce humans to kind of a human in the loop type thing where actually you're kind of plugging in the inadequacies of the machines, but at a machine pace. And that this is fundamentally not a very satisfying job for a human, nor is it something that we're particularly good at checking things at a rapid speed is not something that humans are naturally particularly good at and it's not massively satisfying for people. So that's my kind of negative answer to that. But there is before you say this, I will get ahead of you and say that there are clearly countervailing forces and there are other workplaces where I think AI is kind of pushing work in the opposite direction to become kind of maybe richer and more interesting and more requiring more human interaction. So I think in software, we're starting to see that happen. So obviously sitting down and writing code is now something that most people don't do anymore because in the last, even just in the last year, really, like AI models have become so good at writing code that the job of being a kind of a software developer has changed profoundly. But what I hear from people is that while that is kind of disorientating and it feels a bit existential sometimes, a lot of people are also kind of thrilled and that it's changing the mix of skills that's now required. And actually the people who are kind of thriving in this new world are the ones who have those, like it's brilliant essay that was written maybe a decade ago by a software engineer who wrote about people who are the glue in a kind of software team. The people who basically like look beyond the immediate task of like writing whatever code they need to write to think about the project as a whole, to notice when someone is actually being held up by something that someone else over there would be able to help them with, to kind of be able to communicate with the client about what they actually want rather than just cracking on with the task at hand. And that actually those kind of glue people with those interpersonal skills and those more like project management skills are actually more in demand now and are coping much better with this transition. So that would be an example of where technology might actually be pushing work in a more human direction. So I think you can have both of these things happening at once, depending on all kinds of other factors.

21:43 Unbundling and rebundling jobs

Ben Zweig: Sure, yeah, I mean, just from an economic lens, there's no rule which says that like if technology automates tasks within a job, like there's no saying whether that's gonna be the dehumanizing tasks or the humanizing task. Like it's happened to be the case that AI has automated the sorry, that technology typically in the past has automated the like menial parts of jobs and has led to more fulfilling jobs. But that could very much be we could be in a different paradigm now. I mean, there was that, I don't know if it was a tweet or a saying, but people quote it all the time. I forget who it's from, but someone had said I don't want I wanted AI to do my dishes and laundry so I can write and make art. I don't want AI to write and make art so I can do dishes and laundry. And that is like such a microcosm, like in the home job, if you will. And yeah, like if AI is good at these things that require judgment and humor and humanity, then that just leaves us just being these kind of meat robot reviewers, which is like awful. And so I guess like in some way like, let's say the unbundling of the job is like, that seems okay. Like in this translator example having some machine translation seems very reasonable. I mean, if translation is very good through AI, in some parts of the translation process like that's great, we should take advantage. But if it's like only leaving them to do this kind of like check AI, that is like bad work, but if it leaves them with like, oh, make a funny context aware joke or something, like maybe that's a really good part of the job. So it seems like maybe the problem is just the rebundling rather than the unbundling. Like when you're rebundling a job, you're doing job design, you're crafting something, you're designing something. And I guess that construction of a job kind of happens from the employer, kind of happens from the employee, like there's some give and take sometimes to determine like how that job gets sort of crafted and rebundled. What is your understanding of like how these come to be? Does this TV studio just like hire McKinsey to tell them how jobs should be designed? Do they talk to the translators? What did they actually do? Like how does this get determined?

24:14 Who controls job redesign?

Sarah O'Connor: So I think the difference, like the thing that really makes the difference, obviously you're right, like jobs are being redesigned right in front of our eyes. Okay, and most of the time it's not that someone sat down with a piece of paper and was like, how should I, let's move that task over there to that column and let's do this here. That is not what's happening. It's much more kind of organic and messy and not necessarily thought through. The thing that I think makes the difference in terms of whether it goes well or badly for the workers in question is like, who has the power to make those decisions? Who is actually kind of in control? So for the software guys, they basically have quite a lot of agency most of the time, not all of them, but a lot of the ones that I interviewed, they're like senior developers. They have quite a lot of status in their organizations. They know more about these tools than anyone else. And they're deciding on a day-to-day basis, like, okay, I'm going to get Claude to do this. Actually, oh, I'm not happy with that. I'm going to sit down and rethink it. They're the ones that are really driving it and in control. And so therefore they have a better chance to redesign their job in a way that feels still meaningful and enjoyable to them, albeit it's going to be very different to what it was before. For the translators, they had no control whatsoever. So like some of the translators said to me, like none of them were anti-machine translation per se. What they were against was the fact that they had lost all control over their own working day and their own working conditions, right? There could have been a world in which these tools existed and the freelance translator could have been like, okay, cool, it's up to me how to use these tools. I'm going to use them to rattle through this kind of boilerplate, very simple bit. And then that's good, that'll boost my productivity. And then I can spend a bit longer crafting like the perfect translation of this joke or whatever it might be. I might just kind of rattle through everything quicker and make more money. But actually they didn't have control over it because they are freelancers and they are intermediated by these agencies. There is no voice there, right? There is no kind of collective or individual voice. It's like, this is what it is now and you take it or you leave it. And a lot of people are leaving it. So a lot of the professional translators have said, we're not going to do this post translation work. It's miserable.

Ben Zweig: Exit.

Sarah O'Connor: They exit and they're finding other things to do. And some of the things that they're finding to do is that they're like, okay, I still want to do this but I'm going to cut out the agency model and I'm going to develop personal relationships with clients and I'm going to lean into the fact that what I can offer has things that the AI can't guarantee you like, for example, accountability. So if I'm translating a legal document like I, I'm a human that is looking you in the eye and telling you that this is accurate and if it's not, then you can come and sue me. And that's something that you're not going to get if you just shove it through a machine system which some agency has rented to you. So that's like some of the ways that those guys are adapting. The other thing I was going to say about job redesign is that sometimes jobs right now are being kind of unbundled or redesigned in ways that I can imagine one worker would like and the guy next to them would hate because we all have and value different things about our work. The example that I would give you would be journalism, my job. So there are some like regional newspapers now in the US where they've said to the local journalists, your job now is not to go out, interview people, take notes, come back to the office and write up your articles. Your job now is just to go out and interview people and take the notes. Then you're going to come back to the office and download the notes to the AI and the AI is going to write the article. So your job now has been like cleaved of half of what it used to be, right? There's no writing involved in your job anymore and that this will be more efficient and we'll get more articles out. And hey, it's good for you because you can spend more time out on the beat reporting. Now, personally, I would hate that. I would leave that job if I had the labor market power to do so because I love writing. That's one of the parts of my job that I really love. But I also know journalists who actually, for whom the writing bit is a slog and they actually would love to spend more time having lunch with sources and digging up interesting information and to kind of hand off that bit. And so, it's always really difficult when you're talking about the quality of work and what is good work. You have to always remember that it is so idiosyncratic and we all have our own view of what an ideal job is, which is why I come back to this idea that the only thing you can really do is put as much power as possible into the hands of individual workers and then they can decide for themselves what a good job is. It's not always realistic, but I think that's the kind of the cleanest way is just to sort of maximize people's power to either shape their own jobs or to walk away from jobs that have been reshaped in ways they don't like.

29:08 Do unbundled jobs become new occupations?

Ben Zweig: So yeah, I really wanna, I mean, that's such a fascinating thought experiment. So let's just play it forward. I mean, first, one thought is that it seems very good that there is diversity in how jobs can be constructed. Maybe there's some newspaper that they do this beat reporting and then AI generated texts and maybe Financial Times and then the human writer and like, great, two different markets. It also, I mean, I don't know if you would necessarily draw this conclusion, but to me, that example seems to illustrate a strength in that worker power being at the individual level rather than at the collective level, because if it's collective, then it's trying to represent diverse interests, which may be contradictory. So I don't know, I won't like push you on to disavow, but that is just a thought, but let's just play it forward. So let's say that does happen where you have this like beat reporter and that just like becomes an occupation. And then you have this like full stack journalist that becomes this like reporter and writer combination that is bundled. Are those just like different occupations then? Like, is that how new occupations get created? Like, is that a world with just more diverse career paths and more diverse jobs? Like that sounds kind of nice in an organic kind of way. I don't know, what do you think?

Sarah O'Connor: I mean, maybe personally, I think that the job of being a journalist is harder to unbundle than the people who've decided to unbundle it think. And this is also something that we're seeing a lot, right? Is that nobody knows as much about your job as you. And we all think that we know quite a lot about other people's jobs and how easy other people's jobs are to automate or to change or whatever. So like, I don't think it was the journalists that dreamed up this idea of hiving off the reporting part from the writing part. It was the bosses, right? Who thought this would be more efficient and cheaper? I think that a lot will be lost in terms of quality, quality, accuracy, accountability, when you have someone doing the reporting and then a machine doing the writing. I actually think that will probably lead to a less good quality product. I mean, we'll see. And I think the same thing really about translation, like there have been studies of how the kind of quality of subtitles has declined since the introduction of machine translation. And it's quite stark that there has been a decline in there are more errors, the language is less rich, less interesting, but sometimes these kind of quality losses are hard to detect. And I think translation is a great example of that because if you're relying on a translation, it's because you don't know what the original text said. So you have no way of knowing if you're getting a much reduced kind of flatter, thinner, more meagre version of what the original was. And so there will be errors as well. We kind of experienced losses as consumers without necessarily noticing it.

Ben Zweig: That may be the case for journalists. I could think of other occupations where that sort of splitting would be nice for everybody. Like for example, I'm an adjunct professor at NYU and I've been teaching economics for the last, I don't know, 12 years or something. And like when I first started, I was doing everything. And then like, as you do more, then I could get a TA who could like do the grading and then I could like do this and like sort of split it out and have someone like do the review session. So now I've like reached a point where I just like show up and lecture and like love it. Like I hate the grading. I hate making exams. I hate like proctoring. Like I just but that opens up opportunities for young TAs and like, and maybe it's good for them to have that work. Or maybe that goes to AI, I'm not really sure. But like that's just one case where it leaves the part where I don't think there's, I mean, people talk about there being some value to the bundle of like teaching and grading. I don't personally believe that. I, but like, yeah, like there's some cases where the bundle is just like there are weakly bundled things.

33:34 Toe-treading and the Midjourney study

Sarah O'Connor: One interesting thing I would add there is that the description of the, what's happening with the regional journalists is being driven from the top down, right? Like somebody sat in a board room and figured out that this would make sense to do this. But what we're also starting to see in companies where actually employers have allowed workers to sort of muddle through that here are these new tools, try and use them, see what happens quite a kind of organic bottom up approach. What you're starting to see is lots of like interesting sort of toe treading and people starting to impinge on other people's jobs and being like, well, actually, I might do that bit of marketing myself, or maybe I don't need to go and wait for the in-house legal team to sign this off. Maybe I can check with Claude if it's okay. And we're starting to see quite a lot of that now. And I think that's quite interesting. And one of the things that it, I think makes me reflect on is that there was, I don't think people talk about it as much anymore, but there was initially this argument, I think put forward by Daron Acemoglu and Simon Johnson, that pro worker AI would be to augment workers and anti-worker AI would be to automate. And so we should try and encourage employers to do augmentative stuff and not automation stuff. And I think that the world is just so much more complicated than that. And actually what augments one worker might automate another worker who sits right next to them. Right? Like what benefits me might be a loss for the person next door. So there was, I wrote in the FT just last week about this really interesting kind of ethnographic study that was done by some researchers in the Netherlands where they spent like a year embedded in an advertising team in a media company. And they watched the way in which the introduction of Midjourney, like AI that creates illustrations, how it changed the workflow and the kinds of jobs that they all did. And basically what happened was, so before the AI was introduced, the senior creatives would like come up with some rough ideas and they would like sketch them out in a very rough way. And the client would pick which one they liked. And then there would be a whole load of professionals like graphic designers, photographers, videographers would work together to flesh out what would end up being the kind of the final advert. And that would involve lots of kind of creative to and fro and everyone putting their own input in to come up with what's the kind of final thing. What happened after AI was that the senior creatives at the start of that chain could now use Midjourney to create something at the rough idea stage that looked perfect, that looked beautiful. And that the clients fell in love with and were like, oh, that's great, that's exactly what we want. And from then on, the jobs of everyone else was simply to try and live up to that AI created vision. And so no longer was it a kind of creative to and fro in which everyone had their input. It was the senior creative had basically set the direction right from the start with the help of AI. And then everyone else was just executing on that vision. And when the researchers said that the senior creatives loved this because they felt as if their jobs had got better. Like now I've got away with visualizing my rough ideas because I didn't have those skills before, but everyone else felt as if they'd lost something quite important. And so it just goes to show that often these unbundlings and rebundlings and shifts in the nature of your job, it's not necessarily the kind of the boardroom that's doing it. It might be the guy sitting next to you with a Claude subscription.

37:16 Job quality and job quantity together

Ben Zweig: Yeah, it's a very interesting example because I mean, we started the conversation by talking a little bit about how jobs were kind of narrowing. But you could have some jobs being narrowed because other people's jobs are broadening. Like you mentioned, people just like do, the senior creative person who's doing the prototyping and all that and someone doing marketing stuff. There's this paper that came out of OpenAI, I think a few weeks ago, that was showing that people were doing a lot of work like outside of their core occupation and that's a growing trend. And that seems to suggest some like, at least some cases of like broadening jobs. Like the people that are being augmented presumably are their job is getting more expansive. And then there's people whose jobs are kind of narrowing. And I wonder if there's just like some equilibrium balance where like there's just like only so much coordination capacity that a person has. Like, you can't have someone expand the scope of their job to infinity. That just seems not possible for like a human brain to kind of like coordinate. And then if someone's job is narrow, then like I would think that would like create more incentive for that, for workers to shift between groups, for them to take on more things, do more work, expand their set of capabilities. And I wonder if this really just leads us to like not that much change in the scope of like how diverse someone's job is, but like more of a shift in like just the quantity of workers that are involved. And just like it being it causing some rebalancing in terms of just like net employment on just like the employees that are working on more of the kind of, I mean, I guess going back to like exposure just like where AI is doing more, the spaces where AI is more active versus the space where it's less active. I'm trying to not use the word exposed, but it's hard to avoid. I don't know, yeah, do you think there's some like, I don't know, I mean, I guess I read from your book that like there's a big focus on like job quality and also a focus on like quantity of workers who are working on these domains where AI is entering. And they're both kind of interacting in interesting ways. I wonder if you think they could be like analyzed jointly or like independently. I don't know, I'm just curious to hear your thoughts on all this.

Sarah O'Connor: I think jointly is definitely the right way. And we were just talking about this advertising firm and I've been talking to quite a lot of media companies recently and I think that as an example of where you're seeing both of those things happen. So you are seeing some people who have like the enthusiasm in the appetite and the kind of the creative skills are suddenly kind of ultra empowered to be able to do much more than they were before. And without necessarily needing 10 people in the video production team and 10 people in the sound design team to kind of execute their vision. And therefore you are seeing in areas like that, you are seeing job losses, right? A lot of those companies are retrenching. And also there are, that opens up the possibility that entire companies business models will be disrupted and there will be these kind of AI-native firms will come along that kind of offer to do something similar, but from scratch with a much smaller number of people, we're starting to see a bit of that too. Companies like Meta are saying, well, actually, we can just use AI to make adverts and you don't need to hire a creative firm whatsoever. So yeah, I think that there might well be sectors where you're seeing a dwindling number of jobs which are themselves becoming richer and more interesting. And whether that's a good or a bad thing depends very much on which side of that divide you end up on. But in general, what I would say is like what is frustrating to me and I think what a lot of people feel sort of perplexed about when you take a step back is there was never a problem with the fact that we had lots of people doing creative work that they quite enjoyed. This was not a problem that actually needed a solution. We could quite well have used and welcomed automation which could do like very dirty, physical, complicated tasks, areas where there are like deep, deep labor shortages because there aren't enough people who are willing to do the job for the price on offer. But instead, actually, a lot of the jobs that have turned out to be quite vulnerable to AI disruption are jobs that people quite liked and were quite good at and so that doesn't feel like a great outcome. And more broadly, I think it's just a reminder that just taking the technology as given and allowing it to sort of rip through the labor market based solely on what it is and isn't good at and thinking that the only thing humans can do is like rearrange themselves around the tools, I think is not the best way to think about this. We'd be much better off thinking very hard about what do humans want to do, what are we good at, what benefits us and then figuring out how to mold the tools to serve us rather than the other way around.

42:43 Why AI sentiment is so negative

Ben Zweig: Sounds like a very coherent theory of pro-worker AI. I'm, this is the pro-worker AI I'll sign on to. Okay, so do you think this is, like this idea that AI is kind of hoarding the engaging parts of the job and leaving us with the shitty stuff, is this like a, do you think this is a determinant to like why there is like negative sentiment around AI? It's not popular. And so I guess it's a two part question. One is like, is this job quality component a big explanatory factor in negative AI sentiment? Another is that do you think there's a distinction between this like these vibes in the US and Europe? I'm counting UK as Europe for this example.

Sarah O'Connor: Yay. Yes and yes, I mean, I think there are lots of reasons that sentiment is pretty negative towards AI in a lot of countries. The there's like the fact that the people who are making it keep saying that they think it might kill us all, that's doesn't make you feel particularly positive towards it. There's the fact that they keep saying that they're gonna put millions upon millions of people out of work, which hasn't happened yet, but it's not an appealing prospect to most people, right? But then I think to your point, yes, I think the ways in which it's affecting the quality of people's jobs probably is a factor for some people. I mean, as we've just discussed for some people, I think it's enriching their work, right? So this is not a kind of blanket statement, but I do meet a lot of people who are in jobs where actually either they're being kind of pressurized to use AI tools by their employers in ways that they don't think are particularly helpful, or that the tools are actually not quite good enough for the job at hand, and so it's kind of frustrating that they keep being told they have to use them. And it feels like in some ways it's making their job more complicated or harder because they're having to constantly like mop up after the mistakes that the AI has made. Also this toe-treading thing a lot of people, when I wrote that article last week about that study and about the kind of the broader phenomenon of people kind of using AI to encroach on other people's jobs, it had such a massive response from readers, so many people wrote to me with examples of how this is going in their work. And I think the thing that people find most kind of annoying about that is a kind of like, I think it feels like a sort of professional disrespect that there'll be someone who has built up a whole career in writing software or being a legal expert. And then one of their colleagues is like, well, actually I just asked ChatGPT and it said this, so you're wrong and that is just kind of enraging, right? And so I think there's quite a lot of that going on as well. So there's, and then for writers and musicians and people in the creative arts, they're worried that they're gonna lose the thing that they love most about their work. So I do think that fears and realities about the kind of change in quality of work are part of it. And then your other question about does it vary by country? I think, yes, I mean it's no surprise to me that people in America are particularly miserable and fearful about AI because it's a country that has like zero safety net. It's not a good place to exit your job if it's going bad. You're not gonna get much support from the state. And they, and there aren't great voice mechanisms in most American workplaces either, like trade unions are very weak. Employers can be quite sort of dictatorial. And so in my book, I spent some time in Sweden, as I mentioned, sentiment towards AI is much more positive in Sweden and other Scandinavian countries. Also adoption is really high. And I think a lot of that is down to the fact that safety nets are very strong. The whole Swedish labor market, the kind of theory is, you protect the work and not the job. So the Swedish government and trade unions and employers, they're not trying to kind of preserve jobs at any cost. They kind of welcome creative destruction, but they make sure that workers are offered like very good income insurance if they do lose their jobs and also a lot of support to get a new one or to get retraining. And so that kind of lowers the fear factor a bit and makes exit a bit more of an appealing option. But also there is much more voice in some of those countries, right? So in Sweden, there are trade union members on the boards. They have something called co-determination. So when an employer wants to introduce a new technology that's gonna change the nature of jobs, they have to sit down and negotiate over it first. And the employers still have the final say, but it means that some of those kind of accidental sometimes decisions that get made where a new technology might be brought in with the best of intentions, but actually leads to quite sub-optimal outcomes for workers. Those can be kind of preempted before they happen. And I saw some really interesting examples of that when I was in Sweden.

Ben Zweig: Yeah, yeah, that's a great example. I mean, it is hard to kind of think about why AI, I mean, as someone who's based in America, I have a hard time thinking about non-US AI sentiment because so much of sentiment in the US is driven by like, who are these AI labs that are like involved with our government and like trying to set policies and like asking for a lot of trust. And it seems like the fact that the frontier labs are being built in the US requires a lot of like, faith in Silicon Valley institutions, which there's a lot of skepticism of I feel like it's very understandable for people to say, oh, these are the guys that have been trying to convince us of NFTs a couple of years ago. And like, that was all a bunch of nonsense. So like it's, but in I guess like in the whole world we're thinking about like, employment in terms of like, what's happening. And I think your book really does a great job in describing like, what is actually happening to jobs? And then there's of course, the fear of like, what might happen? And so there's another like confusion there of like are we driven by fear of like this like, these doom scenarios, or are we driven by like, what do we see that's irritating and annoying? And also how much of that is understandable? Like if AI is a great lawyer and you don't need your lawyer, like that's annoying to the lawyer, but like I don't know, like that seems like the reaction is just like protectionism or not wanting it to happen in your backyard, I guess.

49:15 Giving workers more agency

Ben Zweig: So I guess like, final question for you, like what is your favorite policy recommendation? I guess two part question for like governments, but also for employers, like if you're a big company, like do you recommend policies to them?

Sarah O'Connor: I think it would be kind of the same. And I sort of said this at the start, but I think that really the reason that everyone is feeling so fearful right now comes down to the fact that everyone feels as if this is out of their control, that this is like something that is happening to them. The labs don't even seem to be able to control these things. They're breaking out of their sandboxes. They're doing crazy things. You can't control what your boss is gonna decide to do. And so I think that everyone is really feeling this kind of massive threat to their agency and their ability to kind of control events. And so as a policymaker or as an employer, I would be trying to push back on that and to make people give people a bit, not make people feel that they've got control, but actually give them a bit more agency and control in all of this, right? And there are ways you can do that. Policymakers can make sure that it is easier for people to walk away from jobs. Employers can think a lot more about talking to their workers about what they're doing before they do it and listening to them about how it's going. It's amazing to me. I've obviously been doing a lot of sort of book talks and stuff because this book has come out, which has been great. And I've met so many people who come up to me afterwards and tell me the kind of stories of their workplaces and an awful lot of employers bosses, people in the C-suite who desperately want this to go well for their workers but don't know how. One guy came up to me and said that he'd hired a lot fewer graduates this year because he didn't need as many. So he already felt a bit bad about that, but didn't see what else he could do given the competitive pressures in his industry. But also that he was worried that those graduates might be feeling a bit lonely because they weren't kind of getting to interact as much anymore with seniors because they could just ask AI to kind of solve their problems and all of that sort of thing. And did I think that this could be a problem? And if so, what should he do about it? And I said, well, why don't you ask them? Why don't you ask them if they feel lonely and if they do what they would like you to do about it rather than sort of decide that they're lonely and make them go and play football with you once a week or whatever it is. And I think just talking to people, talking to one another and asking people, it doesn't have to be in a formal, like collective bargaining situation, but just in general. And the same goes for all of this kind of toe-treading stuff. Like in that example, I talked about in the Dutch advertising team, in the end, they all sat down and they talked about how this workflow had changed and how some people had felt that they'd lost something. And they figured out a new way of doing things together. They kind of like co-designed a whole new structure whereby everyone would actually be in those initial pitch meetings and they would all kind of sign off on that initial vision so that they still had a way of having some kind of creative input. So often there are ways of navigating this, but it requires us all to listen to each other and have some respect for each other's professional expertise.

Ben Zweig: Yeah, okay. So for policymakers and companies, it's like agency, agency, agency, just like worker voice that is like the solution. And also for, there's like a hidden message for economists. Qualitative research, go out there, talk to people get to know what things are like on the ground.

Sarah O'Connor: Yeah, numbers can't tell you everything, guys.

Ben Zweig: Yeah, yeah. I mean if they can convert it to numbers won't complain. Yeah, okay. I mean, really, thank you so much for being part of this show. I gotta recommend your book, "We Are Not Machines, The Fight For The Future Of Work." It's really an excellent read. And for the record, books written by journalists are the best written books. You can't have much better than the nonfiction books written by CEOs. I really think it was so well written, it was so great. And I gotta recommend your writing in the Financial Times, The AI Shift newsletter, really great stuff. Thank you so much.

Sarah O'Connor: Thank you, Ben. I enjoyed it. Cheers.

Ben Zweig: The Economics of Work is brought to you by Revelio Labs, workforce data for research and benchmarking. If you enjoyed, please remember to rate, review and subscribe. It'll help us reach other curious listeners. I'm your host, Ben Zweig. Our producer is Cole Wagner. Thanks for listening.