The AI Jobs apocalypse is coming | Amit Katwala

If you look at civilizations that have collapsed in the past, right, the evidence suggests that they never really collapse in like one day of horror, right? It's usually like over a drawn-out period of time, like the Mayan Empire collapsed over a course of a couple of centuries. Same with the Roman Empire. It didn't just go bang on one day. And I think that's probably what we'll see is like this slow attrition of things getting worse and worse and worse and AI sort of helping to accumulate wealth at the very, very, very top and life just getting more and more miserable for people until the world explodes.

Yes. Hello and welcome to the tech report. I'm Natasha Banal and joining me today is tech journalist and author Amit Katwala. Welcome.

Hi Natasha.

Hi. We're going to start, um, the episode today by talking about a new report that was released about youth unemployment today, um, that shows among other things the huge impact that AI could have. Amit, what's going on?

Yeah. So, basically, the UK government has commissioned Alan Milburn, who was a former Labour cabinet minister, to basically figure out what's going on with youth unemployment. And he's done this big report which he's releasing in stages. And the first part of the report says that the UK faces like a 125 billion a year financial hit from young people not being in jobs. It also showed that the number of people aged 16 to 24 who are not in education, employment, or training is now more than a million, which is the first time that's happened since 2013. That seems really bad. Um, not, not a great situation for, for the people who have been labeled the lost generation, right? Um, how is AI playing a part, or how could AI be playing a part in all of this?

Yeah, so AI has been mentioned as one of the things that's perhaps exacerbating this. Um, you know, people say that it's taking entry-level jobs away that maybe graduates used to go into. It's also making it harder for people to get jobs because, you know, people end up being funneled into this sort of AI-powered HR system where AI screens their applications and then if they do get through the first round, and they talk to an AI, you know, bot in an interview, which then looks for certain keywords in what they say. And so it's really, really hard for them to like get through to a human. I think that's a function of two things. One, companies being encouraged to use AI and then secondly, that, you know, it's a buyer's market, right? Like there's not that many jobs available, uh, partly because of AI, partly because of other economic reasons. And therefore, companies feel that they can, you know, pull this kind of stuff where maybe, you know, 20 years ago, they wouldn't have, they would have felt the need to actually actively try and attract candidates, whereas now it's more like there are candidates out there and they need to wow that pool down to someone that they can give the job to, and that person will be so, so grateful to have a job offer that they'll just put up with whatever, you know, indignity, you know, whatever they have to face during the recruitment process.

Yeah, it does feel like AI screenings are becoming almost like a de facto part of the recruitment process. And I know a lot of people are really put off by it. You know, I've seen a lot of messages on social media from people saying, I really don't think it's okay for a brand to think I want to talk to an AI for 20 minutes, um, and then not get much feedback at all about my performance and think that I want to work there. Why would I want to work there if that's my first experience of, of the business? So there's definitely a back and forth, but you're right that that it feels like the scales are skewed in favor of, of employers at the moment, just because there's more, uh, there's more people looking for jobs, um, in the market. I guess with, with this situation though, there's, there's also talk about how AI could impact, um, jobs and the kind of things that graduates or people in junior roles could do in the future, right?

Yeah, so I think like, you know, one of the things that people always say about AI is that, you know, it does the work of a, a grad, you know, a graduate, or does the work of three graduates or whatever. And you're right, like a lot of the things that people of our generation maybe would have done when they first started a new job, particularly a white-collar job, you know, making slideshows, or, you know, analyzing data, you know, writing memos, things like that. These are things that AI can do like to a reasonably good standard. So I think those are a lot of the jobs that are being lost at the moment. I do, I think that there is a degree of short-sightedness, particularly for these like white-collar companies, consultancy companies, and, and things of that space, in that if those entry-level jobs don't exist, then there's no funnel to the kind of higher-level executive roles. So I think that's a, a problem that companies that have rushed to adopt AI in these entry-level roles are going to face soon. I guess the other thing is that, you know, AI can maybe amplify the work of like one graduate. And this is again, what a lot of the AI proponents will say is that, you know, it can, you can have one graduate doing the work of five graduates, 'cause they can use AI to upskill or to, to massively increase their productivity.

Yeah, it's like turbocharging their work, therefore they only need one person to do what maybe four would have done in the past. I guess. Is there any way to remedy that pipeline, that that you, that you can think of, that funnel effect, if we're missing one piece of the puzzle? Obviously, those people aren't ascending into the middle management level jobs and, and so on and so forth. Is there any way to fix that aside from just removing AI?

Well, either that or remove middle managers, I guess, would be my, uh, 'cause I think the thing is, like everyone's like, "Oh, well, the graduates are are screwed, but we'll be fine 'cause we're, you know, middle-level executives and what we do is really valuable and no one could ever replace us." But actually, you see people talking about these like tiny teams and one-person companies. And actually, like, if you can replace a graduate with an AI, then why not replace a, you know, mid-level executive? Why not replace your CFO? Why not replace your CEO, right? Like there's no real, like, if, if you, if you buy the premise that AI is, you know, genuinely like as intelligent as humans, which, you know, you can believe or not believe depending on your particular slant on it, uh, then, then, but if you accept that premise, then there's no reason why you should have any, any human employees, really, or just one at the top who generates the ideas and then has a team of agents go out and do, do that for him or where. Yeah, I mean, I think like this sort of, I'm kind of fascinated by this notion of the apocalypse. And I think, you know, in movies and in books, uh, the apocalypse is always this like singular moment where everything goes wrong in one thing and the world explodes. But actually, I think the AI apocalypse, most apocalypses are a much more gradual process. Like if you look at, um, and this is maybe a slight tangent, and I apologize for bringing up, you know, doomsday so early into the recording, but if you look at civilizations that have collapsed in the past, right? The evidence suggests that they never really collapsed in like one day of horror, right? It's usually like over a drawn-out period of time, like the Mayan Empire collapsed over a course of a couple of centuries. Same with the Roman Empire. It didn't just go bang on one day. And I think that's probably what we'll see is like this slow attrition of things getting worse and worse and worse and AI sort of helping to accumulate wealth at the very, very, very top and life just getting more and more miserable for people until the world explodes.

That does seem very doomsday. I mean, obviously, not all things are equal. In this report, they were talking about, um, the kind of jobs that might be more exposed to AI and might be more vulnerable to it. And it, it feels like those are more in the kind of knowledge and office-based work rather than manual and service occupations, which makes sense, right? It's easier to replace someone who deals with Google Docs most of the day versus someone who is like a plumber, um, or, or who does transport or things like that. But I guess, you know, here they're saying it doesn't mean that there's no risk to young people and that the narrowing on one side could have an effect on the other. I guess is this more of the same of actually rather than learning coding, young people should, uh, train to be plumbers, basically, hundreds of thousands of plumbers?

Well, the thing is though, like, yeah, you can train to be a plumber, but like, who's going to be able to afford a plumber when no one has a job, right? What's the right? Like there's no real, no one's really thought about that. Or if they have, they've sort of, you know, waved their hand at something like universal basic income and said, "Oh, well, that will, that will happen at some point." That I think there will need to be an economic transition at some point. And again, if you believe the hype, like if you believe what the AI companies are saying, and if you believe that they can take human jobs and they will, you know, take 80% of human jobs or whatever by a certain date, then, then we need to come up with a system of allocating that wealth through society, otherwise there isn't going to be an economy, right? Like there's, and it comes to like, I think the same thing about this as I think about like Google AI Overviews, right? There's this sort of machine that they've built that like hoovers up content and regurgitates it and then spits it out and completely like cuts the people that actually create the content out of the loop. But actually, well, if you, you know, run that through to its logical endpoint, like what's going to happen when there isn't any more news websites left to feed the AI? How are the AI Overviews going to know, you know, when the Queen's birthday is or whatever nonsense people are Googling if no one's writing the article? You know what I mean? Like, what? So, and it's the same thing here, right? If AI hoovers up all the wealth and all the productivity and all the growth in the economy, then there isn't going to be an economy left because ultimately, like, and that's going to affect the jobs market, too, right? Yes, you can always go and get a job as a plumber or work as a, you know, retail assistant or whatever, but if no one's got any money, then the shops are going to close. Um, and, you know, like Elon Musk isn't going to be shopping at like M&S, right? So like if you're an M&S checkout assistant, you know, you're a bit, you're a bit screwed regardless. So I think that's a, Yeah.

Yeah, you are. I guess the hope is that it doesn't, it happens so slowly, as you say, that that maybe that that we follow in the steps of the Mayans and and take 200 years and therefore all of us will be dead by the time that happens and we don't need to worry. Fingers crossed.

Yeah, that's the dream.

That's terrible. But, um, but yeah, I mean, this, this narrowing, so it's not just that jobs, uh, could disappear. It's that, you know, routes into work become narrower still until they're basically choked out of existence. I mean, look, we're quite lucky in the UK, uh, because we have a welfare system. And I guess, you know, this is all to say this is going to be terrible for the welfare system. We're going to have a lot more people on benefits, a lot more people that find it harder and harder to get into, into work. But I guess, you know, elsewhere in the world, and I suppose we could look to the US as a good example of, of this, the, the effects of, of layoffs and the effects of AI on the workplace have been even more pronounced, right? I mean, we, we've seen a lot of, of, of the effect of that. And I guess what could we, what could we use to, I suppose, tell what might happen next based on what's going on over there?

Well, I think there is, I think that is to, to a certain extent, I think that is a little bit of a fallacy because like, if you look at the report that was published in the UK about these million people not in employment, uh, educational training, like these are not, uh, this isn't distributed evenly throughout society or even geographically, right? There's a much higher rate of this in, you know, the north of England than there is in London, for instance. And I think if you look at the US, you'll probably see a similar effect where like, yes, there are lots of people on benefits and lots of people not in work, but those, those people are not people that necessarily would have been, uh, in white-collar jobs in Silicon Valley anyway, right? Like, in the sense, like I, I think we should be careful not to conflate like youth unemployment, uh, which is a structural issue, with white-collar unemployment because code came in and now you can't get a job in coding. Right? That's a very, very narrow sector and a very, very narrow approach to this problem. I think what you alluded to of the the narrowing of access routes into employment is really interesting. And I actually am of the opinion now that there's almost no point applying for a job if you don't already know someone there. Like literally, if you are on a company's website or heaven forbid clicking easy apply on LinkedIn, you might as well not bother because you are going to go through into an AI screening process and you almost need a connection or a personal connection or someone to kind of like shortcut and jump your way up the queue, which has been the case for a while. That's not like an AI thing. That's always been kind of the way it works, particularly in industries like ours. But equally, like that's terrible for social mobility and diversity and, you know, just for, for, yeah, all these things that, you know, we, we claim to want in our society and in our industry.

It's horrendous. And also, there's an erosion of that too, right? The fewer people in your cohort that get stable jobs that are able to grow through the ranks mean fewer people that you can then turn to to say, "Hey, can you help me out? I need a job. I've been laid off or I can't get into the job market." Um, you know, a lot, a lot of people are in that situation now, and it ever has been thus. But it does feel that that could get a lot worse, right? And you're in a situation where you go, I don't know anyone. I, my family doesn't know anyone. My friends don't know anyone because they, they are all in the same boat as me. And that's, that's, I suppose, that's that tipping point that we were talking about of like where it just ends up being unsustainable and no one can go to the shops anymore. No one can afford anything.

Yeah, I mean, what you end up with is an underclass, right? Like, and you end up with a section of society that has extreme wealth and you end up with an underclass. And you, I mean, you see that in some countries already. Like you see it in like the UAE already where it's sort of enshrined into the, into the way that their their laws work, um, you know, with migrant laborers, you don't have citizenship besides. You see it in the US as well. Like, I, I just got back from, uh, the US and, you know, there's like a, a real divide. Like the, like Americans have so much wealth, but it's just so unevenly distributed. And I think AI will exacerbate that. And, and, and I think the difference is that rather than being like, rather than there being like a geographical split of like the wealth being concentrated in in areas that historically had white-collar jobs like New York and Silicon Valley, like that, that split will be much more apparent because people that used to be in those jobs, those jobs will no longer exist. Um, so therefore, this, the divide will get more and more. And I think the people that used to have what they considered to be, you know, safe middle-class or upper-middle-class jobs will be, uh, the ones who are feeling it the hardest. Whereas I think if people, people that have always been in sort of like, you know, blue-collar or service industry jobs, maybe for them, life will continue pretty much as it, as it sort of has for the last, you know, several decades.

Yeah, it's like a longer tail, right? But I mean, I guess it's important. I know we're talking about the youth side of things here today and, you know, uh, the generation AI that's coming into the workplace, but, um, you know, all these, all these things that we're talking about, they chip away at all levels, right? We're not just talking about, you know, this affecting graduate jobs, but it's the, the erosion of, of jobs across the board, right? You're also going to get management jobs. I know, you know, in, in the past in this program, I've also spoken about, um, you know, older, older people, um, older people, older workers, um, so the people who are more into the sort of, you know, 20, 30 year mark of having been in employment, that also seem to be targeted by, you know, AI layoffs, AI-related layoffs, AI-excused layoffs, um, hugely, hugely problematic for, for everybody because I imagine, you know, for, for other sectors of society, there's, there's just the same vulnerability, right? Whether you're a young parent that is just, you know, that has small children who don't, it, it's not easy to find a job. Whether you, you're disabled or you're sick, you've had to go on leave, or if you're reaching, uh, the end of your career, and it, it becomes harder and harder to find another job because of discrimination that is rampant across recruitment, right?

Yeah, I mean, I think that, look, I think we have a road map for what this sort of extreme wealth inequality looks like for your average person, right? Like go to, you know, before the Industrial Revolution, or, you know, Renaissance Italy, or whatever. There'll be a, you know, a very, very small number of very, very wealthy people, and then everyone else will basically sort of live at their leisure. I mean, your best bet is to get a wealthy benefactor who will sort of, you know, pay you to paint them in repose, or whatever. Uh, you know what I mean? Or like live off the land, right? Yeah. I think that there will be a correction. And I think ultimately, like one thing that's happened is that like jobs and the jobs that people do, I mean, this is like the kind of the jobs argument that book, like the jobs that people do have become so uncoupled from the things that people need, if that makes sense? Like in terms of like, what, what do the jobs that, like, you know, and include us in this as well, like, what do the jobs that we do actually do to, like, contribute to the things that people need to live in terms of like growing food, or providing shelter, or providing power, or, you know, that kind of thing? And so this whole tranche of jobs now that have been like, AI is great at writing PowerPoint presentations, writing memos, you know, writing lists, blah, blah, blah, like all this stuff, but none of that stuff is actually productive in and of itself, right? It's good at ticking the boxes that we've created to like that we've ascribed value to in, you know, society as it was from 1950 to 2008, right? And it's good at that and doing those jobs, but are those jobs inherently valuable? Like we, we sort of decided that, you know, as a society, we decided that those jobs deserved, you know, hundreds of thousands of dollars a year, but is that a given? I'm not sure it is.

No, it's not. And I think it's, it's really interesting what people value and don't value. And I, I know, you know, I wanted to talk about OpenAI's Sam Altman, who this week sort of said that he actually was, you know, wrong maybe about the AI jobs apocalypse and that in fact, he maybe didn't think that there were going to be so many social and economic, um, impacts. And obviously, you know, this is someone who has an agenda. This is only speaking from his perspective, but he did cite something interesting, which is like his own perspective of like, oh, I tried to offload things that he clearly did not think were valuable enough for him to do to an AI, which is like responding to Slack messages, responding to emails, that sort of thing. And then he realized that it wasn't actually that nice. And he sort of reigned it back and decided to do it himself. And I guess that what you're talking about, that value is sometimes not apparent until you lose it and you realize that actually AI is not that good at doing it in the same way. Right.

Well, I think the value, like I think people mistake the output for the process, right? And I think that like the whole point of doing the work, and this this varies depending on context, but like the, like the point of coming up with a business plan is not the business plan. It's the thinking that you did to get to the business plan, right? It's, it's thinking deeply about the business and the way it should be structured and, you know, the problems that you might face and how to solve those problems. If you just type "business plan" into ChatGPT or Gemini and it comes out with a business plan, great. But like, what's, what you know, you haven't actually solved any of the problems that you need to solve. And I think that's, that's the issue. I think in a lot of jobs, like the output gets sort of mistaken for the actual work, which is in the process. And if you're not doing the process of thinking, then what are you actually doing? What value are you actually adding? Yeah, on Sam, I think Sam Altman's interesting because obviously he spent the last, you know, two, three years saying that AI is going to, you know, every, every time Sam Altman or someone like that says that, you know, AI is going to take all our jobs, like, you know, the value of his company goes up. So, I think probably like, I don't know, like maybe he's had a, you know, sort of a revelation and realized that, oh, maybe my company's peaked in value and now my main risk is someone, you know, getting angry with me and like an angry mob turning up on my house, so maybe I should just like dial it back a bit. Maybe he's cashed out. Who knows? Um, but like, yeah, so I do take those statements from, from the creators of AI companies with a large pinch of salt, I would say.

It is really interesting that it sort of took a personal experience. I'm not sure whether that was a thing that was the catalyst of him, but going like, "Oh, actually, I, I, I've changed my mind." But, but it is funny how it's like, unless something happens to them, often, you know, these tech leaders don't really come out and say, "Oh, I tried this and I didn't like it." And actually, you know, it's sort of like, I don't know, it's like a snake oil salesman suddenly having a revelation going, "Hang on a minute, this snake oil doesn't do anything." You know what I mean?

Yeah. Yeah. Guys, I've done some research and it turns out homeopathy is like, you know, not real. Like, who knew? Sorry for misleading you for like 18 months and driving the price of chips and, you know, drinking all your water. Whoops, my bad.

Yeah, I think you have to be very, very, very careful when you, when you pass statements like that from, from people like Altman, personally.

Yeah. And I suppose it's interesting because in, in the case of his company, and obviously there's, there's huge other things going on. there's an IPO looming this year, all that sort of stuff. Um, but, you know, you, you do kind of see where he's coming from trying to attract those corporate clients. He needs, like, ultimately for AI businesses, it's not a good idea to think into that future that you and I were just talking about where nobody can afford anything because they won't be able to afford AI, right? In the end, it all collapses on itself and, as you say, completely destroys itself. So, it's, it's, I suppose the, the rhetoric of it's going to take loads of jobs. There's a, there's a benefit of saying that's not true. Actually, it's going to all be, you know, wonderful and glorious and, you know, there'll be new jobs created, which, you know, there's some evidence that that might happen. Um, but it doesn't mean that that's good news for actual people, right? So,

Well, the people he's talking to when he says, "Oh, don't worry, AI is not going to take everyone's jobs." Is the people that buy his products, right? It's the, it's the procurement manager or the executive who decides which AI model to to spend their budget on. It's going to take everyone's jobs except yours, my client. Like everyone else but you, valued client, you'll be safe. Like, come on. Um, I, I don't know. Like, I think, and I think the thing is with AI, like it's so massively subsidized at the moment, right? And as you know, I'm sure Ed Zitron has mentioned on this podcast many, many times, like these, these companies are operating at a massive loss. And like the model is, yeah, like hook people in with like, you know, free cloud, uh, tokens and then jack the price up massively. Same with Gemini, same with, um, ChatGPT. And like, ultimately, like, yes, for some companies that will be worth it. And, um, but for like individual users, it probably won't be. And I think their hope is that like they'll, they'll hook us in and our skills will have atrophied so much by the time that they jack the price up that we'll feel like we've got no choice but to subscribe. The same thing happened with Uber, right? Like, you know, it, it, it becomes part of like your muscle memory, like to get an Uber. And actually, you don't, if you look at the prices now, uh, compared to when Uber launched, like, and, and how comparably, when Uber first launched in London, it was so much cheaper than black taxis, whereas now they're not, they're still cheaper, but not by that much, right? And it's exactly that sort of thing. And I think people will come to the realization that actually maybe it's better to just do stuff themselves. Maybe it's even quicker sometimes to do themselves. Maybe it's even more accurate because I think ultimately, and, and, you know, for some industries, that will be true, and for some industries, it won't. But I think this sort of mad rush to adopt AI across the board is, is, um, slightly short-sighted in my view.

Yeah, it's interesting that muscle memory. The thought of like, there's, I don't know, you know, with, with our generation of, of people, you know, I don't remember phone numbers. We don't necessarily know how to use maps as well any, as we, as we would have done when we were younger. Um, you know, the thought of, you know, someone using a, a payphone on the street, you know, all these sorts of things are sort of very minor things, but you get used to using your Google Maps, you get used to your phone telling you what to do, you get used to, you know, paying a certain way, living a certain way, using certain tools. And how did we live without them? And in the end, you just sort of forget. Like, if you ask me to, you know, make a loaf of bread from scratch, I mean, unless you were one of those aficionados in the pandemic, you probably don't know how to do it. And it's the same with, with anything else, right? You know, ask me to, you know, go in back in time and use a Rolodex to do my job, and it would be really hard. And so, um, you know, people don't answer the phone anymore. Um, so, so it's that kind of stuff where absolutely, I can see, I can see what you're talking about, where it's like, once, once you reach a point that there's just no way back because we'll be so dependent on that tech that we won't know what to do without it. Right.

Uh, just, just to note that of all the people I know, you're the person that's most likely to own a Rolodex, I would say. I mean that as a compliment.

Thanks, Amit. I do have a file of fax. I don't have a

There you go. See, exactly. I was, Yeah. Bang on. You look so offended. I was actually

I'm just, I'm just, Yeah. I just think it's, it's terrible when someone clocks me straight away and you've always been able to.

But, um, yeah. So, so with, with this scenario there, right? So we've got obviously a lot of stuff going on, um, in the background that might have informed this, like will they, won't they of the AI job apocalypse. It does fly in the face of Christopher Ola, who this week was talking with the, with the, with the Pope, saying, you know, AI is going to have a huge impact on all human life, all that sort of stuff, and positioning himself kind of on the opposite side of the spectrum that Sam Altman's put himself in now, um, not very optimistic, saying that there needs to be guardrails, there needs to be regulation. Where, where do you land in all of this? What, what are your thoughts on what, because obviously all these things can't be true at once?

So my thoughts on Anthropic in particular are that they're just at a different point in the, so, you know, the got in a hype cycle, right? So basically, there's like an AI hype cycle. So it starts with found an AI company, and then get disillusioned by a company and think it's too dangerous. Found AI company, get disillusioned. The second AI company say, "That's too dangerous." Found a new AI company focused on safety. And they just kind of go through the levels. And every time they like, they get spooked, they just start a new company. And then they're going to be like, "This company's focused on safety." Until the money comes in and they're like, "This company is no longer focused on safety. We now make weapons." And then that, like, a bunch of people leave and they start their own company. So I just think that like, yeah, Anthropic is a little bit behind the curve, that a little bit behind on the journey that OpenAI is on of like, you know, they're just like, at the moment, it suits them and that, you know, their, um, shareholders or whatever, to to say that AI is an existential threat. And then in six months, once, you know, they've realized that they've got as much juice out of that orange as they can, then they'll say something else. I don't know. I think that AI could upend absolutely everything, but I don't think that there's any evidence that that is that LLMs are going to be the, you know, the method by which that happens. Again, we, we, we talk about AI as this sort of like, well, at least AI is this kind of broad thing, but when we talk about AI generally, we're talking about LLMs, right? And those are the products that these, these companies are best known for. And I think like, yes, it may upend everything within a narrow segment of society. And obviously, some of the new things that Claude has has come up with are, you know, genuinely really impressive. But, you know, like, what's the stat? Like, only like a really small percentage of the world have ever even used an AI, right? I think we forget that actually, like, most people in most parts of the world are not using AI like day-to-day in the same way that, yeah, like, you know, white-collar workers in English-speaking parts of the world are. So I think that it might upend absolutely everything, but only if your definition of absolutely everything is like a very narrow tranche of like Silicon Valley, London, New York, the Vatican.

Yeah, it's very hard to not make it seem like navel-gazing. Obviously, although LLMs are the, the best known thing, I suppose, or more prolific thing, I suppose, within the world of AI, it's still quite a small segment of, of what, um, AI could represent, right? And I think, you know, that people forget about, you know, other technologies that are being developed that are perhaps more disruptive to other industries as well that are just not as prevalent yet, right?

Yeah. Yeah. Exactly. I think people because AI speaks the same language as, I think people describe it with much more intelligence than it actually necessarily has, right? Like I think ultimately these are probabilistic, uh, word prediction models. And I think there's a, there's a limit to what they can do. Uh, there's a limit to how far beyond human intelligence they can go. And I think that like what we've seen is that they serve a very good averaging function. And that like, if you need something done to the quality that like an average person could do, then yeah, sure, like if I need like a Bolognese recipe, then they're really good at like looking at every single Bolognese recipe that's ever been written and bringing me sort of like the median Bolognese recipe. Great. But like, if you actually want to push the boundary and like innovate and like do the things that companies say they want to do and say that, you know, say that they exist for, then I think you kind of need more than that, right? You don't need the average Bolognese recipe, you need to come up with something else, something new, something different. And I don't know that LLMs are ever going to be able to do that in their current formulation.

Do you often look up average Bolognese recipes?

Yeah, I'm actually making average Bolognese for dinner tonight.

Just average, not the best?

Yeah, just, just medium Bolognese recipes, you know what I mean, though, right? Yeah. Like, no.

Nothing crazy. Don't be putting any like Worcestershire sauce in there or whatever. Like, just

Beef, tomato sauce, salt, pepper, bam.

Um, yeah, I, I get you. I get you. But like, you know, thinking about that sort of average thing, or thinking about that average work. I mean, there was a Gartner report that came out recently that said, you know, that the actual AI that is replacing human beings, whether partially or or or entirely in their jobs may be demoted. And I'm just wondering how likely do you think that is? Will, will that sort of thing start happening where, you know, you'll have Sam Altman-style revelations happening in the world of work where people go, "Actually, this was a huge mistake. I'm going to now demote you to just being a tool again, and I'm going to rehire some people."

I think it depends on the costs and also how competent these things are. Like I think what's happened is that basically like we've hired like a whole tranche of workers without really like putting them through any sort of interview process and then giving them jobs and then we're like, just basically seeing if they can do the jobs or not. So yeah, maybe they will get fired. I also think about like the, the cost is relevant to like the cost of human labor is relevant too. Um, I'm, this is going to sound like a tangent, but I promise you it's not. So when, when I was a kid, like there were like automated car washes everywhere, right? And you would, when you went to get your car washed, like when my, my parents go to get the car wash, they go to one of these automated car washes and, you know, the rollers would come down and all get done automatically. And those don't really exist anymore. Like they're not, there are some around, but like where we live, like there's not any of those around anymore. And instead now when you want to get your car washed, you go to a place and a human washes your car with a hose and scrubs it down and and blah, blah, blah. And that's happened because like it was cheaper to do it that way, to pay a human to wash your car, because the cost of the automated car washes went up or whatever, or because the people washing your car are getting paid cash in hand and not paying tax or whatever. That like, but for whatever reason, that the, the cost of human labor is cheaper than the automation. And automation is not necessarily cheaper because it's automated, right? Often it will be cheaper to just get a human to do the same job. And and I could see that definitely happening with, you know, AI agents as well, particularly as companies like that that are offering the AI agents realize that they do actually need to sort of like turn a profit and they can't run as like loss leaders forever. Like if you look at like what the actual, I've not done the sums on this, but I'm sure people have, like what the actual cost of, um, asking AI to, to, you know, write 500 words versus asking a human to do it. I'd be very interested to see what the actual cost is.

Yeah. And if any of our viewers know that, that please do share it with us. I think it's, you're 100% right. You're in a situation where you've got, you spent billions of dollars or billions of pounds on on a bit of kit. And if most of what it's output is like is is the very low-level stuff, it's not exactly, even if you do it at scale, it's not exactly going to create that ROI that that companies are under huge amount of pressure to to produce, right? So, um, and I guess a lot of the LLMs, especially that people are deploying at the moment aren't exact, they're not exactly reliable or up to scratch to do that higher-level work, um, that would perhaps justify their existence a bit more. I mean, what, what you said, what you've described about, you know, human beings being cheaper than than than AI is kind of what Microsoft has has been pointing out, which is, you know, the AI might be more costly than the, um, human beings that, that they replace, right? And that they've started actually canceling some of the cloud code licenses, right? According to a report from the, from The Verge, and then moving engineers towards other systems, um, because of it. So, I mean, is that something you think other companies are going to think about now and follow suit done?

Yeah, quite possibly. I mean, obviously Microsoft has got its own, you know, systems and things like that in place. So maybe it just prefers that that it's not, you know, spending money on like an external AI rather than, rather than, you know, people using Claude. But, yeah. Yeah. I mean, you could see that. I think like ultimately, like it, it depends on the companies themselves, the contracts they have, but also like what they're using AI for. I think a lot of companies have, you know, spooked by predictions of doom from, you know, Sam Altman and others have rushed into adopting AI without really thinking about whether they need it or whether they don't need it, or, you know, um, they're so terrified of being sort of left behind that they're kind of rushing into this technology without really thinking of of how they're going to use it or what they're going to use it for. So yeah, I think we might see a correction where, you know, some of these licenses, companies might realize, "Oh, actually, we don't need, we don't need to do, maybe we only need this much AI and not like a full, you know, into every single thing that we do." But, um, it's still early days, right? Like I think that the, the technology is still in its infancy and I'm sure it will prove useful. Like I don't, I don't want to come across as like a complete naysayer. I'm sure there are some aspects of of work where it will prove useful. Um, but whether those aspects, whether it will be useful enough to justify the cost is another question, I think.

Yeah, it's interesting that one of the reasons that Microsoft decided to make this change was because, um, they really pushed a lot of their employees to use, um, AI, right? And they, they were like, you know, we would love you to use it. Please do use this. And then more and more people were using it to the point where it got completely unsustainable, right? It's, it's interesting because we know from speaking on this program that, um, there, there are loads of examples of companies that are doing the exact same that, that, you know, tying the use of AI tools to performance reviews, to, to bonuses, to whether you continue to be employed at companies or not. Um, it does, it does feel like this flies in the face of that, right?

Yeah, I mean, companies make really bad decisions all the time, right? Like I think like they're not rational, they're not rational entities. I mean, the other thing with Microsoft, just to, just to add to that, is that like obviously Microsoft has got a big stake in AI being a success. So like it makes sense from their view as like a corporation to like be seen to be adopting AI like really broadly because, you know, that's the whole, they've reshaped their whole business model around that. Um, I think that, um, yeah, like targets for AI adoption, it's just about perverse incentives, right? Like it's like the, the, you know, like mid-level execs will have been given a target to adopt X amount of, to have X amount of workers using AI or whatever because, you know, another, another person above them will have a target of like doing X or Y or Z. And I think they maybe lost sight of what the point of the point of it is, which is like, I think, I think if, if you've, if you've been given a target to like cut costs by 25% or whatever, and you, you decide that AI is the best way to do that, then sure. If you've just been given a target to like adopt AI for the sake of it, that seems like slightly, um, moronic to me.

It's, it's an odd scenario, but it does feel like that is actually, it's, it's, it's taking place, right? You, you've got people who are very much pushed into it, um, and who may just try to do, I mean, I've talked previously on this show about, you know, performative, um, work where you're kind of in a situation where you go, "Okay, I've been asked to do these KPIs that are AI-related. Um, I don't necessarily feel like that's going to help my job, but in order to get ahead in my career, I need to be seen to be doing it." I suppose, do you think those, those sorts of things should be scrapped because it's, you can't regulate, you can't regulate against companies doing stupid things, right? You know what I mean? Like, you can't, like companies are going to do what they're going to do. I think like ultimately, like the, the market, market will decide if that's a good idea or not. And the companies that like overdo it will suffer as a result and have course cracked as Microsoft has done, or or not, or or, you know, or they'll race ahead and, you know, use their massive amounts of productivity to, you know, really zoom to the top of the, um, you know, the charts or whatever. So, yeah, I think the other, the other thing, like, and I think I can't remember if this is Ed Zitron or Cory Doctorow who wrote about this, but like this, this sort of like creep of AI into like every single product that we use, every single software product that you use has now got, um, AI, um, integrated into it, despite the fact that, you know, I don't want Meta AI in my WhatsApp messages. I don't want like AI in my word

documents. And, um, the reason for this is that because these companies are all selling, like, software as a service subscriptions, they basically feel like they need to add value to those subscriptions so that when they're coming up for renewal, like the exec or whoever is like signing that renewal says, "Oh, great. I'm getting more for my money."

So there's this constant creep of like, everything has to get better all the time. Everything has to be improved and new and improved. And actually, like, Microsoft Word has probably been fine since like 2004, right? Like, if they just left it like that and like updated it to work on new operating systems, it would have been absolutely fine. Like, we don't need all this extra stuff.

But because the people that are buying or procuring the products need to be, you know, they've got a target of getting value for money, then that sort of trickles down through the whole system and you end up with these products that like, no one actually wants.

Actually, AI is the thing, the new thing at the moment where it's just getting shoved into every product and, and I think that again, while also being ruinously expensive for the companies that are running the products, is also like a trend that is making people push back a little bit against AI and everything. Yeah, it does feel like, you know, that sort of thing like, if you buy a set of knives, we'll throw in a colander for free or whatever, and that colander happens to be AI in this scenario. And then you've got a situation where you're trying to use a product that you've known and you've liked, or maybe you haven't liked, but you use, you have to use it, and then suddenly AI will be involved. And I don't know if anyone watching has had this experience, but it will correct what it considers to be typos in my work, and they're not typos, and it's really, it inserts problems. There have been so many times when I've gone to send a message and just as I was sending it, you could just see the language being changed. Um, and it can be extremely annoying and change the tone of the way you speak and try to, you know, autocomplete messages and things like that.

And I, I do agree with you. I wonder why anyone would think these things are useful, but I do think that they do perform a function. And I've been thinking about this a lot because, and I'd love to hear your thoughts on it, because it feels like a lot of the time we're talking about scenarios where companies are saying to people, "We've invested a lot in this AI tool. We would love for you to use it." That's it. That's the messaging. There's not a lot of explanation. There's not a lot of like, um, chat about like, "Oh, we'd like to use it for this purpose or that purpose, or we feel that this will be a game-changer." It's just a sort of open-ended DIY scenario.

They, it feels that a lot of businesses still don't necessarily know where it could add value. They just know that they've bought it and it's available and you can use it, right? And so I think, you know, this kind of function of things being included in things that you know might actually perform a good function for the AI companies that are selling these products because it means that rather than having the task of taking evangelists into different businesses and teaching them how to use things and coming up with ideas to make AI useful, you just insidiously insert them into all these different tools that we already use. And therefore, they're there whether you like them or not, and you're using them whether you like them or not. And therefore, it kind of helps, I suppose, with name and brand recognition. What do you think about all this? Because this is the theory that I've been like cooking up in my mind for a while.

Yeah. I don't know if it's that, or if it's just like, I think that, well, it's just an overhyped technology, right? Like, I think that, and maybe it isn't overhyped. Maybe it's going to live up to it, but it's just the latest, just the latest thing. Like it's, you know, five years ago it was quantum. Every single company was like, you know, rushing to like make a quantum computer. And five years before that, it was something else. Like, you know, and five years before that, it was something else. And it's just like the latest craze. Um, and companies feel like they have to. And it does feel really like existential. It really feels like, and this is like props to OpenAI and Sam Altman and Anthropic and the kind of like AI sort of marketing department for really making it feel like, you know, "Get on the train, otherwise you'll be left behind" kind of thing. Um, so I think that's largely that.

And then I think like, if, like, I think as a general rule, like, you know, that old adage about like, you know, if you're paying, if you're getting something for free, then you're the product. Like, I think like if a feature gets added to like, like if AI is getting added to stuff that you use, even though you don't want it, and it makes it actually harder to use, then something else is going on. Like, look at Google, for instance. Like, Google built one of the most profitable companies the world has ever seen. Search engine page rank. Brilliant. Like, completely redefined the way that we access information. Like, completely like built the world we live in today by making it possible for anyone to find stuff that they were looking for. And then they just decided to like break that function completely and bring in like AI overviews, which like, not only get stuff wrong all the time, but also like completely, you know, uh, kneecap the very sources of information that Google, you know, Google promised to organize the world's information, right, when it first launched. And it did that for a long time, made a lot of money doing that, learned a lot of money selling ads against the world's information. And now it's decided to just like turn off the tap for whatever reason. So there's something going on there that like, you know, Google feels it can make more money from, you know, from its AI systems than it could from before. It can make more money from just serving you the ads directly in AI overviews versus, you know, in the ten blue links, which are, you know, getting hidden behind these layers of sort of abstraction. Um, which I think is interesting.

Yeah, maybe maybe it's that evolution you were talking about earlier of like, it's always a different company. First, we'll build you up, we will open the world's doors to you, and now we will destroy you kind of thing.

Yeah. Well, I mean, that is that is like Cory Doctorow's like enshittification argument, right? Like the the the way that companies will give you something for free and then they'll commoditize it, and then that commodity, like commodification, like moves up through the layers of, you know, different, um, layers so that, you know, they take advantage of the users first, and they take advantage of the suppliers, and then eventually they just like hoover up all the value. Um, and we'll see something with that like that with AI as well, where, you know, initially it was free, and then started charging, and then started like actively like selling stuff to you and the answers. And then like, ultimately, like if you are using a free version of like Gemini or ChatGPT, then you are the product, right? Like, you know, ultimately they're serving their advertisers or, you know, whoever is paying money to to surface those results. So then can you really trust that information? Not sure you can.

Yeah. Going back to the jobs thing and and just thinking about what you said about, you know, there being phases and almost, um, trends and fads that we're seeing time and time again. I suppose I wanted to end on, on maybe, you know, is there a note of hope for for the generation that's coming into the workplace and for all the rest of us in the workforce? Um, can we just hunker down and wait for it all to blow over, do you think? Is there a chance of that?

Uh, no.

Fair enough.

No, I'm I'm trying to think about like what might happen. I I guess like there's there's two routes, right? One is like we collectively get together as a species and decide that we need to like redistribute wealth more fairly and we end up with a situation where like, yes, AI is great and it does a lot of the work, but then that income gets redistributed fairly throughout society. A lot like people don't starve and we can all, you know, spend our universal basic income and use our time to become artists and, you know, freemasons and and all this kind of stuff. I don't know. Um, and so that's one route. Uh, the other route is that there's just like the AI, the AI bubble bursts and these companies just go away and like AI goes back in its box. But that feels also like a slightly fanciful notation to me as well. Like I, like now that these tools are out there, I don't see them disappearing. Like, yeah, maybe they won't be as useful as they were before, but like the jobs that they have taken aren't going to be coming back. Like if you're like a translator or, you know, you know, something that can really be really be done very well by AI, then then yeah, probably time to retrain as a plumber, I guess.

On that note.

I don't mean to like single out translators, that that felt me. Um.

Yeah, I understand what you're saying. It's like what's been done cannot necessarily be undone at this point.

You can't put it back in the box, right? Like.

Yeah. Yeah. That's that's that's fair enough. On that cheerful note, Amit Katwala, thank you so much for joining me today.

It was a pleasure. Thanks, Nasha.