Why Doomers Are Wrong about AI and Jobs with Google's James Manyika

So many top AI CEOs keep telling us that AI is about to take all of our jobs. I talked to a senior executive at Google who's willing to make a big bet that they've got it all wrong. This podcast is brought to you by Atlassian RoVO, the AI that takes your team from AI novice to AI native. Welcome to Platformer. I'm Casey Newton. And our guest today on the show is James Manica. He's the senior vice president at Google and Alphabet where he runs Google's research and labs operations, along with what Google calls technology and society, which is basically Google's effort to think hard about the broader consequences of the AI it is currently building. But before we get to James, we want to begin, as we always do here on the Platformer pod, with numbers. We are obsessed around here with the subject of AI and jobs. And each week, we're kicking off the show with fresh data trying to make sense of what is actually happening on the ground. So joining me once again to do that is Platformer fellow and Gen Z AI correspondent Ella Marquianas. Let's bring her in. Ella, welcome back to Platformer. Hi, it's wonderful to be here. How's it going, Casey? It's going great. You know, whenever we talk, I'm always curious, have there been any developments over the past week when it comes to the automating of your own job, sort of any, you know, danger signs on the horizon that you're scanning for? I love to tell you this, Casey. Nothing has progressed towards taking my job this particular week. Good, good. In fact, Claude continues to say false things to me sometimes, which makes me pretty happy. Shame on Claude. Yeah, shame on Claude. We actually caught, this is true, we caught Claude citing a Grokopedia this week as a team, which was hugely disappointing to all of us. Yeah, I do find, I think this is actually more true for Gemini than Claude, but like the thing they currently have weird blind spots on is like Googling. Like if a page looks nicely formatted and it like resembles a source they think is useful, then they just like will cite it to you. And so especially like for a while, I couldn't use Gemini because it was always sending me like these Times of India like articles that were like redigested, slightly false versions of like mainstream news articles. Yeah. Anyway, so kind of the opposite. However, I do have some gloomy statistics for you this week. Perfect, because when we went up, we did a survey of the Platformer podcast readership and we said, what is the thing that will make our show go viral on social media? And they said gloomy statistics. So you're perfectly hitting the task at hand. What have you brought to us this week from the world of AI and labor? Cool. So this will require a little bit of explanation. Basically, I'm going to talk to you about this cool survey from the Forecasting Research Institute where they get like 70 economists, many of whom specialize in AI and labor economics, a few of whom have Nobel prizes, and they aggregate their forecasts about like what is going to happen with AI, GDP growth, jobs, etc. So this is like a wisdom of the crowd thing. It's like Kalshi if it weren't just insider trading, is sort of how this is sounding to me so far. Yeah, verified Kalshi. Yeah. Yeah, excellent. Yeah, I can't guarantee they didn't do any insider trading. Fair enough. There may be some insider trading involved in this, but if it happened, we don't know about it. So that's our caveat. So, okay, so they go out to these economists and what do they ask them? Yeah, so the thing I really like about this survey is like, often when you're talking about economists consensus on AI and jobs, you kind of lump in two things, which is, what do economists think is going to happen with AI? How fast do they think AI will progress? And then like conditioning on a certain level of AI progress, what do, which is like really where you want to hear the economists cook, is like, given a certain series of facts about what AI can do, like, how will that actually affect, you know, jobs, GDP growth, stuff like that. And so basically what this survey did is split up AI futures through 2032 into three buckets, kind of like the slow bucket, the medium bucket, the fast bucket. And the slow bucket, it's like AI continues to basically be a personal assistant that you can't autonomously give tasks to very well. Instead of giving you research reports of like a literature reviews of a bad PhD student in 2032, it gives you research reviews of a good PhD student. You know, nice stuff. We'd love to see it. But like things are pretty prosaic. The medium scenario, which is actually the one that like is the median economist prediction right now, is basically, so AIs can do some dramatic things. Like, for example, they can do projects that would take a software engineer five days, and they can do it at a human level. So like that's already pretty crazy. There's like AI-based lab automation software. That's economists median roughly, although there's a spread, which does tell you that economists are like kind of bullish on AI progress. But then there's the scenario that really interests me, which is the crazy town scenario, basically, which is like- Yeah, let's go to crazy town. Yeah, which is like Sam Altman's dream of curing cancer happens. AIs can do all software engineering tasks better than humans. There's like this section, because the surveyors wrote a detailed scenario. There's this section where they're like, it can write a Nobel quality novel. It can pitch it and get it accepted. It can negotiate for the movie adaptation of its Nobel quality novel. Basically, like a truly crazy scenario for 2032. And now, that's not economists median. I wouldn't agree with them if it was. But they surveyed some economists from October to February. During that time, the economists they surveyed were like, it's about 10% we get to crazy town. They did another one from February to March. Median economist prediction had gone up to 20% crazy scenario by 2032. Wait, so between February and March of this year, the number of economists that they surveyed who think that we're growing to crazy town basically doubled? It's like a little bit more complicated than that because they basically tell them to give you probability distributions. Like every economist you give will say like, I'm 50% on medium scenario. I'm 30% on no takeoff scenario. But it's like the prediction, the median prediction of economists that this might happen doubled. Okay, so that's like pretty shocking, right? And I can imagine that just between February and March, we're starting to see the rise of some very powerful coding models. It sounds like maybe the economists got their hands on Claude code or codex during that time. So that is very notable that these sort of august economists think that. Let me ask, in this sort of rapid takeoff scenario that an increasing number of economists are believing in, what does that do to like labor force participation? Like, you know, is there an effect on jobs here? Yeah, so in terms of jobs, they focus on labor force participation, which is a measure not of how many people who want jobs have it, but have jobs, but of how many people are like looking for jobs at all. And so what they find is the prediction is currently we're at like 62.2% labor force participation in the U.S. The median economist prediction for the rapid scenario is 55%. That would be like a really historic reduction in the amount of people participating in the labor force. Basically, the lowest it's hit in the 21st century was a little below 60 during the peak of COVID. And it's only been anywhere close to 55% in the like mid to early 20th century before women entered the workforce. So basically, if you see this- Wow. And this is the median prediction. This is the median prediction conditioning on crazy scenario. So the median prediction for the median prediction for medium scenario is it goes down a few points. It comes closer to the COVID low. You see something like unusual, but the prediction for the like really strong scenario, which is only around 20% for people right now, is a really dramatic one. This is great. So this is great. So our message to the world is if you want to think through the median AI scenarios in a world where things just continue to improve at the current rate, it's sort of like we're having a global pandemic. It's like sort of what you're sharing with me right now. I think it's like, it's interesting because like kind of many people disagree on even what the current rate or level of progress is. Like, I remember when I was looking at scenarios, I was like the slow scenario, some people might say is already happening today because they look at these AI outputs and they're like, this is like a decent PhD student. And other people are like, we're nowhere near that. Right. Yeah. And then I think my takeaway here is like economists think there is like in the 20% range chance of like something above the job impact of a global pandemic. Yeah, 20%. So like lower probability, but you know, one in five is still very significant. So those are some fascinating stats to consider and I think are going to ground the conversation that is about to follow because our guest today is not somebody who is worried about massive job loss due to automation in the near future, although he does think that AI is going to change jobs quite a lot. So Ella, thank you for bringing us some more fascinating news from the world of AI and jobs. And with that, I think it's time to take some of these ideas to this week's guest, Google's James Manica Montreal, Toronto, Jeff Hinton crew. So he came back and suggested that I should actually build a neural network for my undergraduate thesis. So that was actually the very first thing I ever published in my whole life. So... And what year was this? It came out in 1993. Yeah. So well, well before, you know, folks like me were spending every waking hour reading and writing about this. What was it about the subject that captured your interest? Well, it was two things. I mean, I grew up on Star Trek, so the idea of AI and science, I'd watched 2001, A Space Odyssey. There was a, so there was a part of it that was just fascinated by this. And on the other hand, I was just intrigued by the idea that, you know, would it be possible to build systems that can actually do advanced cognitive tasks? Seems interesting to me. So I just began reading up on it. So when I went to Oxford, I did my PhD in AI and robotics. So to kind of continue and pursue this. Yeah. Well, you have since spent a good portion of your career trying to measure how technologies change economies. You spent a long while at McKinsey where you wrote a paper called Jobs Lost, Jobs Gained almost a decade ago. Now you're inside Google where you can see what happens when these tools actually land in workplaces. When you look at the debate that we're having right now about the future of AI and jobs, there are folks that are very bullish on the future of the sort of worker in the economy. There are other folks who think that a lot of jobs are about to go away. Where are you in this moment? Well, I think it's such an exciting moment. I mean, clearly the technology and its capabilities is expanding at an incredible pace, at an incredible pace. But I think when you then try to translate this into what this might mean for work and jobs and occupations, I think you get a very mixed view. And it's roughly what that paper you mentioned said, 10 years ago, which I think is still correct, which is there will be some jobs that will decline. There will be jobs that will grow. And most importantly, a third aspect, a lot more jobs will change. And I think it's worth thinking about all of that because, you know, whether you're looking at the aggregate economy level or sectoral level or by occupation, you get a different mix of those three things happening. But I think all three things will happen, is what, you know, and the research, you know, hasn't changed very much. I think the debate that people have is what's the mix of those three things as opposed to are these three things going to happen? Right. Let me just name a dynamic that I'm sure may be on some listeners' minds. You are now employed by one of the biggest beneficiaries of the current AI boom. So when you think about what's coming, you look at the landscape, how do you tell when you're hearing the labor economist in your head and when you're hearing SVP at Google? Well, I actually hear both things. I think, you know, so and it's actually less SVP at Google more. So the AI researcher, computer scientist in me is extraordinarily excited about the pace of the technology and the advances and the capabilities. So when I, that part of me sees the pace and I think, oh, my goodness, this is going to be extraordinary. This is going to be extraordinary and it's going to happen very, very quickly. The labor economist part of me says, hang on a second. These things don't actually play out that quickly in the economy and the dynamics are a little bit more mixed. So I almost see two speeds proceeding here. And so often, I often think of this as AI researchers, our community often tends to overstate what happens in the labor markets based on what they're seeing on the technology and the frontiers of AI advances. I think these are two very different conversations. On that front, when you were at the McKinsey Global Institute, you did some research. And several years back, you found that about 50% of tasks you believe would be automatable through AI, but that only about 10% of occupations would be fully automatable. I wonder, a few generations of AI later, does that ratio still sound right to you? Or do you think that now that we have agents, it's starting to move? I think all of the pieces have actually been moving, by the way, because keep in mind that at the time when we did that work a decade ago, we took a task view, which by the way, seemed novel at the time because people were making these whole job predictions without actually saying, well, actually, if you look at what occupations are really about, it's a combination of tasks, essentially what make up what you and I do, Casey. So if you take a task view and you try to understand the impact of AI and automation technologies, then you get the view that I had 10 years ago. I think what's happened since then is that at the task level, many more tasks are now possible to automate. And I think for that picture's moved pretty quickly. But when you look at what makes up the composition of occupations, and if you go through the typical, so the Bureau of Labor Statistics kind of tracks somewhere between 850 and 900 occupations. If you look at all of those and you take some of the data sets in the ONET data set base, which looks at the task composition of each of those, you come to a conclusion that basically says, well, how many of the existing occupations look like they have the majority, call it 90% of their constituent tasks automatable? And that number is still under 10% in my view. And most researchers will still say that versus if you say, okay, so how many tasks look like they're going to be hard to automate, partly because the AI can't do that yet, or that some of these are coupled tasks where the weak link will slow down the combination. So if you take two tasks and you can automate one of them, but they need to be done in a coupled way, guess what? You'll only go at the speed of the last weakest link in that combination. And so most jobs have these coupling of tasks that it makes it very difficult. The other thing, by the way, Casey, I'll mention, which I'm glad our, you know, the research community is now paying attention to, as are some of the labor economists, is that it actually matters to look at how long tasks take. So this is one of the things that's moved quite a bit. If you had asked the question back in 2017, I remember looking at this to say, of the tasks that are possible to automate, you'd find that some of them that are very short duration tasks that require 30 seconds or a minute is about as long as you can predictably do that task, you know, in an automated way. But now we can now do some of those tasks for up to four plus hours. So the task duration with reasonable predictable completion of the tasks has also made a tremendous amount of progress. And I think we should take that into account. Yes. There's the meter time horizon chart is, you know, maybe the most looked at chart in Silicon Valley right now because it measures this exact thing, which is how long can an AI system reliably do a task? So what I'm hearing you say and what is so interesting, and what I feel like gets us into the heart of this discussion, is that if you were to measure the sort of tasks that are automatable now, that number might be much higher than 50%. Or at the very least, it's trending to be much higher than 50%, let's say, you know, within the next year or so. But at the same time, you're saying that other number, that how many, like what percentage of jobs you could automate, it's sort of stubbornly holding in roughly the same place that it was 10 years ago. So to me, that feels like the nut of this whole debate. Like what is your best explanation for what that divide is? Well, I think part of the divide is that first of all, we now, and we understand more fully that whole jobs tend to have a much, much more complex mix of tasks. And this idea of weak links or coupled tasks actually matters a great deal in most occupations. It's most occupations. So therefore, if you look at across the whole economy at most complex tasks, the, you know, we can't automate most of them. And so the automatable jobs, when you say, okay, so which whole jobs can you automate more than 90%? It's still a relatively low, small number. In fact, most of the debate amongst the labor economists debate whether in the next decade, is that number more like two or 3% or is it more like nine or 10%? I don't think anybody who's looked at whole task, whole job automation would say it's more than, it's half or 50% or any of these extraordinarily large numbers. And it's the reason, Casey, why I come back to the view that this idea that three things will happen, yes, there will be some job declines, jobs lost over time, but we should keep in mind that there'll be jobs that'll grow and the jobs growth part of this, the jobs gained is a function of two things. One is existing jobs that will become, that will actually grow in demand because the technology often changes the demand picture for some activities in the economy. And then also new jobs that will be created. I think we forget that, for example, you know, David Autor has done this work. If you'd gone back to 1945 and you look compared to today, something like over 68% of the jobs we have in the economy today didn't exist. And many of the ones that have I did. But I remember, I used to say when people would ask me, what should my kid do back 10 years ago, and I'll tell you why I'm saying this because I don't think my answer has changed. I would say, tell me how old your kid is, would be my first question back. If they told me their kid was 18, I'd say they should learn to code. When they told me that their kid was two, I'd say, well, hang on a second, you should think about what kind of skills are going to be important because this AI thing is going to make it big. It's going to make lots of progress. And I think this brings you back to the entry-level question you asked. I think the thing we've said at the time, which is correct, but it may no longer be true, is that coding was going to be important in the mechanical sense of learning to, you know, churn out lines of code. Now that the systems are able to do that, that doesn't mean computer science as a field has gone away. And I think, I remember when I studied computer science as an undergraduate, the coding part was just one slice of what I had to learn to do. I had to learn algorithm design. I had to learn all of these other things. I had to learn there's so much more I had to learn. I think we may need to go back to that because even today, I'm finding that, and we're finding that often it's the more broadly educated, skilled computer scientists who are a lot more interesting than just simply the ones who, their only claim is just the ability to generate lines of code. So I would say to the new graduate, guess what? There's lots of opportunities. The preparation and skills required for that have actually changed. So part of that is to keep up with those skills. I'll say one more thing, Casey. You know, I was looking at the data the other day. The demand for software development jobs is actually going up. Yes, it is. It's actually going up. So it's not as if these jobs are going away even in this moment, but I think the skills required for it are changing. Yeah. And it is, you know, for what it's worth, like, this is another, um, attention in Silicon Valley right now that I'm quite interested by. Because just within the past week, I've had a chance to talk with, well, you, um, Aaron Levy from Box, and Nikesh Aurora, the CEO of Palo Alto Networks. And both Nikesh and Aurora said, please send me more engineers. I don't have nearly enough engineers for what I want. At the same time, you look around the valley and it's earnings call time and they get on the earnings call, other CEOs do, and they say, um, well, you know, we're getting rid of 5% of the workforce, uh, you know, to prepare for the AI future. Um, so I understand, you know, maybe some of those folks are not, uh, or, or, you know, sort of want to have a nice story to tell the market. Um, but you, you do hear both things. And I would say I'm hearing them pretty equally at the moment. Yeah, but I think there's so much more going on than, than, than the AI effect. I mean, I, in fact, I'll, you know, as somebody who's super excited about AI's impact on the economy, I'll say not much has happened yet. And the reason I say that is, I'm saying that both on the positive side and also on the negative side. On the positive side, I think, I think at the economy level, we've yet to quite see the productivity gains, which I'm, I'm looking forward to and I'm excited about, but we also haven't seen much of the AI-driven labor impacts that everybody's talking about. I mean, I think there's been quite a few papers now that have been written. I don't know if you, there was a time when I think there was a paper that came out, uh, I think this might have been the Canaries in the coal mine paper. And what I found interesting about that paper was, if you look at where the sharp declines happened and they showed in those charts, uh, they're all like in October, 2022. ChatGPT didn't come out until November, 2022. And in fact, adoption didn't really happen in the enterprise space until maybe 2023. So if the sharp declines in entry-level hiring happened in October, 2022, I think you'd have to believe something else was going on, I think. Right? And I think there's now been analysis that showed that there were a whole bunch of monetary effects around what was going on in the labor markets. Uh, there were also just leftover hangover stuff from COVID. So I think when we look at what we're seeing in the labor markets, uh, there may be a tiny sliver of that that's AI-driven, but a lot more of it is driven by, quite frankly, other macroeconomic effects and things that are going on. That's not to say, by the way, that we, we should therefore not worry about the labor market's effects. We should. I just don't think they've happened yet at the scale that, uh, anybody's, is, is concerned about. That, that seems fair to me. Um, in that case, let's move further into the realm of speculation. So you were talking about job change earlier, and this is another point that I had been spending some time trying to think through. Um, when I talk with folks whose jobs are beginning to change, uh, in part due to AI, it seems like an increasingly large part of their job is reviewing AI output. And whereas once they spent a lot of time, let's say, manually writing code, now they spend more time reviewing code. And I think they're like analogs for this and in different kinds of knowledge work. Um, and the thing about reviewing work is it doesn't always scratch that same creative itch, maybe as like writing code in, in the first place. So I'm curious if you've given some thought to that and whether we may, you know, job change winds up being a different kind of job loss because we may see some jobs that once felt very creative that now just sort of feel like tedious drudgery. Uh, yes and no. So, uh, I, I, I think, yes, you're correct to say that on the one hand, I think there's a fair amount of work that's now reviewing the outputs of these systems, making sure you're guiding them and you're nudging them in the right way to say, no, no, no, you're going down the wrong path. Do this. So yes, there's a fair amount that's that. But I also think that the creative output is also extraordinary. I think the creativity becomes quite different. It isn't about figuring out which has the right code repositories to look at and dig into, but it's kind of more problem solving. So I, you know, I actually like this moment in the sense that you're now having to be creative about how, what systems should you be building? What questions should you be asking? Which experiments do you want to run? Uh, if you've got 10 agents working for you, uh, what different directions are you going to send them in? Uh, what orchestrations are you going to run? I mean, what kind of tournaments do you want to run, uh, for your different agents to experiment in different ways and compare the results? So the nature of the creative problem solving is a little different, I think. Uh, so I don't think the creative part goes away. And by the way, the reason the creative part is exciting, uh, for, for, for many folks is, you know, it's going to be the scariest thing and the hardest thing to do. But I can guarantee that people are going to come up with great ideas doing that, are going to be built extraordinary, exciting systems. I mean, I've seen this in, in, in our own teams, right? If you, if I think about the thing, the team, the teams in Google Labs that are creating these extraordinary new AI first products. I don't think, sure, they spend some of their time reviewing what's, what's come from these systems, but a lot of the time they're dreaming up new things. To build. Yeah, and often, you know, a nice side effect of being able to hand off sort of automatable tasks to an AI is that you have more time to do that kind of thinking and creating. Um, you know, your, your answer had me thinking about the kind of work that I do as a journalist. That is a creative profession. I could, if I wanted to, and I don't want to, ask an LLM to generate a column for me. Um, but even when I have just tried to do that, you know, as a sample to kind of say, like, can this thing mimic my style? I find that just reading through it, you know, it, it gives me a bad feeling. On the other hand, when I've used these systems to say, can you help me be creative? I find that they are very good at that. And, and in small ways, I start to, I, I sort of feel like they're helping me get a little bit better at what I'm trying to do. So I think, I think there's something interesting there. Oh, I absolutely do. In fact, there's, there's a, there's a wonderful chart in a paper that David Autor, the, an economist at MIT, kind of put together, which is when you, if you think about a task instead of tasks, and, you know, and, and you ask this question about what's automatable or what's not, you can either, you can automate the tail end tasks. Those are the tasks that has always stuck with me was that one of the mistakes of the globalization era was that even though, you know, the so-called China shock in aggregate really only impacted something like 4 million jobs, by the way, which at the scale of the U.S. economy, you could say that's not very large, but if you're one of those people in those locations, in those occupations, it was everything. And I think what we did in terms of either safety nets or wage insurance or transition support to help those workers, we didn't do anywhere near enough. And I think that's a place where policy can make a big difference. So even though, back to the original question you asked about work, Casey, which is, I'm relatively, I mean, there's going to be a lot we're going to deal with, but I'm relatively sure there'll be work and there'll be activities in the economy, but I'm also sure there's going to be a lot of change and a lot of that change is in the nature of skills and expertise needed, but also in people either changing occupations and occupations changing. And I think policy can play a role in assisting and cushioning some of that. So I worry that that's the part we won't do. So even though, so if you ask me, if you ask me my, you know, what things keep me awake at night about AI and work, it's not job loss, quite honestly, for the next decade. It really isn't. But it's these questions about how do we support the transitions that work and workers are going to need to go through, and how do we make sure either companies like us, but also employers and policymakers keep that in mind as we navigate through this moment. That's what I worry about. Absolutely. And I think this is a huge point. You know, I'm sure you're looking at the same surveys that I am. There's a story as we record this today that 70% of Americans now oppose the construction of data centers in their communities. And my sense is that while they have some concerns about AI that go beyond the economic ones, I think a lot of the fear is just rooted in the fact that if they were to lose their job today, there is not going to be some obvious, you know, government helper standing by to help them navigate into the next point in their life, right? And I think it's really interesting to think about what the conversation about AI would look like if the average American felt like there was a plan for them if something was about to happen to them. Yeah, I think that's one of the things we have to do. I mean, I also think we have to think about, you know, obviously AI has a growing energy footprint. I think we have to think about that. And we also have to make sure that as we build this infrastructure, it doesn't actually increase the cost of energy for people. I mean, you've seen, I hope you've seen, we've made, we've had a pledge about making sure we're going to bring our own energy. We're going to make sure, you know, it doesn't raise the rates and cost of energy, but at the same time, I think having a confidence that there is a plan which involves everybody, by the way, to support the transitions and change that are going to happen in the labor markets, I think is very, very important. I think that's what we may perhaps be failing to do. So I think it doesn't help, by the way, back to your question about, you know, AI wiping out jobs, 50% of the jobs. I think it doesn't help when we in the AI field talk that way. First of all, I don't think it's going to happen. As I said, I'm willing to take bets with anybody if you wanted to take a two-year bet on wiping out half the white-collar jobs. But I think, to the extent that the economic research and analysis mostly doesn't say that. I think we're, you know, we're probably impacting the possibilities of this technology having extraordinary impact by quite frankly, scaring everybody when in fact that fear is actually quite unfounded. So let's not confuse the pace of technological development with how quickly this plays out in the economy. And I think we should be much more thoughtful about how we think about, well, create a plan, but also make sure that we're being clear about what we see the economic impacts are of this in the labor markets. Let me ask you another policy question, which I think is maybe semi-related. In 2023 at the UN, you co-led a project called Governing AI for Humanity, which laid out a fairly ambitious vision for global governance of AI. I'm curious, three years later, what is your view? Have we made progress there or are we moving backwards when it comes to global governance of these systems? Well, we haven't made much progress in the following sense. I think, you know, in that project, I was co-chairing the UN's High Level Body, and what was fascinating by that work, by the way, is because of the nature of what that body was doing, and I was the co-chair of it, was we were engaged with all the countries in the world, all 190 countries of them. And it was actually quite interesting that the attitudes towards AI were very different around the world, by the way. You know, people in the so-called global South, you know, in Asia, in India, in Africa, in Latin America, were generally more positive about AI and its possibilities. Of course, there are other concerns and issues around infrastructure and access and making sure, you know, their needs and uses were reflected in how this was being done. You know, Western Europe and the U.S. were probably the most negative, by the way, which is kind of interesting. But I think to your question about have we made progress? I think at the time, there was an understanding, which I think is still the case, that there are some basic things we should all agree on. Things like this technology should be based on fundamental human rights, that we should, the respect of international law, that we should think about both the possibilities as well as the harms, not one or the other, of this technology, that we should think about how we make sure everybody can participate both in the development of this technology, but also in the use of it. So you have these kind of basic core principles that I think, I think and I hope the world still agrees on because at the time there was general agreement by most countries, by the vast majority of countries, about that. The question then became, okay, so how do we put this into practice? I think as the process thus taking much longer. I'm happy to say, though, that one of the recommendations that we made is now in place, which is the creation of a scientific panel that now works on some of the scientific questions involved with regards to AI. I'll tell you one thing that I learned from that process, by the way, which is fascinating for me, was that, you know, in discussing the risks and possibilities, you know, from this technology, there are the ones you might expect, right? So risk of misapplication and misuse. But many members of the body, including especially the ones with the global South, insisted on adding mist use as a risk, which I thought was interesting. As in not using it. Yeah, as in not using it, as being an actual risk because, you know, that actually got written into the recommendations. The idea was there's some people who live in places, countries, and communities where the risk of not using this AI is actually far greater than the current circumstances they live in because they may live in a, if it's in the U.S., in a county where access to specialist doctors is very low. I mean, I don't know if you know this. If you ever looked at the map of the U.S. at the county-by-county level, I was amazed at how different access to expertise in healthcare is by county, at the county-by-county level. There's some counties where certain types of expertise just aren't there, and some counties where you have 10x the number of oncologists or 10x whatever, right? So these questions about, for some communities and some countries, the risk of misuse is actually quite high. So often when I hear people say, you know, AI is not good enough, I often ask the question, compared to what? Yeah, what is the baseline? So if I have access to Stanford Medical Hospital, I live in San Francisco, yeah, it may not be as good as that, right? But if I'm in a place where I don't have that, I don't have access to the world's best oncologist or the best doctor, I'd ask the question compared to what? I think this is a fair point. You know, I think that we in the press are often focused on moments when the AI is giving bad advice. And I do think it's happening in, like, single-digit percentage of the time. And some of that can be, like, quite bad and dangerous and harmful. It is also the case, though, that I think particularly if you already have some expertise in a subject, AI can be a powerful guide. And if, as you say, you live in a community where you do not have access to a lot of expertise, the thought that you might miss out on that forever, I could understand why some countries are concerned about that. Yeah, I mean, you could even apply the same questions to driver's cars. If you say, you know, they're not safe, I'll say, well, compared to what? Right, right. We already know in the case of Waymo, for example, that, you know, the instance rate is 30 times better than your typical human driver, right? So I think these questions of, now, this isn't to say AI is perfect. Far from it. I mean, this technology still has lots of gaps, I think scientists are also going to be able to connect ideas beyond their own disciplines. So we've been building a whole bunch of tools. I mean, we published a paper that's in archive called Co-Scientist, where you can now compare ideas from multiple disciplines, even in biology. So you're no longer as a scientist constrained to just what you know in your domain or your part of the discipline or your part of the field. You can explore ideas in many more papers, in many more disciplines, theories across the board, and then bring them all together with the use of these agentic systems and then construct experiments. So the ability for scientists to ask questions and construct experiments and decide what to explore is going to be vastly different than, you know, take AlphaFold. Scientists used to spend three or four years working on one protein, doing X-ray crystallography to figure out the structure of one protein. Now they can just look it up in AlphaFold, right, in AlphaFold 2 and 3, right? And then, sure, there's still some validation to be done, but you're starting in a very different place. And so therefore, the diseases you can study are much wider. The reason we have 190 biologists using AlphaFold is because before, if you were in a place or a country where you didn't have a lab and you needed to understand the structure of a particular protein because it affected a particular disease, you had to wait until somebody had figured out the structure of that protein you needed to study in order to understand a particular disease. Now you just look it up. So I think, so that's, I know we spent a lot, I spent a lot of time on the science. I think you're going to see that kind of change in the nature of work across many, many occupations, back to even software engineers. I think the way software engineers are going to spend their time, we're already starting to see glimpses of that today. I think that's going to continue to change. And there's so much more software to build, by the way, so. I believe that. Well, you know, my last question was going to be, if I had you back in a year, what is, you know, the single number or fact that we would want to track to see, you know, whether the worldview you presented today was valid. But I sort of think we already have the number because you told me right at the top of the show that we are not going to see 10% of jobs automated away in the next year. I don't think we will. I'm happy to take bets. All right. Well, it is a bold bet and it is what I want to hear, but I think you've made a great case. So James, thank you so much for joining us here today. Well, thanks for having me, Casey. And there's so much that's exciting about this topic, but I always, you know, if I, if I say one last thing, if I could, is I think our biggest challenge is to how do we hold two kinds of ideas in mind, both the exciting possibilities, but also the complexities and challenges that we're going to need to grapple with. I don't think one is, it's not one or the other, it's both. Yeah. Well, for what it's worth, I feel like the world is very much now on the page of challenges and complexities, and they're looking for more reasons for optimism. So I think anything you can do on that front, the world is ready. Thank you. Also, one last thing. And one of the challenges we've got on the optimistic side is that, you know, I mentioned all the things in science and in the societal implications. The challenge is that a lot of those are going to be indirect for most people, right? Because what are you going to get, Casey? You're going to get a flood alert on your phone to say, get out of the way. And you'll say, thank you, whoever sent me the flood alert. You won't say, okay, this wasn't possible before because of AI, but now is. You're going to get a great cancer screening and you'll say, great. But, you know, so I think a lot, and the challenge we've got is that many of the beneficial impacts that are already becoming real, most people don't think of them or experience them as directly as AI. So I think we have to, we have to work on that part of it too. Well, a huge part of that is going on more podcasts. So something for you to kind of think about as you're scheduling your time. Well, thank you. Thanks for having me. Thanks, James. Platformer is produced by Lindsay Chu and edited by Fitz Harris at Story & Sound. You can watch this whole episode on YouTube at youtube.com slash Casey Newton. My email is casey at platformer.news. And we'll see you next week. Take your team from AI novice to AI native with Atlassian Rovo. Go to rovo.com to learn more today.