 Hello everyone. My name is Sadia Zahidi. I'm a Managing Director at the World Economic Forum. A big topic on the minds of many that are here in Davos is artificial intelligence. It's also the sustainability revolution and the green transition. And it's also what the economic picture currently looks like and how all of that then has an impact on jobs, on people's skills, which sectors are affected, which specific tasks are affected. I'm going to tell you a little bit about our Future of Jobs report, which we launched in mid-2023, looking out at the next four to five years. So most of these projections are through to 2027 from the time that we did it. Now, just a little bit about the report, we surveyed more than 800 firms and it covers about 11 million employees and covering 40 economies. And in total, when we then applied that data to how it was structured across the workforce using ILO data, this covers about 673 million jobs. So we're really able to say quite a lot about what is happening in the global workforce. We could have the next slide. Thank you. At a very aggregate level across all of that data, we find that the green transition is likely to be a positive. Yes, there will be jobs displaced, but many more will be created. So it's a net positive. When it comes to technology, the picture is a lot more mixed and you'll hear more about that soon. Yes, a number of new roles will be created, but there will be roles displaced because what technology can do now will disrupt many tasks within roles, but in some cases, entire jobs. And then finally, the economic outlook, the uncertainty that currently exists, the lack of new investment that currently exists. That's really what's disrupting that longer term outlook and is probably the biggest risk when it comes to creating a net negative outlook for the future of jobs. Now as a whole, across those 673 million jobs that we looked at, there's a 23% churn. So about a quarter of all jobs will be affected. Slightly higher numbers that are declining and slightly lower numbers that are increasing. But on the whole, we're looking at a picture where roughly speaking, there will be almost the same number of jobs that will grow as the number of jobs that will decline, but it's not a direct transfer between those jobs. So the types of jobs that will decline include, for example, people that are bank tellers or people that are administrative assistants, but the type of jobs that will increase are, for example, people who are sustainability specialists or teachers or agriculture workers. The largest absolute gain is going to come from agriculture and from education. So it's not necessarily a direct transfer between those jobs that are declining and those jobs that are growing. Now what does that mean for people's skill sets? Even for the middle set of jobs that are not changing too much, there will still be a massive change to skills. So out of everything that you and I do every single day in our workplace, nearly half of that is going to be changing over the course of the next four years. So 44% of the core skills that are existing in current jobs. Now again, this is aggregate and it depends which specific job you look at. Now what does that mean in terms of the types of skills that are in demand and what the biggest upskilling needs are according to the businesses that we surveyed? Number one and number two are deeply human traits, analytical thinking, creative thinking. That is what businesses are looking for in order to upskill their employees. But following on closely from that, artificial intelligence and big data, technology literacy, number seven, number eight, design and user experience. So quite a lot of technology skills in there as well. There's also quite a lot about self-management. So resilience, flexibility, agility, curiosity and lifelong learning, empathy and active listening. So quite a lot in there that is also about the softer skills that we all need to develop. That's what's coming up in the aggregate across all of those jobs that we analyzed. Now we also looked at large language models and about 11,000 tasks across 800 jobs. And we tried to understand what might happen due to large language models potentially being able to disrupt those set of tasks. And that roughly divides into three types of categories of tasks. There's high potential for augmentation, not surprisingly, when there are administrative or clerical activities that are repetitive or things like designing databases or analyzing data to improve operations. A lot of that is actually automatable now. But there's a huge scope for augmentation, evaluating personnel capabilities or performance, collecting data about customers, reading documents or materials to inform work processes. So augmenting our productivity. And then finally, there are many, many, many jobs which have very low potential for exposure to artificial intelligence, specifically large language models. And that are things that are service work. It's people who are in customer-facing roles. It includes things that require a lot of collaboration. For example, in the development of educational programs. So there is a lot that is also for the moment unexposed depending on how fast the technology moves. Now, this is one way of looking at large language models and what they might do to jobs. There are others as well. And I'm going to turn it over to my fellow panelist, Eric Bilneurbson, to tell us a lot more, the bigger picture, this research, as well as his. Thank you, Sadia. It's such a pleasure to be here. It's been a pleasure working with you, Sadia. You know, all of us have been playing with mid-journey and chat GPT and a lot of really cool technologies. If 2023 was the year of playing with those great technologies, I think 2024 is going to be the year of harvesting real productivity gains. In other words, it's a year that AI gets a body and starts actually being able to do things, transforming the world, transforming work, transforming productivity. And it may seem like it's an overnight success, but like most overnight successes, it has at least a decade in the making. Going back at least 12 years when Jeff Hinton and his team introduced deep learning techniques and won the ImageNet competition, and I see Andrew Eng here in the back who was a pioneer in using these techniques as well. And they've been working on improving these techniques. And the first big set of successes was actually not in generative AI, it was in predictive AI, mapping a set of inputs X to a set of outputs Y. And if you have enough data on the inputs and the outputs, you can find, a machine can learn the relationships and how to map them together. And there have been many applications that have been quite successful. About six years ago, Tom Mitchell, Daniel Rolwalk, and I tried to look at using this task-based approach that we've been talking about to understand what kinds of tasks and ultimately what kinds of occupations, what kinds of industries, what kinds of companies would be most affected. First by predictive AI and then generative AI. And here's the chart for 950 occupations. Each of these dots here is one of the occupations in the Onet taxonomy. And we found that almost all of them would be affected to some extent by predictive AI. There wasn't a single one that was unaffected. At the same time, there wasn't any that were completely transformed that were automated entirely. Most of them had different levels of transformation. And you can see, for instance, up here, cashiers were ones that are relatively low paid. They're on the left side of the chart, have a lot of potential for transformation. There are other occupations that are relatively high paid. Like over here, we've got pilots. There's a lot of potential for using predictive analytics and predictive AI there as well, even though it's relatively high paid. And I couldn't help looking at economists and I wanted to see where we stood. I was a little surprised. I thought that a lot of the tools would transform our occupation more, but we're actually kind of on the lagging end of all that. But the other thing is, if you look at the big picture, there's a bit of a pattern there that the lower paid jobs tended to have more tasks that were suitable for predictive analytics. Well, now we have generative AI. And what you see with the next slide is that the pattern is very different. That although the, again, we have the wage on this axis and the susceptibility to the tool or the exposure to the tool here, now the curve is somewhat more upward sloping. Economists are much more affected by these new tools as are a lot of creative professionals. So we used to think we're relatively immune, like marketing professionals and many administrative and management jobs, many professional jobs, things that doctors and lawyers do. So we have a different set of folks being affected. But it's still that basic task based analysis. And so if you remember that a occupation or a job is a bundle of tasks, you can break that down and see how much each occupation is going to be affected. For instance, if you look at a marketing specialist, they have to write marketing copy. They have to do market research. They have to train new employees. They have to do about 20 other tasks according to the owner taxonomy. And my company, Work Helix, has broken that down into like 200,000 tasks now for the whole economy and hundreds for each occupation. And if you wait each of those by their exposure, you get a sense of what the total effect is likely to be for that particular occupation. And in turn, you can roll that up to a company and see where the exposure is for a company or you can roll it up to an industry or to a country. And now we're beginning to be able to have a plan for harvesting that productivity that I talked about. It's not enough just to play around with the tools. But if you know what tasks are most affected, you can prioritize what parts of your company or parts of your country you want to work on first. So if we go to the next slide, what you see is that this exposure can happen in a number of ways. As Sadio was mentioning, it can happen through augmentation. And that means you don't try to replace the entire task or the entire occupation. You use the tool to augment it. And so we did this, for instance, in a call center we studied how you could use generative AI to help the call center operators do a better job. And within three to four months, they were already on average about 14% more productive, more calls per hour. They also had higher customer satisfaction. They also had less employee turnover. So stockholders, customers, workers all were better off. And interestingly, the least skilled workers actually had the biggest lift from that in that particular case. You can go through every of the tasks in your organization and get a prioritized list of what you want to do first and where you want to do that transformation. In other cases, there are opportunities to do more full automation. And in those cases, you don't need to have the human in the loop nearly as much. I actually find that most executives, most policymakers, spend too much time focusing on this column and not enough on this one. And then there's another category that they mostly don't even think about at all. If we go to the next slide, we see that there are new jobs that are emerging, entirely new tasks. And by definition, it's hard for us to think of those because they don't exist yet. But they often emerge organically. And if you allow your employees to work with the tools, they end up finding new ways of using them more effectively. So ultimately, what we have now is an opportunity to take these remarkable technologies and transform work, create higher paid jobs, not just more jobs, more good jobs, have more productive companies, countries and economies. And I believe that 2024 is going to be the year where a lot of that starts happening. And we're seeing a move towards a new trajectory for the economy. So at this point, I think we have time for questions, right? So I'm happy to take questions for about, I think we have about 10 or 15 minutes for any questions you may have about this. Just feel free to raise your hand. Let's go over here. Well, let's use the microphone here. Thank you. In the work that you're doing around tasks, have you looked at how you can take parts of old jobs and then path them into new jobs like re-skilling upskilling? That's a great point. So a job, an occupation is a bundle of tasks. And what you may find is that some of the tasks can now be done better by a machine and you may want to re-bundle some of the other ones in a different way. So maybe a radiologist, maybe reading x-rays is now something the machine can understand very well, but communicating that to the patient may be something you still want a human in the loop for. So maybe there's a different kind of a job there for interacting with humans. They take some of the radiologist's job and some of the nurse's job and some of the doctor's job and they bundle them. And that's really an important job, an important task for entrepreneurs, for policymakers, for good managers is to rethink work because we're not going to have a place where everything gets automated. We're far from that kind of a world. Instead, we're seeing a world where some tasks are automated, some are augmented, and humans are still important. So there's a new set of combinations that are needed. The companies that do that effectively are going to harvest the greatest productivity gains. Other questions? Yes, right here. Sorry, you said that humans are important. Sorry, say again. I said you had mentioned to the earlier question that humans are important. How do you deal with adoption of what has been stated on the board in terms of AI? Because I think there's a big concern on the human part as to how adoption is receptive or unreceptive. Well, I'm very excited about the potential of AI to transform work and employment and industries. And so I encourage companies to adopt it, but there's a lot of hesitation because of some of the risks. And in particular, employees may be worried. The most common question I get is, is it going to eat all the jobs? Are we going to be wiped out? And I don't think that's going to happen anytime soon. And the reason is that, as I talked about, it can often be used to augment workers, but it's also the job of managers, entrepreneurs, to when they introduce the tools to do it in a way that they're focusing on how they can make the worker's job more productive rather than replacing them. I think ultimately that creates more value for the company, but it also helps with that organizational adoption. If you go to that radiologist earlier, as one of my colleagues did, and she was a very smart machine learning researcher at MIT, and she said, hey, we can replace radiologists with this machine. Guess what? The hospitals, the radiologists were not dying to adopt that. She realized that a better approach was what we just talked about. Hey, we can help the radiologists do their job better, spend more time communication with patients, more time on higher level abstractions. And the call center example is another one. Some companies tried to completely replace the operators. Others said, hey, we can make your job better, have the customers more happier, so that your job is more pleasant. Guess what? The second company had much more adoption than the first one. So my advice is not just in terms of increasing productivity, but also in terms of speeding adoption. Think about that augmenting, how you can make it an assistant and helper rather than replacing work. Yes, over here. Thank you. You said negotiation. Sorry, I don't have a headset. Go ahead. So negotiation. You said it's low exposure to AI and language learning. But I see in negotiation as negotiator. But when you have a machine learning, it can calculate for you all the possibilities for a good negotiation. So why it's low exposure? Yeah. Well, so when you go into any particular task, it's rapidly evolving. And so one of the things we've done at the World Economic Forum and at Stanford and at my company, Work Helix is we're just constantly updating what kinds of capabilities the machines have. And so all of you as managers, you need to keep up to date on what's doable. I showed you predictive AI and then I showed you generative AI. And there's a different set of tasks that are being affected. And even within generative AI, there are new tasks. It's a constant evolving. So the technology is not standing still. And one of the things you want to do is partner with companies that can help you identify what the opportunities are. That said, I don't think every company needs to always be using the very cutting edge technology at any given time. There's frankly so much low hanging fruit right now that if you simply implement, if something horrible happened and all the technology froze right now and it never advanced, I think we have another 10 years of implementation and of productivity gains just to use what we already have. So my advice to companies is set a priority list, go for some of that low hanging fruit and things like customer service, coding, sales operations, a lot of management applications. You can implement tools right now and get some productivity gains, but then keep your eye on how they can be used in new capabilities like in negotiation. I think we have, is there anybody on this side? I just want to be fair. Okay, looks like the questioning people are on this side. Let's go ahead. I didn't want to be... Okay, we can hear you. Hi. Is the low adoption or the resistance from current job holders is going to some extent hinder the development of AI, you know, and the production of the... So, well, I just, I have a paper coming out looking at adoption across the entire United States that was done in the Corruption with the Census Bureau and it was like single digits, but that data is already out of date because now the number of users of chat... Well, first of all, let's just find a little adoption. How many people in front of me here have used ChatGPT? If you're not raising your hand, yeah, okay. So, within 60 days, 100 million people were using ChatGPT and obviously it's a much bigger number now. So, I think the adoption is not as low as it was a couple of years ago. Now, I think the real challenge, as I mentioned at the beginning, is translating that adoption into productivity. It's one thing to play around with it and write a love poem for your wife like I did for Alina over there or had ChatGPT write it for me. But, you know, which is I guess a productive in a way, but also companies have a lot of opportunities as well. And by the way, you know, Saadi, why don't you come on up here because Saadi has been working with a lot of this stuff. She's got a whole team at the World Economic Forum that's been studying some of these issues as well. So, are there other questions? Yes. Philippe. Just want to know is it the continuation of race against the machine or is it a real disruption? Race against the machine. So, 13, no, 12 years ago, I wrote with a book with Andrew McAfee called Race Against the Machine. And I made an argument very similar to what Saadi and I have been making here, which is that you can use technologies as competitors and try to replace people. And you're going to lose that race or you can use it to compliment people. Race with the machine was what we said. I don't know if it quite works, but the idea is usually to help you go better. And I believe that even more now than I did back then, if that's possible. And that's the advice we've been giving everybody. You know, back to that call center example. One of the reasons that the successful company, Cresta, was able to have their technology work is because they didn't try to do everything. They let the humans do what they were good at and the machines do what they're good at. And it turns out that if you look at all the tasks that a call center operator does or an economist does or a pilot or anybody does, they can be described in a power law. There's some tasks that are very, very common. And then there's a long tail of tasks that hardly ever happen. Machines are very good when you have got a lot of data. That's machine learning needs data. But when there's an imaginary graph behind me right now, when there's a long tail where there's only one or two examples, maybe you've never seen it before. Machines are terrible. Machine learning doesn't work for that. At least not yet. Humans, however, are pretty good at it. So there's a natural division of labor of we are better at exceptions and unusual things and figuring out things. Machines are great when they've seen the example before, maybe a million times before. And that's going to be true for some time. How much time do we have left? Okay, we still have some eight minutes. So other questions, comments? Maybe Eric, I'll just come in on that particular question. We've been doing this report every two years, the future of jobs. And when we started out in 2016, it was all about this concern around the rise of industrial robots and what that was going to do to jobs and factories. And yes, there's been an integration of more of that technology and advanced manufacturing, but it hasn't done that disruption that everybody was deeply concerned about some, but not fully. So there's that element that we have to take into account. It takes time for some of this to get adopted. Second thing, I think we shouldn't forget the huge potential of artificial intelligence in actually personalizing learning and education, which is one of the solutions to helping people move from their current roles that might be declining or that might be disrupted into new roles. Yeah, that's a great point. I want to emphasize that last one about using these to train and educate people. They can be very personalized. Every person can get their own tutor. There's research that shows that if you get personalized education, you get about two standard deviations, faster progress than if you're getting the one-size-fits-all education that most people get today. It used to be much too expensive to give everyone personalized education, but with a large language model and other types of generative AI, you can give very personalized education. And even unintentionally, to give you that call center example, we did some research, not in the paper yet, but I'll give you a preview of some new results. It turned out that sometimes the system went down like any technology does from time to time. And for an economist like me, that was another great experiment because we got to see how do people react when they didn't have access to the system unexpectedly? And I was worried that they would be lost, that they would become so dependent on it. The exact opposite happened. It turned out that the people who had been using the system ended up being better at answering questions than the people who hadn't been. Not quite as good as they were when they had the system live, but they had learned from the system. The system had been coaching them for weeks, months in some cases, and they had learned what some of the right kinds of answers were, and they had retained some of that. And the thing is that the system was giving the answers exactly when they needed it, when it was important to learn them. Previously, they'd gone through training months earlier, or maybe at the end of the week, the manager would look at their transcript and say, oh, you know, you could have said something different here. You could have mentioned a different product. The ability of these tools to give you real-time personalized advice is one of the reasons that I think it's a great tool for training even unintentionally. Other questions? Yeah, right over here. In the jobs transition, there's often a gap between job losses and the new jobs arriving and the skills coming through. Is the clock speed of these AI developments meaning that gap is getting smaller or larger? Both. So one thing that's happening is that the technology is definitely accelerating and it's creating new needs and people have to reskill faster. But there are also two offsetting factors. One, as I just mentioned, it's allowing people to learn new skills faster. It's also allowing with platforms and policymakers better matching and be able to identify who needs, who can move to new areas. Let me give you a concrete example. You know, during COVID, companies around the world laid off millions and millions of people. And then a few, you know, years or months later, they were hiring back some people with some slightly different skills. And it turns out that the people who they were hiring back, they had a new set of skills. But if you look at it, there's a lot of overlap. In many cases, it would be an 80 or 90% overlap. With these tools that go down to the task-based analysis, you can see where those overlaps are. And instead of laying off one person and hiring an entirely different person, which is very expensive, or trying to train someone up from scratch, you can take the first person and say, hey, we only need to teach you another 10 or 20%, and now you've got new skills. You used to pull copper wire. Let's teach you how to do fiber optics. You used to do data science. Let's teach you how to do machine learning. And these transitions are much easier than having somebody learn from scratch. And one thing that's helping make those transitions easier is we have better predictive AI. We have better generative AI. And so both the challenge is greater, but also the tools to address the challenge are much greater. I think we have maybe time for one more question. All right, over here in the back. Yeah, so my question is that since technology is improving at such a rapid rate, and how does the college know or how does the student even know what to study? Because by the time you're out of your college for years later, you are already, you know, what shall I say? What am I looking for? You're not required anymore. Yeah, right. Well, education has to change. I mean, my industry, I showed you all these industries, my industry has been one of the slowest to change and it needs to change faster. And I'll point to Andrew Ang again in the back here. One of the other companies he did, as many of you know, Coursera is a tool for bringing stuff online much more rapidly. He has new courses on generative AI that are coming out. I have a course, some of the stuff that I did at Stanford is now coming out online on Coursera. And people can be lifelong learners and continuously get the skills. They can take nano courses where they learn one very specific skill. They can take broader skills. But the era where you go to education for the first, you know, 10 or 20 years of your life and then you're done and then you just work, that's gone and we need to change our model. Stanford University needs to change and the online courses are part of the solution. Companies can help with this by having continuous education and learning. And of course, all of us ourselves, we have to have a different mindset that we're going to be continuously learning and learning those new skills. Do you want to follow up? You know, I do get it that once you're in a job, you're continuously learning. Learning doesn't stop. I'm talking about the students who are in college, you know, are getting into college and they are in for the next four years and you know, they are saying, okay, do I do this engineering? Do I do that economics? So what do I do? I mean, am I going to be required at all? You do want a foundation and there are some broad skills that are broadly applicable. You know, if you're interested in math and statistics, that's going to be useful for lots of different verticals. If you're great at creative thinking or logical analysis, those are skills and techniques you can apply to lots of different areas. But then you're going to have to continuously learn on top of that as well. Don't think you're going to learn just one narrow craft and be done. It's going to be something where you continuously learn. Well, let me just wrap up by saying, I said that 2024 is going to be the year of productivity where AI gets a body and takes action and starts transforming each of your companies and countries. That's only going to happen if you take action. And I really want to inspire you to make an effort to do an analysis of the opportunities in your company. You can do the task-based analysis that I described and see what kind of roadmap you have for getting some of that low-hanging fruit. If you do that, I think you're going to find surprising productivity gains. We found double-digit productivity gains within a few months in some of these applications. I'm pretty confident that there are applications like that in each of your companies and countries as well. And so this is not a good time to just sit back and wait. This is a time to start using the technology and beginning to harvest some of those gains. And I wish you the best of luck in doing that.