What no one tells you about the next wave of AI jobs

All right. Hi everyone. Thank you so much for joining this virtual event series. We have Mark, Carrie, and Arun ready to discuss lots of topics regarding MicroOne and the future of the social economy. So thank you so much for coming, and I'm passing it off to Mark to get the event started.

Thank you so much, Sydney.

Welcome.

This is our first series that we're launching as the summer is wrapping up towards the end. My name is Mark Esposito. I work as a chief economist at MicroOne, and I'm also an academic at Harvard Kennedy School focusing on the intersection between policy and technology. I could have not wished for a better start. Two extremely prominent voices in the conversation around AI representing not only the industry and academia but also a lot of convergence space. Let me do the proper introduction to today, and then we'll get the conversation off the ground. So Arun Sundarajan is a professor at NYU Stern School of Business. He's also the co-chair of the World Economic Forum's Global Future Council on Data Frontiers. And Arun has been also part of the AI Global Alliance at the World Economic Forum, which is the group that's been studying governance. I've been lucky to be part of the same group he has been part of, therefore kind of a prominent voice across multiple sides of the conversation around AI governance, role the business should have on on this. On the other side, Carey Maya is the VP and Chief Technology and Innovation Officer at Capgemini Invent. Carey is also, about a year ago, joined the World Economic Forum as an executive fellow. He was in charge of the Convergent Technology Initiative that he has spearheaded and pioneered at the forum. And I equally been lucky to be part of Carey's initiative at the forum. So by pure coincidence, we were in two working groups, one at the global AI governance group with Arun and one with the convergent technology with Kerry. And I think this has created a common space for us to have a conversation today. And I'm very pleased, gentlemen, to be hosting you today. Look, I think the conversation is almost like a hot potato these days, right? If we talk about AI, we do talk a lot about the development we see out there in the industry. We're seeing generative AI taking a significant role in our conversation globally today. If you're looking at where conversations are going from governments to boards, right? It is about AI, AI adoption. But one story that keeps on being part of it is whether this adoption of technology is coming at the expenses of jobs. And so at the same time we're talking about AI, we are inherently talking about jobs as well. I'm going to start from you and from the work you have done. There is a conversation that you were having with us when the three of us met prior, right? You were talking about the fact that, yeah, technology is great and we can talk about technology a lot about the productivity side. But you mentioned something about the intellectual debt, which is a story that I think many of us would like to know more about it. Just give us your sense about where you see this global economy going when we're looking at the impact on jobs, but also what you meant by intellectual debt. And welcome to this podcast or to this webinar, whatever this.

Okay, well, thank you, Mark. Thank you, Sydney. I'm delighted to be joining. Thank you to MicroOne for putting together this event. I have recently started using MicroOne and I'm enjoying the experience. I guess, you know, stepping back, it's always dangerous when you give a professor such a broad range of topics to talk about. You know, I could take the whole 45 minutes, but I'll try and keep it brief. We may come back to, you know, sort of the questions that are arising around the productivity gains from AI. It seems to be sort of a mixed bag. A lot of people are saying that they don't see the productivity gains, even though they see a lot of the investment. And, you know, that's not surprising. In some ways, because at least for the latest generation of AI, I mean, we've been using AI in business for 30 years now, or using machine learning-based AI in business for 30 years now, AI even longer. But this sort of the last 3 years of generative AI, and, you know, this, these sort of new transformer-based capabilities, it's still very early days. And in the early days of of, you know, the adoption of a new revolutionary technology like this, the focus is often on speeding up things while keeping the basic processes the same. So you sort of take out the old machine, you plug in the new machine, and you see sort of some task productivity gains. People are doing things faster, perhaps doing things more efficiently, but because of adjustment costs there are often no broader sort of either labor or total factor productivity gains. Those come later. They come when you're, you know, you've reached the point where you've understood the technology well enough to not just take out the existing, speed up the existing way of doing things, but to completely reimagine the process. And, you know, there are plenty of historical examples of this from you know, warehouse robots that move shelves to people rather than trying to build robots that go to shelves. That's a reimagining. But you can go back more than 100 years to the transition from steam and steam power or waterwheel power to the internal combustion engine. And again, early on, there were no productivity gains because they were just taking out one power source and putting in another. Once we realized that everybody could have their own personal power source, then you could organize the work around not being close to the power, but, or being close to the river, but around like, you know, sort of division of labor around scientific management. That's when the productivity gains came. So any news about a lack of productivity gains, I wouldn't worry too much. So the productivity story is not something that I think is resolved or will be resolved for the next 5 to 10 years. The intellectual death story is the one that I think is not being told enough. And is perhaps sort of more worrying. And what I mean by that is, you know, as we went from expert systems to machine learning to deep learning, we got increasingly comfortable with an AI environment where we don't understand why the machine is doing what it's doing. Explainability has gone away, transparency has gone away. You know, we all know the stats. I mean, like, you know, any of these large language models often have over a trillion parameters. There's absolutely no way of understanding how they're doing what they're doing. And what ends up happening there is that humans start to progressively cede understanding to the machine, allowing the machine to make decisions where the human doesn't know why these decisions are being made. They just sort of work. Right? I mean, so we talked about supply chain management, which was math-based 30 years ago, and today is deep learning based. And so supply chain management systems, the people who are managing them don't understand why they're making the decisions they are. And as a consequence, it's sort of like borrowing from the future. You get these gains today, but if something changes radically in the future, you've sort of lost the knowledge that you would have gained by doing it, doing the work in a more sort of transparent way. And, you know, there's often a parallel drawn, like, you know, by Jonathan Zittrain, sort of from the Berkman Center, between this kind of intellectual debt in AI and intellectual debt in the pharma industry, where, you know, we often release pharmaceutical drugs without understanding how they work. They just work and we establish it through a randomized controlled trial. But in some ways, we're borrowing good health today from something happening because we don't understand how things work. And, you know, because the black box deep learning systems simply work better than the other models by and large across a broader and broader range of tasks, I think we're going to be incurring more and more of this intellectual debt. And for a company that is chasing like, you know, the best possible outcomes in the short term, there's no regulatory mechanism that will pull them back in favor of transparency because then they'll fall behind their competitors. But societally, we are just sort of ceding in some ways knowledge and understanding to the machine. And like, you know, the debt part comes from like, you know, borrowing from the future.

That's right. This could not be a better start for a number of reasons. One is because I think you gave us an overview that also resembles a lot what we have done with the financial system when we started to borrow from the future. And we know where that eventually went well and didn't go too well at the same time. So I think it's great for us to understand exactly where the very same opportunity that we had in having access to resources now became in some circumstances spirals of debt that we have never really resolved. I'm going to move over to Carrie on this. Carrie, look, you work for a large company, technology company, and a large technology company, but you always been a top leader in the work you have done. So how do you navigate this conversation that we're having today, Harry, on the AI becoming definitely part of the conversation, how you're dealing with many of the clients that Capgemini is dealing with? At the same time, there's almost like a call for a moral compass on the fact that just because technology works, it doesn't mean it's the right thing for us to engage fully. And therefore, the oversight mechanism— just give us a sense of where you see this conversation makes sense in your side of the world.

So let me just preface this answer by saying that all thoughts that I express over here are my own and don't represent those of my firm. I'll definitely use a lot of the experience that we've been having. And the first thing that I'd say is you need to have a severe sense of pragmatism when it comes to actually deploying these systems. What these systems actually provide are very, very powerful tools. But at the same time, when you work in any organization which is large and has got, I don't know, 20, 30, 50 years of knowledge and know-how, the main problem that they have is I'm exposing all of this IP, I'm exposing all of this knowledge to the tool providers. And my role as a solution provider is actually going to get diluted because these things have a tendency to learn. So initially what comes in to actually have some kind of delta in terms of productivity can very quickly become a situation where a tool provider is now becoming a competitor because in 2 to 3 years, once they've kind of done this across, you know, different categories of different industries in the same area that you operate in, they're learning. And now what's stopping them from being able to actually go ahead and provide technology services or consulting services or whatever it is to the same end clients that you're actually trying to work with? Now, this is something which needs to be understood because whilst we do have, I don't know, 700 million users or something around that much of one of the most popular LLMs out there, the reality is when you start actually looking at deployment within an organization, you get hit with a lot of different kind of blocks. So on one side, this is not cheap. If you're using it at scale, it's expensive. So your primary conversation is not going to be with your CTO, it's going to be with your CFO. Secondly, you're not just looking at, you know, giving random access to everyone. There is a lot of work that needs to be done in terms of budgets, in terms of controls, you know, who's getting access, what are they actually using for it. If you're working at the enterprise level, there's SLA contracts that come into play. So that's SLA 1, 2, 3, fee. And all of these things actually affect a lot of the adoption. So whilst today there's a lot of noise about saying that, you know, I think there's a number which is around 10% of the firms, large firms worldwide, are actually using this. The reality is if you want to actually have these kind of tools be deployed at a large enough scale to see that famous productivity impact that you're looking for, you're going to actually have to deal with these very, very tangible roadblocks that you're going to come across. The second thing is today, I think the flavor of the week is all about agents. Agents are great. They're quite simplistic in certain ways. They're essentially software wrapped around an AI model, and they do things. While the LLM can kind of like infer or make sense of, you know, a question or a prompt that's given to it, if you want to transition from getting an answer to actually executing upon it, you're essentially right. One agent or a series of agents that then take this information and execute certain things on top of it. Increasingly, the main thing that makes an agent versatile or powerful is going to be in terms of tool use. Which tools can it connect to? And whilst a lot of people are kind of going around and saying, oh, agents are going to, you know, really give an extra nudge in terms of the productivity improvement that I'm looking for, The reality is it's kind of like a software nightmare. It's like a place where this can— easy for you to be able to get malicious code. There's a lot of tools which are out there, and a lot of the ways that agents use tools is they connect to something on GitHub. There's little to no control today in terms of someone actually going and writing a malicious piece of code, which then gets used by someone. The agent is still doing its job, but behind the scenes, there's a penetration attack that's actually happening that can lead to malware, it can lead to system compromise, et cetera, et cetera. Et cetera. So your main challenges today in a, let's say, 3-year span moving forward in terms of adoption is not in terms of how much this is going to cost. That's definitely going to be, that's always there. But it's also the fact that your CISO is going to call you up the moment you start kind of using these tools. And so there's, the reason I'm kind of stating this is because it's important to get like a bit of a pragmatic understanding of what's entailed in terms of adapting to these tools, using them. And then of course, the main reason we are having this conversation in terms of what does that actually mean for employability in terms of like, you know, skill reswitching or skill updating, etc., etc.

All right, thanks for this. And while I'm going to go the conversation for a few more minutes, I mean, I see over 120 people connecting, watching us right now. And please make sure you're using the Q&A button to place your question, Kerry. Arun, they come not necessarily to spend too much time with me, but mostly to engage with the Micro One community. And so please use that to, to allow me to facilitate some of your questions. I know you're not a shy community at all, and so I look forward to eventually inject some of that input into our conversation. Let me do maybe another round of conversation between Arun and Keren, then let's see how the questions are, are, are, are like shaping up. One of the you have a lot of experience working with organizations, right? And so one of the things that we often see is that the AI labs that are generating these technologies, they are definitely operating at the pace that is extremely fast. So you can call it exponential, you can call it neck-breaking, but we know that the technologies come in at an extremely fast pace. But we also know, a bit mirroring and echoing what also Carrie said, many organizations work in workflows and the adoption is not always as fast. So you could argue that adoption is at best incremental when the AI labs are operating exponentially, right? So what is in your view, right, the best way for this conversation to be mitigated into a working space where we're not necessarily getting to the point where we're going to be completely cut out or we're going to keep on chasing without necessarily knowing Therefore, back to your first point, it's like we keep on adopting and we are borrowing from the future, but we're shifting a cost somewhere that we don't know yet where, right? So how do you see this in the, in the conversation you're having with the many company you're dealing with, with the work you're doing at the forum? And then let's, let's have a dialogue, folks, right?

This, Arun, do you mind if I go first and on this?

Uh, not at all. Go ahead.

So I think what you've tapped upon is the fact that diffusion is not even. Right. When these tools kind of get released, obviously there are certain functions, there are certain parts of an organization that start adapting it first. And this leads to collaboration and coordination issues. The biggest problem that we have, and this is particularly true in large firms, is you start getting replicas. There'll be someone sitting in one division or in a different geography who will build an AI tool using all these new agents and MCP servers and whatever new model that's out there. And feel very proud of themselves, not tell anyone because they want to protect their competitive moat. And 2 months later, you'll find out that there's someone else at the other end of the earth, sometimes even within the same division, who's done exactly the same thing. So there's a lot of double work that kind of actually occurs. I think if you are an organization as a leader— so let me talk about it from a top-down perspective— you have to actually play a balance in which on one side you want to be able to give some kind of leeway to the local teams to take ownership, to do some experimentation, because that's how you learn, right? It's only when you get your hands a little bit dirty and start actually working with this that you come up with your own opinion. And that's probably the most important thing to have. Like, don't regurgitate what someone else is saying. If they are saying something which is very bullish or they're saying something which is very alarmist, see which company board they're on or which company they've invested in. Because it's probably true that they won't be able to push that stock price up. So you have to take that with a big pinch of salt, develop your own opinion. Secondly, as a leader, whilst you allow these teams or you kind of give the mandate that, hey, go ahead and figure this out, whether it turns in terms of build, partner, or buy, at the same time, I think top-level leadership requires not knowing how this is going to have a productivity increase, which it will, But you know, which technology doesn't? But more importantly, and how is this actually changing your sector, your industry, and your firm more specifically? There's that old saying, you know, don't push the puck to where it is, but focus on where it's going. And I think with firms, this is very important because you've now got 3 years of adoption data. You've gone to all the conferences, you've had all the dialogues, you've seen YouTube video and LinkedIn posts talking about this fact that This is some kind of general purpose technology. It can be used for everything from healthcare to production to education to whatever it is. So you're dealing with a general purpose technology in some sort. And if you're going to be using a general purpose technology or you're immersed in a world in which, you know, there's a general purpose technology getting released like electricity, very quickly being able to understand what is the new nature of the firm, to to, to CodeCourse, that's actually more important because this is a tool. Yes, it does things and it does some things extremely well, but how is that fundamentally changing the way your organization works? If it is alleviating certain constraints which exist within your value chain, well, where is the new constraint that's actually going to get created and how do you think about that more strategically? Right. That those are kind of like the questions which you need to talk about. And then you balance it by giving ownership and a little bit of independence to the teams because they are trying to do it at a task level perspective. They're trying to optimize for certain tasks. They're trying to make— get better at their jobs. You want them to be able to do that. But at the same time, you have to be able to take those recommendations, take those suggestions, you know, take all of that information in and still be able to decide where your entire firm is kind of transitioning to and how all of this bottom-up work that's actually happening is allowing you to turn that vision that you've got where things are moving into an actual reality.

No, I—

this is—

I will go ahead, please.

Yeah, yeah, you don't need me, guys. This is a conversation. Go ahead.

Yeah, I'd love to hear your story. You know, I mean, sort of reinforcing Carrie's point about experimentation. We're in an active experimentation phase and, you know, there is a reticence, you know, like technology organizations within companies are reticent to, you know, sort of seed control to end users. I mean, this is a battle that's been going on for decades, right? But the, you know, it's through that experimentation, you know, by the people who know what your organization is doing. It's from that that the reimagining of how things could be radically different and the big productivity gains will come. And it's from that that your employees also start to sort of learn how to ask AI the right question and what to do themselves and what to delegate to the machine, because these are the two things that the employee of the future has to be good at. I also completely agree with you, Carrie, on the, like, you know, sort of looking ahead to and not getting caught up too much in how I'm doing what I'm doing. I mean, certainly you want to be reimagining what you're doing, but you have to be sensitive to fundamental changes in the business models of your industry that might be being caused by AI, right? And, you know, in some ways, I look at automation in retail, for example, or like, you know, sort of the workforce impact of digital technologies in retail. And a lot of people look at this through the lens of, you know, sort of human replacement technologies like, you know, the self-checkout machines, so on. But, you know, many years before those came, the fundamental changes that technology was causing in retail didn't have to do with changing physical world retail, but creating an entirely new e-commerce model. Where the mix of what the machine did and the human did changed. You didn't need people stocking things on shelves. You didn't need people checking things out, but you needed people to deliver, to put things into boxes. And so how does AI change the business models of my industry and the mix of things that machines and humans will do is a much broader question than simply looking at it through the lens of how you are doing things now or in the future. One of the things that I think a lot of organizations should watch out for is an excessive desire to chase the leading edge. And what I mean by that, to your point, Mark, we're seeing these new model releases every few months, right? And they're always better on some metric that typically is computer science-y and has to do with, well, it can pass the MCAT now, it can, you know, sort of play this game better. It can sort of solve these logic problems, these math problems. But every time there's a model shift, there's also potential unpredictability that comes into the system. And, you know, there's work I've done with my colleagues, Jiao Sadok and Ruben Li, where we've been studying how good AI is at assessing trustworthiness in a human. Like, you know, how is it good at— is it good at sort of distinguishing between the client you should like give the refund to and the client that you shouldn't? You know, the person who's like, you know, genuinely got a problem and the person who's trying to rip you off. And we've found that as models sometimes improve, the trusting behavior changes in the opposite direction. And that's not part of the metrics that the computer scientists are measuring, but that's part of what is really going to be central to your organization. And so, you know, once you've found a model that is sitting— this is a platform, right? And so you're building on top of some underlying LLM or like, you know, generative model. You know, I'd be cautious about saying, well, GPT-5 is out, time to sort of like, you know, sort of unplug 4.0 and plug in 5.0, but to sort of take it in a more measured way. And the final thing that I'm having most conversations, a lot of conversations with organizations about, and this is to your early point, Mark, and also to what you raised, Gary, is about how is what humans— how are the responsibilities of what I assign to humans in my organization going to evolve? And while in the short term I might be thinking about how do I get people to experiment enough so that they sort of become good at using AI, asking the questions and so on, there's a much broader, longer-term issue, which is that a significant fraction of your workforce is going to transition to a new role. Within your organization. Potentially, a significant fraction of your workforce is going to transition out. And so an organization has to think about what are my responsibilities in terms of adequately preparing them for this massive transition that I've been talking about for 10 years as being in the future, but is now in the present. And I don't think that there is a more important sort of society-wide issue that corporate America is collectively on the verge of, then this, you know, millions, maybe tens of millions of people are either going to leave their companies and go on to something new or transition into new roles. And it's not clear that we have the infrastructure outside of the organizations to prepare them for a sort of a transition with dignity, because reskilling doesn't do it. You have to like wrap the scaffolding around and like, you know, give them abilities to discover what they're good at. And if they're transitioning out, find their next role. And so I think these are like, you know, sort of experimentation, not chasing the leading edge, looking out for business model changes. And most importantly to me, like, you know, sort of managing the workforce transition and preparing your executive suite for that. I can't think of a more important thing.

The only probably— point that I would state to what Arun says. I think Arun is just a nicer guy than I am. This is also—

well, that remains to be seen, right?

It's still early days. It's also your own vicarious liability, right? And I'll give you an example. So we were speaking about how the puck is moving. If you are going to be seeing more and more addiction to AI, more and more use of AI, Then very quickly you realize, okay, fine, I don't want to give all this information to some provider, some hyperscaler where this model is actually sitting. I want to probably think about how I can bring the computation or the intelligence to where the data is. This is a real problem. There is so much information and data that we just throw away because it's just too expensive to store it. So on one side, you've got this problem in which people are saying we're running out of actual internet data. To train these models to go from GPT-5 to 5.5 to 6 to whatever else is out there. But on the other side, you've got this space in which there's so much information that we've got that we don't use it just because it's too expensive to store. Data movement costs money, and it's probably related to IP or privacy or whatever else it is. So very quickly you start realizing, huh, if I am someone who's thinking about how I can run, you know, I still want to ride this wave of, of AI tech adoption because I want to be a contender and not a spectator to it. Maybe what I need to do is take ownership of I understand this fact and start moving into a realm in which you think about how do, how do people actually move competition to where the data is. And very quickly you're going to find yourself in the world of Edge AI increasingly. And I have a bias in saying this because I do have a unit that works specifically on Edge AI. We actually see this is something in which, yes, you've got all these models which are coming out. Number one, select the right model. What are you using? Number two, own it, right? Return to on-prem, you could say, to a certain extent, which is increasing in a lot of companies. And number three, what role do I play either from a technical perspective, from a product management perspective, in order to actually build the solutions where the AI can be used at the point at which it's, it's being made? You can also have that in terms of—

in many ways, like, you know, that problem is coming to the individual as well, right?

Correct. But I feel that that's something that the individual should have ownership on because you cannot let someone else guide your career. Sure, they can offer reference points, they can offer some kind of a supporting arm, but at the end of the day— and this is where I wanted to say that I disagree with you— is the fact that there needs to be a certain sense of ownership. There is enough and more information out there in the world. There are talks like this that are out there that you can access freely. How are you using all that information for you to take ownership about how you're going to deal with this rather than, you know, kind of being a little bit more a victim side of it? I think that's the point.

And, you know, I You know, I certainly encourage people to sort of like, you know, sort of embrace a sense of agency and, um, you know, become more entrepreneurial. I think the number one piece of advice I give entering students now is like, you know, take entrepreneurship courses, experiment with side projects, go for hackathons, design sprints. Even if you don't plan to be an entrepreneur, you've got to learn entrepreneurial because this is like, you know, the future of work. But that's a slice of the workforce, and especially sort of early-stage workforce. I think that there's sort of a balance between leaving it all to the individual and putting in place some sort of institutional support for the individuals who, because like, you know, sort of in academic institutions, in the consulting world, in the banking world, in Silicon Valley, they're sort of a biased sample of people who, you know, have self-selected into this world because of like high levels of accomplishment, intelligence, entrepreneurial, but, you know, a lot of the workforce needs help. I mean, like, you know, they can be determined to try and make something of like, you know, this transition, because I've seen this among people who were displaced by the manufacturing automation that has sort of rolled out, and there's certainly no lack of motivation for them. They just don't have the resources or the conduits to be able to say, okay, here's how I get to what's next. I can't just sort of go online and become a Python programmer. I don't know, you know, am I going to be even good at it? But I thought you were getting to a different point, which you and I have discussed sort of in the past, actually, like, you know, the first time we met about individuals taking ownership not just of their future, but of their intellectual capital.

Yeah.

You know, when you were talking about, you know, when you gave your disclaimer that like, you know, this is my opinion, not that of my company, I was imagining in a future in which my digital twin is talking and saying, these are my opinions and not that of my creator, Professor Sundararajan. But I do worry also, like, you know, that's a— drift into this intellectual debt world. There's also a ceding of the ownership of our human capital that is underway. And it's not yet clear to me, like, you know, how that is going to shake out because, you know, we're kind of running out of ways in which you can train AI at scale. And so a lot of the important training data is going to come from individuals now. And individuals might be making the trade-off between, like, you know, do I want to improve my productivity? Versus do I want to, like, you know, create something that is going to make me less important to the organization because embedded in it is my human capital. And to me, a smart organization is sort of managing that trade-off as well. Mark, you've got the look of like, you know, we have to go to the audience.

That's right. And I have to say, this is such an incredible, incredibly easy, you know, job that I have right now to the point I could really be replaced easily because you guys have an incredible chemistry in the conversation and the dialogue is there, right?

Let me, let me just give a preface to this because I know we are time-bound, but it's an interesting idea. And Arun and I discussed this. We came at it from two completely different schools of thought. I never met him before, I hadn't read his work, but the concept that I was starting to think about is artificial intelligence is not really artificial and not really intelligent. There is a lot of training that has to be given to it and especially when you come to real-world scenarios where, yes, there is now a tool that can actually do certain tasks or subtasks that a person is actually, you know, doing. What if you— the fact that you're actually training an AI to do that allows for a kind of a stipend to be paid back to you as it starts kind of like cannibalizing that task that you were doing? And since you were a trainer for it, the productivity gains that it creates gives some kind of like trickle-down economics to the person that actually helped with that training over the life or the span of that AI actually existing.

Which is, uh, Kari, which I think is, is what, uh, you would call it a universal basic dividend rather than income, right? Because it's almost like the idea— I know, I know, I know it's not your words, but I'm trying to say that's why I think it's interesting, the idea that we're rethinking also fundamentally the infrastructure of economic value. When we are becoming agentic in who generates the value simply by the nature of the conversation we're having. So guys, believe it or not, we are just a few minutes before I have to wrap this up, and I'd like to bring it over.

This sort of goes back to Karl Marx in some sense, that, you know, the distribution of ownership of capital in society is what drives inequality. And in some ways, humans have always held their human capital You know, if we go down a path where there's excessive centralization of capital ownership over time, that not just sort of like, you know, leads to greater inequality, but also starts to sort of challenge the motivations that humans have to sort of build new human capital. So like, you know, when I was studying the sharing economy 10 years ago, the promise that I saw was that it would decentralize sort of structural capital ownership, creating millions of micro businesses. And, you know, the results have sort of been a mixed bag depending on, you know, which industry you look at. But it's sort of the same concern here, which is like, you know, sort of how do we strike that good balance between the efficient centralization of different kinds of productive capital and the, like the necessary decentralization of it So that there is some equity to some extent in how, like, you know, the returns from the economy are shared, right? Yeah, there are, there are a bunch of questions, and that's right.

And I mean, I, I don't think, guys, we can, can cover all of them. I, I think to just to link it a bit to, uh, to the kind of experience I had so far with MicroOne is a company that is now fundamentally engaged into recruitment with the support of AI. But similar to what you were saying before, and both of you kind of COVID it in a different way, it's not about recruitment for the sake of more efficiency in recruitment because that's easy to measure is rethinking and reframing the narrative about recruitment, which is I think opening up and I can see this in some of the questions. So I hope that by summarizing a couple of things you mentioned that I can also address some of this question is first of all, expanding the access to this technology to a wider number of people that are not necessarily contemplated by the traditional transformation. A little bit of what I really was saying before are people that would naturally not even be able to be reskilled. And back to a conversation that you mentioned, Carrie, when we also had alignment call, people that never even have a chance to be part of the conversation because they don't even qualify for entry-level jobs, right? And so I think there's a space for people to eventually start expanding the idea of training the data from profile that usually are not part of the conversation. And so yes, academia, you can say, has been adopted to some degree because of the integration of online learning. Banking has been a pioneer in many ways, consulting. But what if you're coming from fields that were not necessarily part of the first wave of transformation? I think to the point where we see a lot of research in Micro One is to see people that will come from an entirely different background that now are integrating into training the data of fields that were not necessarily digitizable before. But this is one side of the story. The other side of the story that I love in the conversation that we had is ownership on one side and entrepreneurial mindset is critical. But to your point, right, there's an institution and actually it's mirroring There's an institutional response that has to come either in redefining where the equity is going to generate some form of offsetting mechanism, or in what do you— and that, Carrie, that's kind of where you kind of instigated the conversation before. And Arun, you mentioned before, we have to take into account the fact that some people might not be in a position, regardless of their motivation, to do that. And so there's a question from the chat that I think it kind of fits into where we can maybe have a quick exchange. So imagine both from a perspective of being in the industry, Kerrie, or being in the advisory capacity, Arun, or both of you, right? You are asking, a company asking, say, how can I navigate this displacement of labor that is coming from the fact that the dependency is still on the economic thinking, right? From an economic perspective, yes, AI is improving the efficiency. But I want to actually expand my measurement of value value differently from economic thinking so that I can rethink about the job differently. How will you advise them to navigate these challenges? A company that might be facing a decision that displaces a job, but maybe they want to basically own the conversation before it's becoming inevitable. What would you say to a company that comes up with a story like this?

Well, it's, it's, it's sort of a tough one, right? Because, uh, like the lack of predictability in what AI is going to be capable of in a few years. And anyone who says they know what AI is going to be capable of in 2 years doesn't understand the technology, right? And so, you know, it sort of starts to, it starts to make a sort of like, you know, here are the roles that I'm preparing people for kind of workforce strategy. A challenging one because, you know, like all of the organizations that said we need more data scientists, we need more sort of like, you know, people who can do simple programming 5 years ago are realizing that a lot of that data science and a lot of that programming is being done by AI now and the people sort of aren't needed. So to me, like, you know, I tell them to sort of first, you know, start thinking about this and start the conversation in the organization. That it be open, that there be a mindset of acceptance that this transition is coming and that everybody should be thinking about it and coming up with their plan and then start to think about the scaffolding. You know, the skills can be acquired. You know, I've been a university professor my entire career. You know, we teach people things, but there's a whole bundle of other things that are wrapped around it. Right of passage, networking, credentialing, mentoring, critical thinking, learning how to learn. So think about how am I going to wrap those things around my workforce's transition so that they can understand what they're good at and they can sort of come up with the networks and the capabilities. So how do you sort of wrap this scaffolding around? And I'll stop there, but I've expanded on this sort of some detail in a Brookings article that I'm just going to pop into the webinar chat if anybody sort of wants to dig a little deeper.

Great.

And actually interesting, Diana asked two questions, but somehow in the way you address this question, you covered that one on the unnecessary urgency to think about the next 2, 3 years in a way that we're thinking about predictive modeling or technology. We can do that, right? There is a nonlinear development in many ways to the— for any predictive form based on historical data will likely be sufficient to understand it. And second, the idea that really you want to, to build the scaffolding, right? It's the, the cultural capacity, the strategic capacity before it becomes a technology output out there. And I think this is great. Carrie, on your side, I mean, we can cover any of the part of the conversation that we have, but we are also almost like on closing remarks, right? On this displacement kind of challenges, but, but connecting it to the, the flow of the conversation we had so far.

Sorry, what's the question?

So basically, the question that initiated this was like, how would you advise an organization to deal with this sense of displacement, right?

So firstly, I'd add more nuance to what was just said. Um, who are we talking to? Is it a private organization? Is it a publicly listed organization, an NGO, a startup, This is very important because your only KPI with any innovation movement is going to be based on cash. Any innovation has to have a dollar value amount attached to it at the end of that row. And being able to understand who it is you're talking to, what is their primary KPI, which will come down to cash at some point of time, is very important because that helps you understand what's the scope of what you're trying to do. Secondly, you have to understand, are they trying to do something which is like a systems-level change, or is it something just in terms of increasing the productivity or the output of a certain unit or a certain department? So without actually having a clear understanding about this, there's no point going forward because you need to actually see how is the buy-in happening, why are they trying to do this, why are they trying to essentially pivot. With the use of this technology. And then based on that, you would have something which is very staged. On one side, you're dealing with technical burden. This stuff is hard to understand. It is even harder to integrate, right? If you're trying to integrate this new technology into your existing workflows, your existing OT or IT infrastructure, that is a serious commitment and it is going to take a long time. And that's just the operational aspect of it. Once you've done that, then you have to move to the people. Who are the people who are going to use this? How do they need to think about retraining? Every job is part of a set of tasks that falls into a value chain. If the AI is coming and cannibalizing or augmenting or changing certain tasks or subtasks, what are the other tasks where you require the person to, to continue to, to operate? And how does the people and the process aspect come together to create what I mentioned for the new product? Right. What is the new face of the organization? So I would actually go with massaging of these nuances to get very, very granular.

Granular.

Yes. That's going to make it much more pragmatic because, as I mentioned, every change that you're going to put inside over here is going to cost a lot. And that J-curve, when that ROI starts getting positive, if you can hit it within 3 years, that's a success. Yeah, but I think, I think you say something that is so interesting, right?

First of all, it's the granularity of the conversation so that it doesn't become just a general distributed narrative where people are going to be benchmarked against something that is too far from the nature of their life. I think this is critical. The fact that we are up for and need to absorb the cost, at least for a foreseeable future, before this can look in something that might be productive. But this, and it came from both of your, your, your, uh, like, inputs. There is a, an investment on the social capital that is not based only on, on conforming, but fundamentally transforming. And I guess, and, and I love to, to have our, our closing remark. It looks to me that we are about to almost philosophically re-engage the narrative about what is labor, after a dependence on the industrial policy that has been dominating for decades, right? And so if we do that, then probably this is still a transformation that will not come without a cost, but we'll be much more aware of the cost and what is being built out of it. I'd love for you folks to give me your final, like, closing comments about, you know, where you'd like you know, our narrative to keep on being shared with other— let's say, you know, we're building a bit of a brick of knowledge in our conversation when we're doing this kind of events, right? Where do you think we should look at the next evolution of the conversation? What would you like as a line of engagement? Where should companies eventually spend more of their time, resources? Just give us your, your final closing advice.

I wouldn't talk about it from the perspective of the companies at this point because we're talking to people who are looking at, you know, this AI adaptation in terms of individuals. My closing remark would be to channel my inner Richard Feynman: there's plenty of room at the edge, right? As we have more and more people using artificial intelligence, the edge use cases, the ability to move intelligence to places where it's not there, that's going to open up pockets of opportunities. And there's going to be a general migration. We're already seeing this with a lot of the conversations that I have with clients and like how people are talking about it. There's a serious return to on-prem, a very large, great amount of thinking in terms of like return to analog in some cases. How do you write, do inferencing with analog AI chips, for example? That's very interesting. And then secondly, think about it from the perspective that if everyone starts adopting AI, Well, yes, AI sucks up a lot of the oxygen in the room today, but there's many other technologies, right? Like the work that I did with the World Economic Forum is all about how different technologies come together. AI was just one of them. There's synthetic, there's robotic, there's all these other things. Now, definitely there's an acceleration effect which AI does in certain functions of these other technologies. But these technologies by themselves have got other things to do. So if you're interested in technology, whether you come at it from a technical perspective or whether you come at it from a business perspective, there's a lot of other spaces where you can divert your attention rather than myopically getting bombarded with a lot of the AI news and kind of feeling that, oh my God, this is a lot of dismay that's happening. So that would be my best guess.

I'm similarly optimistic. You know, I think that there's, you know, if you think about the turn of the 20th century and like, you know, the enormity of change that was happening then, you know, people had telephones, they were able to speak across a distance, they had automobiles for the first time, they had electricity in their homes. Yes, AI is revolutionary, but I think, you know, societies tend to sort of underestimate the impact and importance of the technologies of their time. They're actually thinking about just how incredibly transformative the world— I mean, things must have been for people who are suddenly seeing all these changes in their lives, like, you know, 100, 120 years ago. So my belief is that the progress in the economy and the growth of the economy is driven by chasing unfulfilled human aspirations. And there's no dearth of those. You know, like 200 years ago, we had no healthcare industry, unfulfilled. Now, like, you know, it's a double-digit percentage of employment in the United States. We had no tourism industry. And so like, what will people do is often a question that can be answered by looking to, you know, what are the unfulfilled human aspirations? And where is there going to be growth? So there's no doubt in my mind that there will be radical expansion of the economy, but also no doubt that the next 10 years are going to be tough. You know, there's a period of adjustment and, you know, you have to sort of— we've already seen entry-level jobs get hit. I think that will continue. And then there'll be a backlash because companies will say, well, we need people with 5 years of experience and there's nobody with 5 years of experience. We thought we could hire from someone else, but they haven't been hiring either. And so there's no equilibrium there. So you got to steel yourself if you're job hunting or transitioning over the next 10 years. But like, you know, sort of once we settle, you know, I'm optimistic, like, you know, the glass is more than half full. And, you know, this will have massive expansionary effects over time.

Carrie and Arun, look, I just personally really enjoy the conversation, right? This is a conversation that if we had another hour, we can fill it up with even more. I'm happy that we were given by MicroOne the opportunity to start this series and we'll keep on developing it moving forward. But it will be remembered as our inaugural kickoff was the two of you, right? Really setting the stage on this level conversation. And I think this is where a company that is in the AI space definitely is busy day by day on the product side, but is also the ability for this company to step back and say, well, what is the top leadership required for us to navigate this? Because this is not just— this is not an add-on, right? This is not peripheral. This is transformational to the point that it requires at least an engagement that is more than just the efficiency on the last reports metrics. So thank you so much for having been the guest of this first episode. Look, we had capped about 120 throughout the entire session, right? And it's just now that we went a little bit over time that I think people were probably allocating time for what they have to do next. But it's been amazing that we have so many people. And I know we have I haven't been able to answer all the questions, but hopefully you gave enough inputs, right, for people to keep on getting curious about this. Thank you for being with me. We wish you a great— so if you're in Europe, Terry, enjoy the evening. Rune, if you are in my same time zone on the East Coast, enjoy the rest of the afternoon. And Sidney, thank you for having set this up for us and for having been our host today. From Mark, all the best, and thank you very much. Bye-bye now.

Thank you.

Bye.