Preventing Jobless Growth
21 Jan 2026 08:15h - 09:00h
Preventing Jobless Growth
Session at a glance
Summary
This panel discussion at Davos focused on preventing “jobless growth” – the challenge of achieving productivity gains from AI and technology while maintaining employment opportunities. Erik Brynjolfsson, the moderator, framed the issue as two sides of the same coin: while AI can dramatically boost productivity (the numerator in the productivity equation), it risks reducing employment (the denominator). His research found that workers aged 22-26 in AI-exposed occupations experienced a 13% decline in employment, though those using AI for augmentation rather than automation saw employment growth.
The panelists, including business leaders, policymakers, and labor representatives, largely agreed that historical technological revolutions haven’t led to permanent job losses, but acknowledged significant transition challenges. Ravi Kumar from Cognizant emphasized that productivity gains require business reinvention and integrating human-machine workflows, noting that less skilled workers often benefit more from AI augmentation. Laura D’Andrea Tyson warned about the uneven distribution of productivity benefits, citing how previous technological advances increased inequality and reduced labor’s share of income.
European Commissioner Valdis Dombrovskis viewed AI as an opportunity for Europe to catch up in productivity, particularly given the continent’s aging workforce. Jonas Prising from ManpowerGroup stressed that successful AI implementation requires process redesign and widespread reskilling, especially for smaller employers. Labor leader Elizabeth Shuler advocated for worker-centered transitions, emphasizing the need to include workers in AI development from the beginning rather than leaving them to adapt afterward.
The discussion concluded with consensus that while technology-driven unemployment isn’t inevitable, proactive policies for reskilling, better incentive structures favoring augmentation over automation, and ensuring workers share in productivity gains are essential for successful transitions.
Keypoints
Major Discussion Points:
– AI’s Impact on Productivity vs. Employment: The panel explored the tension between AI driving productivity gains (14-35% increases in some sectors) while potentially eliminating jobs, particularly affecting younger workers (ages 22-26) in AI-exposed occupations who saw 13-16% employment decline.
– Augmentation vs. Automation Strategies: A central debate emerged around using AI to augment human capabilities rather than replace workers entirely. Companies like Cognizant reported better outcomes when AI amplified worker potential (especially for lower-skilled workers who saw 36% productivity gains) versus simply automating tasks.
– Skills Training and Workforce Transition: Panelists emphasized the critical need for reskilling programs, with AI skills becoming as fundamental as Microsoft Office proficiency once was. The discussion highlighted the importance of contextual engineering and teaching AI to integrate into existing workflows rather than just applying it to old processes.
– Sharing Productivity Benefits: Labor representative Liz Shuler raised concerns about ensuring workers receive fair shares of AI-driven productivity gains, noting that 40% of US workers lack $400 for emergencies despite technological advances. The discussion addressed the risk of benefits flowing primarily to capital rather than labor.
– Policy and Institutional Changes: The panel discussed needed policy interventions including tax incentives that favor augmentation over automation, stronger aggregate demand policies, and the importance of including workers upstream in AI development rather than as an afterthought.
Overall Purpose:
The discussion aimed to address how society can harness AI’s productivity benefits while preventing “jobless growth” – ensuring that technological advancement creates shared prosperity rather than widespread unemployment and inequality.
Overall Tone:
The tone was cautiously optimistic but realistic. While panelists generally agreed that AI wouldn’t lead to permanent mass unemployment (citing historical precedent), they acknowledged significant transition challenges. The conversation became more urgent when discussing worker displacement and benefit distribution, with labor representatives emphasizing the need for worker-centered approaches. The tone remained collaborative throughout, with participants building on each other’s points rather than engaging in adversarial debate.
Speakers
Speakers from the provided list:
– Erik Brynjolfsson – Panel moderator, Professor at Stanford (mentioned as neighbor to UC Berkeley)
– Laura D’Andrea Tyson – Professor at the Graduate School of Business at UC Berkeley
– Jonas Prising – CEO of the Manpower Group
– Elizabeth Shuler – President of the AFL-CIO (largest labor union in the US), representing 64 unions and 15 million workers
– Ravi Kumar S. – CEO of Cognizant
– Valdis Dombrovskis – Commissioner at the European Commission, focus on economy and productivity
– Audience – Audience member who asked a question (Simon O’Connell, SMB Global Development Partner)
Additional speakers:
– Simon O’Connell – SMB Global Development Partner (identified himself when asking a question from the audience)
Full session report
Comprehensive Report: Preventing Jobless Growth in the Age of AI
Panel Discussion at Davos World Economic Forum
Executive Summary
This panel discussion at the Davos World Economic Forum addressed one of the most pressing economic challenges of our time: how to harness artificial intelligence’s productivity benefits whilst preventing widespread unemployment. Moderated by Erik Brynjolfsson, Professor at Stanford, the discussion brought together diverse perspectives from academia, business leadership, policymaking, and organised labour to examine strategies for achieving shared prosperity in an AI-driven economy.
The central tension explored was what Brynjolfsson framed through the productivity equation – whilst AI demonstrates remarkable potential for productivity gains (14-35% increases in some sectors), it simultaneously poses risks to employment, particularly affecting younger workers aged 22-26 in AI-exposed occupations who have experienced a 13% decline in employment. The panel’s overarching conclusion was cautiously optimistic: whilst AI-driven unemployment is not inevitable, proactive intervention through reskilling programmes, restructured incentives favouring augmentation over automation, and policies ensuring workers share in productivity gains are essential for successful transitions.
Panel Composition and Key Perspectives
The discussion featured a strategically diverse group of stakeholders:
Erik Brynjolfsson served as moderator whilst contributing significant research insights on AI’s employment impacts, framing the discussion around the fundamental productivity equation and presenting new data on AI’s effects on young workers.
Laura D’Andrea Tyson, Professor at UC Berkeley’s Graduate School of Business, brought historical perspective on technological transitions, warning that productivity gains from previous digital revolutions disproportionately benefited capital owners rather than workers.
Jonas Prising, CEO of ManpowerGroup, offered insights from workforce management, emphasising the complexity of AI implementation and the challenges facing smaller employers in managing transitions.
Elizabeth Shuler, President of the AFL-CIO representing 64 unions and 15 million workers, provided the labour perspective, highlighting current worker insecurity and advocating for worker-centred transitions with upstream involvement in AI development.
Ravi Kumar S., CEO of Cognizant, presented a business transformation viewpoint, arguing that AI implementation requires complete business reinvention rather than simply applying technology to existing processes.
Valdis Dombrovskis, European Commissioner, viewed AI as an opportunity for Europe to address its productivity lag, particularly given the continent’s ageing workforce where “we are aging continent, so labor supply is shrinking.”
The Productivity Equation Framework
Brynjolfsson established the conceptual foundation by reframing the productivity debate: “Productivity is output divided by input. And for some reason, a lot of people focus on the numerator, and other people focus on the denominator… as you increase productivity, that ratio of output to input, you can either have the numerator grow or you can have the denominator shrink or just lay flat.”
This framework proved crucial throughout the discussion, revealing why optimistic growth narratives and pessimistic job loss concerns often talk past each other – they focus on different parts of the same equation. The challenge is achieving productivity gains by growing output and value creation rather than simply reducing employment.
Current Evidence: Productivity Gains and Employment Effects
The panel examined concrete evidence of AI’s effects across different sectors and demographics. Brynjolfsson presented striking productivity figures: “We’re seeing 14% productivity improvements in call centres, 35% in some cases, much higher in software development,” with some coding applications showing “even triple-digit” improvements.
However, the employment picture proved more complex. Brynjolfsson’s recent research, described as “canaries in the coal mine,” revealed that “young workers, those 22 to 26, in occupations that are more exposed to AI, we found about a 13% decline in their employment, maybe as high as 16%.” This data, which “started in 2024 and 2025,” represents some of the first concrete evidence of AI’s employment impact.
This finding sparked considerable discussion. Kumar expressed surprise – “Really?” – when hearing these statistics, noting that Cognizant “hired more graduates this year than we’ve ever hired in the history of the company.” He argued that entry-level workers actually adapt more easily because “they’re learning the new way of doing things from day one.”
The apparent contradiction highlighted different measurement approaches and timeframes. Kumar emphasised that AI is “probabilistic rather than deterministic,” requiring “context engineering” and fundamental business process changes that take time to implement effectively.
Historical Context and Transition Challenges
The panel extensively discussed historical precedents, with general agreement that past technological revolutions haven’t led to permanent job losses but have caused significant transition challenges.
Tyson provided crucial historical perspective on the digital revolution: “If you look at the technological revolution, say the digital revolution… you see very significant polarisation of the labour market… The productivity surplus is going to capital. It’s going to capital.” She referenced David Autor’s work and noted that transitions occur over “40 years,” creating substantial interim challenges.
Shuler challenged the panel’s historical optimism with a pointed intervention: “Can anyone give me an example of a transition that really went well? I think we should learn from our past… when we lost manufacturing, hollowed out the industrial Midwest, and left workers behind. That did not go well. Why are we not learning the lessons from these previous transitions?”
This question significantly shifted the discussion’s tone, forcing acknowledgement that whilst technological unemployment might not be permanent, poorly managed transitions can devastate communities for decades.
Prising offered a more measured perspective on adoption speed, citing blockchain and driverless cars as examples of technologies where “progress speed is often overestimated, with human preferences maintaining roles even when automation is possible.”
Augmentation Versus Automation: The Critical Choice
A central theme throughout the discussion was the distinction between using AI to augment human capabilities versus replacing workers entirely. Brynjolfsson’s research provided empirical evidence for this distinction: “We found that the augmenters, those who are learning new things, had growing employment. So they both had more output and more employment, so more shared prosperity. Sadly, that was a minority of the folks.”
Kumar expanded on this theme, arguing that successful AI implementation requires fundamental organisational change: “This is not about applying technology to old stuff we already had… We have to reinvent the business, reinvent the process, reinvent the flow.” He noted that AI’s impact is particularly beneficial for lower-skilled workers, with “bottom 50 percentile workers seeing 36% productivity gains.”
However, Tyson identified a crucial structural challenge: “For why does a company decide to use the technology to augment skills versus to automate skills? If you have a high percentage of labour costs in your overall cost structure… your incentive is to automate it away.”
This observation highlighted the gap between what might be optimal for society (augmentation leading to shared prosperity) and what current economic incentives encourage (automation for cost reduction).
Skills Development and Workforce Transition
All panellists agreed on the critical importance of comprehensive reskilling programmes, though they differed on implementation approaches.
Dombrovskis reported that “one-third of job vacancies requiring AI skills” in Europe, representing both challenge and opportunity. Prising detailed practical training approaches, noting that AI skills are becoming as fundamental as Microsoft Office proficiency once was.
The challenge of scale proved significant. Prising noted that whilst “large enterprises will handle reskilling,” policy focus was needed on “majority of smaller employers” who lack resources for comprehensive training programmes.
Shuler advocated for worker-centred approaches: “Working people need to be involved upstream in how this technology is developed and deployed.” She used the example of WNBA players to illustrate worker concerns: “WNBA players… they don’t want AI to be evaluating whether they’re a good player or not. They want humans making those decisions.”
Distribution of Benefits and Inequality Concerns
The question of who benefits from AI-driven productivity gains emerged as a central concern, with significant disagreement between speakers.
Shuler reinforced inequality concerns with current economic data: “40% of working people in this country don’t have $400 for an emergency” despite technological advances, emphasising that “AI is just adding to the insecurity that working people already feel.”
Kumar presented a more optimistic view: “AI commoditises expertise and can be an equaliser by removing entry barriers.” He argued that AI could democratise access to high-level capabilities previously available only to highly skilled workers.
Tyson warned about historical patterns, noting that without policy intervention, AI could exacerbate existing inequality trends where productivity gains flow primarily to capital rather than labour.
Global Development and Sectoral Applications
The discussion explored AI’s potential across different economic sectors and global development contexts. Kumar identified significant opportunities in “agriculture, manufacturing, healthcare, and construction sectors.”
Tyson highlighted development implications, referencing a World Bank president’s presentation on “edge AI applications in agriculture, healthcare, and tourism” that could help developing nations increase productivity.
An audience member, Simon O’Connell, proposed that “shared productivity gains model could replace traditional development assistance frameworks,” pointing toward new models of international cooperation based on measurable productivity improvements.
Policy Implications and Structural Changes
Throughout the discussion, speakers identified numerous policy interventions needed to ensure successful AI transitions.
Brynjolfsson advocated for changing how AI development is measured and incentivised, suggesting the need for benchmarks that measure human-AI collaboration effectiveness rather than AI-only performance. He also proposed “tax and policy incentives that favour augmentation over automation strategies.”
Shuler emphasised worker rights: “Workers need voice in how AI is implemented, with collective bargaining as a tool for data usage decisions.” She argued that workers should have input into how AI systems evaluate their performance and make decisions affecting their employment.
Kumar introduced the concept of “macro-delegate and micro-steer” as a management approach for AI implementation, emphasising that successful deployment requires fundamental infrastructure changes since current systems are “built for humans, not machines.”
Areas of Consensus and Persistent Disagreements
Despite different backgrounds, the panel achieved remarkable consensus on several key points:
– AI will transform rather than eliminate jobs long-term, though managing the transition period is crucial
– Augmentation approaches are superior to pure automation, requiring fundamental redesign of workflows with worker involvement
– Comprehensive skills development is critical, requiring strong partnerships between businesses, educational institutions, and workers
– Worker involvement is essential in AI implementation, with surprising agreement from business leaders
However, significant disagreements remained:
– Impact on young workers: Brynjolfsson’s data showing employment decline versus Kumar’s experience of increased graduate hiring
– AI as equaliser versus inequality creator: Kumar’s optimism about democratising expertise versus Tyson and Shuler’s warnings about historical patterns of benefit concentration
– Speed and complexity of adoption: Different assessments of how quickly AI will be implemented and its immediate versus long-term effects
Unresolved Challenges
The discussion identified several critical challenges requiring further attention:
– Structural incentive problems: How to incentivise companies to choose augmentation over automation when labour costs are high
– Benefit distribution mechanisms: What specific policies should redistribute productivity gains from capital to workers
– Transition management scaling: How to extend successful transition models beyond large enterprises to smaller employers
– Worker agency balance: How to provide meaningful worker voice whilst preserving innovation incentives
Conclusion
The panel discussion revealed both the promise and complexity of preventing jobless growth in the AI era. The most significant insight emerged from recognising that optimistic growth narratives and pessimistic job loss concerns address different parts of the same productivity equation – the challenge lies in growing output and value creation rather than simply reducing employment.
Shuler’s intervention about past transition failures served as a crucial reminder that positive outcomes require proactive intervention and structural changes. The surprising alignment between business leaders and labour representatives on worker involvement suggests potential for collaborative approaches.
The path forward requires addressing structural incentives that currently favour automation over augmentation, developing new frameworks for measuring human-AI collaboration, and ensuring productivity gains are shared rather than concentrated. As Brynjolfsson’s research on “augmenters” showed, it is possible to achieve both higher productivity and growing employment – but this requires deliberate effort to shape technological development toward inclusive outcomes.
Success will depend not only on technological capabilities but on policy choices, business strategies, and institutional arrangements that align private incentives with public benefits. The goal of shared prosperity in an AI-driven economy remains achievable, but only through sustained commitment from all stakeholders to ensure that AI serves to augment human potential rather than simply replace human workers.
Session transcript
Hello, and welcome to this panel on preventing jobless growth. This has been the big topic here in Davos, and we have a lot of drivers of both the growth side and the job side, technology, geopolitics, demographics, and other factors. There’s really two narratives going on here that are surprisingly disjoint.
There’s the positive narrative talking about growth and how we all have more wealth creation. There’s the negative narrative about where the job’s going to come from. And I would like to highlight that these are actually two sides of the same coin.
Productivity, what is productivity? Productivity is output divided by input. And for some reason, a lot of people focus on the numerator, and other people focus on the denominator.
On the numerator side, we have some amazing progress on a lot of dimensions where, as productivity grows, we have the potential for not just more wealth creation, but cleaner environment, addressing poverty, longer lifespans in health care.
And technology has been helping with that. For instance, we did a study looking at how LLMs were affecting call centers, and we found that there was about a 14% average increase in productivity when people are using it, up to 35%. There have been studies in software and coding where you sometimes get even higher double-digit or even triple-digit productivity levels.
And recent data, it’s early in the United States and other countries suggest that we maybe see the early inklings of a productivity revival. The other side of productivity, though, is the denominator. And there, the data is more worrisome, because as you increase productivity, that ratio of output to input, you can either have the numerator grow or you can have the denominator shrink or just lay flat.
And that’s a bit of what we’ve been seeing lately. Job growth has been somewhat discouraging. We did a paper we call Canaries in the Coal Mine, looking at some early indicators of what was happening in jobs that were highly exposed to artificial intelligence using data from ADP.
We found that those in the youngest group, aged 22 to 26, in the most exposed occupations, like I mentioned, call centers, software, we ranked all 900 occupations based on their tasks. That age group actually had about a 13% decline in employment relative to before LNMs, especially in 2024 and 2025, we started seeing that effect grow bigger. The older age groups did not have that effect, so the overall effect was fairly muted.
And some encouraging information, if you sliced it based on people who are augmenting, using the LNMs and generative AI to augment their work versus people who are using it to automate work, what we found was that the augmenters, those who are learning new things, had growing employment.
So they both had more output and more employment, so more shared prosperity. Sadly, that was a minority of the folks, but if we can get more people doing that, maybe that’s a potential path. It’s still early, that was only about 13% I mentioned for that one age group.
We just got some new data in, it’s not in the new paper yet, but it’s now about 16%. I’m not sure whether it’s going to continue to grow over the coming months and years, but it’s certainly, as I say, a canary in the coal mine to look a little bit more closely at it. Our goal, of course, is to prevent jobless growth, to get the benefits of productivity that not only increase the top line, but also don’t hurt people by eliminating jobs.
And we’ve assembled a set of the world’s leading experts here. I’d like to take a minute to introduce them, and then we’ll each have them weigh in on these issues. So let me start over here with Valdez Stromboskis.
He’s a commissioner at the European Commission. His focus is on the economy and productivity. Ravi Kumar is the CEO of Cognizant.
Jonas Prising is the CEO of the Manpower Group. Liz Shuler is the president of the AFL-CIO, the largest labor union, and Laura D’Andrea Tyson is a professor at the Graduate School of Business at UC Berkeley, my neighbor at Stanford. So, welcome all of you, and let me just start with a question.
I think we’ll start with you, Ravi, if that’s okay. I’d like to ask you, you know, we’ve heard a lot about, I just mentioned, about the potential for productivity boost from artificial intelligence. But we don’t really see it in the aggregate statistics a whole lot, that there’s been these amazing capabilities, but the aggregate numbers are still somewhat muted.
Can you make it a little bit more tangible for us and tell us a little bit about how AI is augmenting work, replacing work, affecting productivity? You’re very much on the front lines in your organization, and what can organizations do to harness the potential better? Thank you.
Thank you, Erik, for this question.
You know, we did this study with 18,000 tasks and 1,100 occupations from the ONIT database. Yes. We are pretty sure now that the technology is pretty advanced.
I mean, the amount of innovation which has happened around it, the drift of value to businesses has not happened yet. No. And that’s, you know, that’s true for most technologies.
It takes like eight to ten years. In this case, the drift has been slower than what we think it should have been. And the reasons why the drift has been slower, and therefore the productivity has been a little bit of a flattish curve.
You know, first and foremost, this is a technology which is very probabilistic in nature. It is unlike traditional software, classical software, which is very deterministic. The implementation of it and the reinvention of businesses, which was relatively easier when we applied enterprise software 30 or 40 years ago.
When you have probabilistic software, and you’re trying to do things which are very different to what classical software did, which has human judgment integrated into the decision making, into the work patterns, into the problem solving, you would have to do contextual engineering of the technology which is available.
So one of the things we’ve been working on is to make sure that how do we teach AI to be a part of your team? And until you do that, productivity is not gonna be teach the AI to be a part of your team. And I’ve written a HBR article on it.
Have a look at it. I mean, it’s in the public domain. How to teach AI to be a part of it, how to teach AI to understand the hustle of the company, the work patterns of the company, the tribal knowledge of the company.
And that is integral to what we’re trying to codify and create digital labor using AI. Remember, productivity has been very high for software development. If you just take that specific function.
For decades. Yeah, companies like ours, we’ve seen a significant bump on productivity just on software development cycles, but that’s very deterministic. That’s deterministic work.
It’s writing logic, writing code. In fact, I would say one of the points you made, early careers have got impacted. In our case, we actually hired more school graduates last year than ever before.
Really? Yeah, and we think we can amplify the potential of people at the bottom of the pyramid, eliminate entry barriers, create more productivity, more throughput, provided you can use it to amplify the potential of the people at the bottom.
I’d like to hear a little bit more about that. And we did the stat for the bottom 50 percentile, we had a bump of 36 percent. And the top 50 percentile, we only had a bump of 17 percent.
And that gives us hope. that you could broaden the pyramid, you could take paths to expertise much quicker, learnability is much faster.
Can I just ask on that point in particular, I mean, by the way, we found a very similar pattern in our call center study that the less skilled workers actually had the bigger boost from using the technology.
But what you’re doing and continuing to hire a lot for these entry-level people could be a roadmap for a lot of people. So I wanna hear more about how you’re doing because it is true that a lot of the skills that they were doing are increasingly done by machines. So are you upskilling them in a way?
Are you teaching them new things that they can do so that they can continue to be productive for you?
I think the good news is early carriers, they do not even know how the old stuff was done. This is the first time they’re starting to write software code, and this is the only way they knew it. So it’s a relatively easy transition to this process.
The second bit I would say is, you know, code-assist platforms, which helped people to work with machines together, were very synchronous. We have now started to get to asynchronous work, which is agents do the work, you macro-delegate it, and you micro-steer it. And if you macro-delegate it and micro-steer it, you’re sending agents for work, delegating them, and you’re taking it and you’re steering it.
So I’m gonna touch a few other points so that I don’t hijack the time for the rest of the panelists. So context engineering is a space we have deeply researched on. We think that’s very important in the contextual computing era.
You know, the past was, you know, we wrote technology around the microprocessor. We’re gonna write technology around the LLM, which is very probabilistic in nature. Context engineering is a very important point.
You know, integrating work between machines and people is very important. The infrastructure of enterprises has predominantly been built for humans. It’s not been built for machines.
And if you look at. What’s happening with autonomous cars, you know, to implement autonomous cars has been harder because the infrastructure has been built for humans. And now you’re trying to play, you’re trying to build an interplay between humans and machines.
If it was just machines or just humans, it would have been relatively easier. If it’s machines and humans coming together and amplifying each other, it’s going to be harder. So that infrastructure, integrating work patterns, in fact, AI is going to be in the middle of a workflow.
You’re going to have people on the front and people at the back. People on the front doing, you know, authentication, problem finding, you know, and ideation and stuff like that. People at the back doing validation and verification.
So to integrate work, productivity is only going to come if you start to integrate the work from human and machine labor. The last point I want to make, and we can open it up subsequently after all of us talk, this is not about applying technology to old stuff we already had. So what will really happen is the old stuff will remain as is, and we’re going to do it in a cheaper way.
We have to reinvent the business, reinvent the process, reinvent the flow, and that reinvention and reimagination will drive productivity. So integrating workflows between machines and humans, context engineering, you know, amplifying the potential of humans, and reinvention of businesses rather than just eliminating work or just trying to apply this on top of something which is already bad.
Historically, that’s been where most of the biggest gains come from.
And I’m sure we’ll come back to that point a little bit, but I want to get some of the other folks involved. Let’s go to you, Laura, because there’s, you know, I talked about this tension between productivity and jobs. But you know, economists know that historically, whether it’s in the Industrial Revolution or all the changes since then, every time there’s been an amazing technology, it hasn’t eliminated jobs.
People were worried about it. You can, you can… I quote all the time, but, you know, there have been more jobs historically.
Is this time different? Are you worried? What are your expectations?
So first of all, I want to say, yes, I certainly espouse those views. I certainly feel that past technological revolutions have actually led… There’s no such thing in the evidence so far of long-term technological unemployment.
I mean, what happens is the technology over time, along with changes in demand, changes the composition of employment, changes what people do, changes what sectors are demanded to grow, and labor is hired to do that.
So that’s the first thing. The second thing, though, is, and I think it’s fair, and it explains some of the concern that people have, is the disruption effect can be pretty big. The disruption effect can be pretty big.
Even if you’re moving from one set of jobs to another job, that’s not the easiest thing to do. Yeah. I mean, you know, it’s like you…
Historically speaking, again, the technology oftentimes will eliminate or reduce the number of jobs. It might increase productivity in that place, but the number of jobs actually declines, whereas the new jobs are someplace else, or the new jobs require some set of skills that people who are displaced don’t…
I mean, what’s an example of that in the past that we’ve seen? Have. I mean, I would say that the best work here is really by David Autor, and you can kind of see all the new jobs, but again, it’s a 40-year process.
It’s not a 40-day process, or four years. So I would say that the concern people have is about the transition. And then the policymakers, I mean, I tend to look at this all in terms of policy.
So what should policymakers do? Well, number one, I just want to start with the notion that we really need to have very strong aggregate demand. I mean, if we have a weak labor market, then the growth of new jobs is going to be slow.
So first of all, let’s just say you’ve got to get macro policy. But then you have to do things like help train. And I heard this morning, and I think this has been really important, California does a lot of its training through community colleges.
And so basically the kids going into community colleges are very concerned about your kind of data. Oh, my God, I’m not going to get into an entry-level job in this kind of office situation. So what the, I would say what community colleges and educators need to do is have very tight links with the business community.
What do you, what’s happening with you? What sorts of jobs do you feel you are going to create? Like the set of, like the Ravi described, the 900 different occupations.
My view would be work with Ravi. The policymakers should absolutely, because in our market system, these kinds of decisions are going to be made ultimately by the companies. The policymakers actually have to respond.
So I would go that far. I do want to say, though, that I have a dystopian view in the following sense. I certainly believe that the productivity benefits are there and can be substantial.
The question is, how are the productivity benefits shared? This is a huge issue going forward, because if you look at the technological revolution, say the digital revolution that came in between the industrial revolution and where we are today, you see very significant polarization of the labor market.
You see a loss of middle skill, middle income jobs. Many people benefit with education at that point by moving top, but a lot of people fall bottom, fall bottom. And the productivity growth of the economy, measured in a variety of ways, the real wage growth does not keep up, does not keep up.
What has happened, and I’m really just talking about the digital revolution. Where you can see that not only is there polarization, but the labor share of income declines. The capital share of income rises.
The productivity surplus is going to capital. It’s going to capital. And so I think those are things we really have to worry about.
It’s not just the number of jobs. It’s not just that the composition of jobs is going to change. And we are beginning to see in the data.
It really is how you share the benefits. And therefore, from a policymaker’s point of view, it’s a challenge to think about how to do that. But I think we should get that on the table as one of the things to discuss.
I believe in the productivity benefits. I believe that employment will change over time. I believe that there’s no long-term unemployment.
But my concerns are about real wage growth, about polarization. The division of the benefits. About the division.
And the transition. Of course. And the transition.
The transition is very important.
And let me just hold that there and let’s get some other voices in. And so you mentioned policymakers. And so we have one here, Valdis.
If you want to weigh in, like, you know, how are you thinking about, you know, this risk of jobless growth? And what are the policy targets we should be thinking about? Is that on your agenda?
Well, good morning, everyone.
Obviously it’s on our agenda. Well, all in all, what we are looking, if we first look at the productivity, that in terms of productivity, EU is lagging behind other major economies already for a few decades. So there is some catching up to do.
And if we look why we are lagging behind in productivity growth, it’s largely explained by tech sector. So from that point of view, we are viewing AI as an opportunity. And we are now investing a lot of actually developing and applying AI, setting up what we call our AI Continent Action Plan, how we…
promote and develop the AI, including AI gigafactories and so on. And second, apply AI strategy, where we facilitate the use of AI across different sectors of economy and also the public sector. So in terms of jobs, jobs right now, a labor market in the EU is robust and what we had seen in post-COVID recovery, actually, our economy has created more jobs than you would normally expect with this kind of economic growth we are having right now.
So our- So it’s almost the opposite of jobs. Our employment growth continues to, our employment continues to grow and unemployment is at historically low levels. And also, if you look at the OECD data, what the OECD data says that approximately one-third of job vacancies are highly exposed to AI in terms of requirement of AI skills.
And it probably will increase to two-thirds in the coming years. So obviously, as some colleagues mentioned, an important question here is about managing a transition, how we equip people with the right AI skills, actually, to be able to use it as a factor of production.
And in Europe, in any case, we need to offset it against another tendency. We are aging continent, so labor supply is shrinking. Also, to sustain the level of prosperity, we need to increase productivity.
So we mainly see it as a challenge of properly managing transition, making sure that the labor has the right skills for also deployment of the AI.
Great, that’s a terrific list there. So that’s the policy makers. Let’s talk about employers.
I think they also have a responsibility, and we have maybe the world’s biggest, one of the world’s biggest employers, you can tell me, at Manpower. Jonas, can you tell us a little bit about how you think about it? Like, what is the responsibility?
What’s the potential for employers to address this?
You know, I think, so as a starting point, I think it’s great to hear that, you know, we all seem to believe that long-term jobless growth is not going to be the main issue to contend with. The transition, certainly, and how can we ensure that this is evenly distributed? To your early comments, though, Erik, when we look at youth unemployment in the U.S., for instance, we see this disparity now.
But we always see this disparity when, you know, the labor markets are going through a tougher time, as they have in 25. I know the economic growth is strong in the U.S., but the labor markets have been weakening all the way through 25. So not only college graduates are having a tougher time finding their first job, and high school graduates.
They’re kind of on the front lines of this. And this is what happens as companies go through an economic cycle. They hire experienced workforce first, or specialized skills.
So I don’t know if your research sort of eliminated that, because I… Well, let me just, very briefly, yeah.
So we looked at that age group, but we also looked at the exposure to AI. And the folks that are most exposed had a big fall, the less exposed had a lesser fall. And then, interestingly, the least exposed, like home health aides, they actually had growing employment.
So it wasn’t just a level shift, it was kind of a twist, depending on how much they were exposed. So that’s a little bit more concerning. I have to be clear, this was not a causal test.
We didn’t, like, you know, have an A group that was exposed to AI and a B group that didn’t get exposed. So it’s all correlational, but the patterns are a little concerning. Yeah.
Well, we’re not really, we’re seeing that in some job categories, and you mentioned the two, really, the only ones where we can see some effect. But broadly speaking… Value realisation is lagging the strength and the advancement of the technological…
Yes, as Ravi said as well. Just as you said, Ravi. In fact, we did a quantification of that by looking at those 18,000 tasks.
In the United States, $15 trillion is the labour economy or the labour value. $4.5 trillion is already theoretically exposed to AI. But that $4.5 trillion is not realised yet in a way.
And I think a lot of those reasons is that to get the full benefit of AI, it’s not about an AI applied to a task, but it is about an AI applied to a workflow. And that application of the design, the process redesign on how work gets done will create new jobs, will require new skills, and it will take time. The reinvention.
The reinvention. It’s easy to reinvent with five people in a start-up. Large organisation, public sector, it takes time.
The culture has been established. And the ability to attract and grow the skills, which comes back to the point you asked me about around employees. And of course, employers have a big responsibility to avoid this notion of a divide.
Now, when it comes to the digital divide, I have a bit of a contrarian view, because AI commoditises expertise. So, instead of creating a divide between the haves and the have-nots, everybody has access, essentially, to an infinite extension of your mind. Of their capabilities.
Of capabilities. So, if you let me finish. So, I think that is a positive.
But what we’re observing today that companies are already doing, those at the front edge, they are really training their people in AI skills. And you’re saying, what is an AI skill? If what is true is a process redesign is needed, what kind of skills are you training for?
Well, they’re being very pragmatic. You can think about this almost as the evolution of, you know, Word, Excel, and PowerPoint. First of all, know how to write a really good prompt.
Second, know how to integrate a document and analyze this document. Draw the insights, ask conclusions, you know, do that. And then last but not least, moving to the last phase, is how can you augment your skills when you’re accompanied by a large language model and you can now get work done in a different way.
I like that. So that’s sort of the thinking that we are starting to see coming through. So you can think about, you know, the skills or rather you can think about the skills being the technology infrastructure equivalent.
So they are needed. And the training is really all about the R&D and how people can use it. Use training.
The most successful candidates for jobs today list, I know how to write a prompt. I know how to do chat GPT. I work with Claude.
Much as you remember seeing, yes, I know I can do the Microsoft suite of Office very well. You know, this is now just ubiquitous in applications. And those that have those, employers will say, oh, well, this is a person who’s learning new skills, much more employable.
I think we are interested in that candidate.
This is so important because as you’re highlighting, you know, you can use AI to augment yourself and humans and machines working together. It’s hard. It’s harder than just having one separate.
But ultimately, if you do that, now the person becomes an extension. They’re creating more value. And ultimately, that makes them more employable.
So that’s a very promising path. Just to add to it, I mean, you raise this point. The divide which was created for digital technologies was also because there were skills needed to get those jobs.
And it wasn’t as commoditized.
I mean, I actually think AI is an equalizer in many ways because it takes the entry barriers out. You can diffuse this fast and you can actually access jobs which you couldn’t access if we pivot the skilling in the right way.
I just point out, we should go on, but a question I have in my mind. If we live, now let’s go to a very specific structure. The structure we live in is a structure in which companies make these decisions, okay, they make these decisions.
For why does a company decide to use the technology to augment skills versus to automate skills? If you have a high percentage of labor costs in your overall cost structure and it tends to be in fairly low skill categories that can be easily automated away, your incentive is to automate it away, I believe.
You’re both talking about companies that are doing something different. I think what I want to understand is what can we do in the structure to encourage companies to behave that way because there’s a natural incentive. I’ll give you a chance to think about an answer to that question.
I have some thoughts as well.
We’ve all been talking about workers and here we have Liz Shuler representing the largest labor union. I think we should hear from, you’re the anchor person here. Tell us a little bit about, given the scale and some of the things you’ve just heard here, tell us about how you think we can ensure that workers get a fair share, have a voice in all of this.
What are you seeing? I’ve been sitting here waiting to speak because we represent all of the people that you’re talking about.
In the U.S., the AFL-CIO is an umbrella organization of 64 unions, 15 million workers, ranging from every industry from professional athletes to actors and bus drivers, nurses, health care workers, the people making this meeting happen behind the camera.
So we absolutely… are paying attention to what is happening with this transition and this discussion. We are seeing just as a backdrop, the economy, at least in the US and around the world, isn’t working for working people now, right?
We have inequality at its highest levels. You know, people are working harder and harder for less. They’re working two and three jobs just to keep up.
In the US, workers, 40% of workers do not have $400 for an emergency. Now, put AI on top of that, the insecurity that we’re all experiencing, the fact that people are waking up and some new technology is landing on them in their jobs without training, without them having a say.
Of course, they’re gonna be anxious. Of course, they’re gonna be feeling insecure about what the future holds. And so, I think we really need to stop and say, who are we doing this for?
What are the results we want and how do we get there? Well, we get there by including workers in the process. They shouldn’t have to be at the end of the cycle.
They should be actually upstream in the cycle. And in fact, we have a partnership with Microsoft for that very reason, to get workers in the labs to say, if we’re gonna start using AI in transportation, we should have bus drivers in that lab working with developers to make this as successful as it can be.
Because we are not anti-technology. If you think about every industrial revolution that we’ve been through, working people have helped us make that transition. And it’s really because we’ve helped tame the technology.
We’ve helped figure out how to use it in the most effective way. So, I think your question about augmentation versus replacing, that is the big question we have. Because if we can all agree that this is to make our jobs better and safer and easier, more productive, then we’re all in.
But if you’re looking to just de-skill, dehumanize, replace workers. put people out on the street with no path forward, then absolutely, you’re going to have a revolution. So I think that’s something we all need to be very real about.
And thinking about if we are going to have productivity gains, working people, the ones who make these industries happen, need to share in that. So there’s not been a lot of discussion about that here. Of course, in terms of how we create policies, how we create tax infrastructure, whether or not we are redistributing, that word is a dirty word around here, but we need to talk about it and confront how we’re going to make sure that working people share in the gains of these technologies.
And if you look at the numbers of jobs, let’s talk about job quality. Because yes, maybe there are a lot of jobs created, but what kind of jobs are we talking about? Are they jobs that can sustain a family?
Is it a job that people can actually work one job? One job should be enough. So I think these are the things that are on people’s minds.
And also, I have to say a word about privacy, data security, and how the data is used. I’m going to use the WNBA players as an example. Right now, professional athletes who are fighting for a fair contract, data and AI in their workplaces is part of it, because AI is actually judging how players are making decisions and how they play a game.
Are we now saying that that’s what we’re getting to? That an athlete can’t use professional judgment, an artist can’t use professional judgment without an AI system evaluating them at every turn, saying whether they made the right decision to make a pass to this player or that player?
So collective bargaining actually is a tool that really undergirds this and gives workers a seat at the table to determine how that data is used.
Well, I think we all agree we need to work towards shared prosperity. I want to pick up on that theme of augmentation. But first, I want to just give a heads up.
I’m going to open the discussion up to questions from the audience. So if some of you have a question you want to raise, you formulate that. But let me just pick up on what you’re saying.
Almost everyone here was talking about. this distinction between automating and replacing workers versus augmenting them and extending what they can do. And, you know, I’ve done some work on this, and one of the challenges is that, you know, we can have much more shared prosperity if we use AI to augment people, but there’s a set of institutional incentive policy levers that are steering us excessively towards automation.
The other direction. And the wrong direction. And I wrote a paper called The Turing Trap that many of you may be familiar with the Turing test, this idea that we try to make AI that perfectly imitates humans and can replace them.
It’s been an inspiration for a lot of technologists. I think it’s a terrible direction for the technology. We should be doing, as Ravi and others said, finding ways for humans and machines to work together.
Most of the benchmarks out there that the AI researchers are working towards, they are a black box of just how well can this AI do some particular task. And then they take it and they put it in a hospital and in another organization. It turns out it doesn’t work nearly as well as it did in the laboratory.
Surprise, surprise. You know, if a medical imaging system, you know, says, oh, you know, cut out the patient’s left lung and the doctors or nurse says, well, why are you saying that? Well, probability 0.87.
Like, what do you do with that? You know, do you operate or not? It has to be designed from the beginning to work with the workers and have them coordinate.
So we’re creating a new set of what we call centaur benchmarks where it’s not cheating to have the human help the machine or the machine help the human. It’s the whole point is the two of them need to work together. A whole group of them work together.
And we have some tax incentives that stray too much towards automation. A lot of CEOs, I think they overly focus on some of the bottom line metrics of cutting costs when they should be thinking about how they can create more value. So there’s a whole set of cultural and financial incentives that we can change to lead us more towards this augmentation.
I think that would be a great takeaway from this panel is we can have productivity growth increasing that numerator I talked about before and not simply cutting the denominator. Sure. Vadis, if you want to weigh in, then let’s get a question from the audience.
Very quickly, just one point I wanted to make to put more emphasis also on labor skills. Because we have been dealing also in previous years with digital skills more broadly, where is a huge shortage of digital skills. And we are seeing right now the same with AI.
Indeed companies are, we see it also in the microdata, massively retraining their workforce because there’s not enough graduates with the necessary AI skills. So if we want to have this transition successful, we also need to focus very much on having the right skills of people entering the labor force and also people already in the labor
force. Absolutely. So is there someone in the, back here, if we get a microphone over here.
Here comes the mic. And just say your name and keep the question brief.
Brief as I can. Simon O’Connell, SMB, Global Development Partner. Fascinating discussion.
I love the kind of framing moving towards the shared productivity gains. Official development assistance last year was about $212 billion globally. It’s been dismantled, unraveled, depleted, et cetera, et cetera.
I wonder if there’s a whole new framework here around shared productivity gains where those doctors, nurses, teachers, even farmers in more affluent societies who gain time and the companies gain in growth through that increased productivity, that can be shared.
It’s really metricable. It’s really evidencible. And it can be shared.
I’m sure it can be shared, you know, in more affluent societies as well. But a whole different model moving away from the more traditional official development assistance structures to shared productivity gains. And I’d love to hear thoughts from the panelists on that.
Thank you.
Thank you. I just heard a really compelling presentation from the president of the World Bank. And what they’re doing is they believe that And that you can use edge AI models to help increase productivity in the key sectors that are for to bring a country or bring a nation, bring a continent out of poverty.
So he sort of talked about the edge AI in agriculture. So helping small-scale, low-income farmers in Africa who had a tendency to just leave the farm, sell the farm, go to urban, not be employed. This now makes them much more productive.
He talked about that in terms of health care. He talked about that in terms of tourism. So I think it’s possible if we, here we tend to focus all on the big LLM models and the big corporations, but actually there’s amazing set of things being done at the sectoral level.
And that probably also would be involved with you. I mean, you talk about voice and labor representation. If we’re going, if the technology is developing, getting the workers to work in a sector with the technology, I love the bus transport example.
I love it. My husband is a writer. What about the Writers Guild Association and how you protect your intellectual property?
So the basketball players, it just strikes me that we can actually get a huge increase in productivity, performance, wages, everything else, if we actually design the edge applications to really work in that.
So we’re running low on time here. Can I add one? Yeah, yeah.
I just want to give everybody a chance to weigh in, maybe for like a minute or so. And let me start with you, Ravi.
Yeah, so I think you raised this very important point on higher demand. Higher demand. And that I think is a trigger to, I mean…
Why are corporations focused on bottom line? Because they’re not able to do growth. So the ability to take the pivot back to growth and create a flywheel on that, and productivity is a flywheel, I mean it creates deflationary growth.
Of course. So the point you made on sectors to look at I think is very important. I mean a lot of productivity is looked at in context to the knowledge industry and a lot of jobs which got created in the last two decades were to the knowledge industry.
The reality is you need more productivity in agriculture, manufacturing. Care. Sectors, health care.
And the throughput you can get there is significantly higher and we did exposure scores and velocity of change in the last three years. The exposure scores are very low in construction, transportation, but the velocity at which that exposure is gonna go up is very high there. And you’re gonna create more jobs because the productivity is gonna go up.
I mean let’s take software development. Productivity is very high. Yes.
So the exposure scores are very high. The velocity of change is gonna be lower because it is so high. I mean in software, at least our study says the exposure is 70%.
Which means the rate of change every year is gonna be much lower. It’s already re-baselined. The jobs of the future are already here.
While manufacturing, health care, construction, transportation. Construction. These are sectors where you could do digital enhancement of physical jobs using AI technologies.
Yeah, and some of those physical changes are gonna happen a little bit later. We’re seeing it first in the cognitive side. Jonas, you wanna weigh in with some closing thoughts?
Yeah, you know what strikes me?
Over time, we always get very mesmerized by the technology itself. 10 years ago, we’re all worried. Blockchain, we were curious.
The whole Davos Trust, it was all about blockchain. That lasted one year and then that was it. Then we moved to.
driverless cars, what are all the bus drivers going to do, what’s going to happen with them, and you know, we’re nowhere near there. And when we are, the buses or the trucks have a driver in the cabin, because as humans, we prefer that certainty that just in case, just like airplanes, just in case, at least let’s have three pilots, even though evidence and data would say pilots are the ones that, you know, cause issues more than the technology in most cases.
So in the end, the technological progress and the speed of which, I think, is all about humanity. The work of the future enabled by technology is decided by the workers of the future and the skills that they acquire, yet we always tend to move into the notion of what can technology do and, you know, what is going to happen.
So this notion of we need to take care of the people, provide humans the edge, which we believe is going to be the case, augmenting with AI. So the human edge and the human age is really what we need to focus on, because the notion of reskilling at scale, we don’t have to worry about large enterprises. Large enterprises will retrain, reskill, upskill their people, because they’re competing and they have the resources.
We need to worry about all of the other employers, which are the majority of employment providers What are they going to do? And that’s where policy makers, distribution policies, and things like that are so important.
This amazing technology gets a lot of our attention, but making all those other changes, really that’s where the rubber hits the road.
That’s where it comes to life.
And that’s been lagging much more. We can close that gap and boost the productivity, that will make a big difference. Let’s go to Valdis, and then Liz, you get to wrap up, and I’ll say a final word.
Actually, to come back to the original topic of our discussion on jobless growth, I think we all concur that there’s not going to be a jobless growth. But we have to admit that there is also going to be some replacement of work. Professions will change.
It will not be able to preserve all the works. But the nature of jobs will be changing, and it’s also about managing transition, about reskilling the people, because there will be jobs which will be phased out. It has been the case since the first industrial revolution.
But the question is what happens on jobs with aggregates, and there are reasons to believe that there will be jobs. And the question is how we properly equip people for those new jobs. We will not be able to hold to every old job we had.
Thank you. Liz?
Can anyone give me an example of a transition that really went well? I think we should learn from our past. And I think about the U.S.
when we lost manufacturing, hollowed out the industrial Midwest, and left workers behind. That did not go well. Why are we not learning the lessons from these previous transitions and choosing a different way?
So we have the opportunity to do that this time. And so I would argue that this transition needs to be worker-centered, that we put workers at the front instead of them always the losers, and that training partnerships, not just top-down but workers again at the table, and then thinking about guardrails, guardrails to make sure that the technology is not running over us and that working people again are the ones driving it.
Absolutely. And that’s the interest not just of workers but all of us in society, and enlightened self-interest. So we need to do that.
And I think one of the things that we can do is get better visibility into how the world is changing, better data and statistics, and trying to align the incentives a little bit better. So thank you very much. This has been an amazing discussion, and we’re very privileged to have all of you be able to weigh in on this.
I appreciate it. Let’s give them all a round of applause. Applause.
Erik Brynjolfsson
Speech speed
191 words per minute
Speech length
2273 words
Speech time
713 seconds
AI shows significant productivity gains (14-35% in call centers, higher in software development) but aggregate statistics remain muted
Explanation
While AI demonstrates substantial productivity improvements in specific applications, these gains haven’t yet translated to broader economic statistics. This represents the gap between technological capability and widespread implementation.
Evidence
Study of LLMs in call centers showing 14% average increase in productivity, up to 35%. Studies in software and coding showing double-digit or even triple-digit productivity levels.
Major discussion point
AI’s Impact on Productivity and Employment
Topics
Economic | Future of work
Young workers (22-26) in AI-exposed occupations show 13-16% employment decline, while augmentation users see employment growth
Explanation
Research indicates that younger workers in jobs most susceptible to AI replacement are experiencing significant job losses. However, workers who use AI to enhance rather than replace their capabilities are seeing employment gains.
Evidence
Paper called ‘Canaries in the Coal Mine’ using ADP data, showing 13% decline in employment for ages 22-26 in most exposed occupations like call centers and software, with effect growing to 16% in newer data. Augmentation users had growing employment.
Major discussion point
AI’s Impact on Productivity and Employment
Topics
Economic | Future of work
Disagreed with
– Ravi Kumar S.
– Jonas Prising
Disagreed on
Impact of AI on young workers and employment patterns
Companies should focus on augmenting human capabilities rather than replacing workers entirely
Explanation
The optimal approach to AI implementation involves designing systems that enhance human performance rather than simply automating tasks. This requires rethinking benchmarks and incentive structures to promote human-machine collaboration.
Evidence
Reference to ‘The Turing Trap’ paper and development of ‘centaur benchmarks’ where human-machine collaboration is the goal rather than pure automation. Example of medical imaging systems needing human interpretation.
Major discussion point
Augmentation vs. Automation Strategies
Topics
Economic | Future of work
Agreed with
– Ravi Kumar S.
– Jonas Prising
– Elizabeth Shuler
Agreed on
Augmentation approach is superior to pure automation
Ravi Kumar S.
Speech speed
162 words per minute
Speech length
1316 words
Speech time
484 seconds
Technology adoption requires 8-10 years for value realization, with AI being probabilistic rather than deterministic, requiring contextual engineering
Explanation
AI implementation is slower than expected because unlike traditional deterministic software, AI is probabilistic and requires extensive customization to work within specific organizational contexts. This complexity delays productivity gains.
Evidence
Study with 18,000 tasks and 1,100 occupations from ONIT database. Reference to HBR article on teaching AI to be part of teams and understand company culture and tribal knowledge.
Major discussion point
AI’s Impact on Productivity and Employment
Topics
Economic | Future of work
Disagreed with
– Jonas Prising
Disagreed on
Speed and nature of AI adoption and technological change
AI commoditizes expertise and can be an equalizer by removing entry barriers, with bottom 50 percentile workers seeing 36% productivity gains
Explanation
AI democratizes access to expertise and disproportionately benefits lower-skilled workers by providing them with capabilities previously available only to experts. This can reduce inequality rather than increase it.
Evidence
Bottom 50 percentile workers had 36% productivity bump while top 50 percentile only had 17% bump. Hired more school graduates than ever before.
Major discussion point
AI’s Impact on Productivity and Employment
Topics
Economic | Future of work | Development
Disagreed with
– Laura D’Andrea Tyson
– Elizabeth Shuler
Disagreed on
AI as equalizer versus creator of inequality
Entry-level workers adapt more easily as they learn new AI-integrated processes from the start
Explanation
Young workers entering the workforce have an advantage because they learn AI-integrated workflows as their baseline, without needing to unlearn previous methods. This makes the transition to AI-augmented work more natural.
Evidence
Early career workers don’t know how old processes were done, making transition easier. Code-assist platforms enabling synchronous and asynchronous work with agents.
Major discussion point
Worker Transition and Skills Development
Topics
Economic | Future of work
Disagreed with
– Erik Brynjolfsson
– Jonas Prising
Disagreed on
Impact of AI on young workers and employment patterns
Business reinvention and process redesign are necessary rather than just applying AI to existing bad processes
Explanation
Maximum productivity gains from AI require fundamental reimagining of business processes and workflows, not simply overlaying AI technology on existing inefficient systems. This reinvention is where the biggest gains historically come from.
Evidence
Need for context engineering, integrating workflows between machines and humans, and reinventing business processes rather than just eliminating work or applying AI to existing bad processes.
Major discussion point
Augmentation vs. Automation Strategies
Topics
Economic | Future of work
Agreed with
– Erik Brynjolfsson
– Jonas Prising
– Elizabeth Shuler
Agreed on
Augmentation approach is superior to pure automation
Higher productivity potential exists in agriculture, manufacturing, healthcare, and construction sectors
Explanation
While knowledge industries show high AI exposure, sectors like agriculture, manufacturing, and construction have low current exposure but high potential for rapid change and job creation through productivity improvements.
Evidence
Exposure scores are very low in construction, transportation, but velocity of change will be very high. Software development already has 70% exposure so rate of change will be lower. $4.5 trillion of $15 trillion US labor economy is theoretically exposed to AI.
Major discussion point
Sectoral Applications and Global Development
Topics
Economic | Future of work | Development
Laura D’Andrea Tyson
Speech speed
156 words per minute
Speech length
1219 words
Speech time
468 seconds
Past technological revolutions haven’t led to long-term unemployment but change job composition over 40-year periods
Explanation
Historical evidence shows that technological advances eliminate some jobs while creating others, with the net effect being job transformation rather than permanent job loss. However, this transition occurs over decades, not years.
Evidence
Reference to David Autor’s work showing new job creation over 40-year processes, not 40-day or four-year processes.
Major discussion point
Historical Context and Future of Technological Unemployment
Topics
Economic | Future of work
Agreed with
– Erik Brynjolfsson
– Valdis Dombrovskis
– Jonas Prising
Agreed on
AI will not lead to long-term jobless growth but will transform job composition
Strong aggregate demand and tight business-education partnerships are essential for successful transitions
Explanation
Successful management of technological transitions requires robust macroeconomic policy to maintain strong labor markets, combined with close collaboration between educational institutions and businesses to align training with actual job needs.
Evidence
California’s use of community colleges for training with tight links to business community. Need for policymakers to work with companies like Ravi’s to understand job creation patterns.
Major discussion point
Worker Transition and Skills Development
Topics
Economic | Future of work | Development
Agreed with
– Valdis Dombrovskis
– Jonas Prising
– Elizabeth Shuler
Agreed on
Skills development and retraining are critical for successful AI transition
Productivity gains historically go to capital rather than labor, creating polarization and declining labor share of income
Explanation
The digital revolution demonstrates that while productivity increases, the benefits disproportionately flow to capital owners rather than workers, leading to wage stagnation and increased inequality despite overall economic growth.
Evidence
Digital revolution showing significant polarization of labor market, loss of middle skill jobs, real wage growth not keeping up with productivity growth, and declining labor share of income.
Major discussion point
Distribution of Productivity Benefits and Inequality
Topics
Economic | Future of work | Human rights
Disagreed with
– Ravi Kumar S.
– Elizabeth Shuler
Disagreed on
AI as equalizer versus creator of inequality
Edge AI applications in agriculture, healthcare, and tourism can help developing nations increase productivity
Explanation
Sector-specific AI applications designed for particular industries can significantly boost productivity in developing countries, particularly in key areas like agriculture and healthcare that are crucial for poverty reduction.
Evidence
World Bank president’s presentation on edge AI models helping small-scale, low-income farmers in Africa, preventing farm abandonment and urban migration. Applications in healthcare and tourism sectors.
Major discussion point
Sectoral Applications and Global Development
Topics
Development | Economic | Future of work
Valdis Dombrovskis
Speech speed
140 words per minute
Speech length
597 words
Speech time
254 seconds
EU views AI as opportunity to address productivity lag, with one-third of job vacancies requiring AI skills
Explanation
The European Union sees AI as crucial for catching up with other major economies in productivity growth, particularly given that the EU has been lagging for decades primarily due to technology sector gaps.
Evidence
EU lagging in productivity for decades largely due to tech sector. AI Continent Action Plan including AI gigafactories. OECD data showing one-third of job vacancies highly exposed to AI, expected to increase to two-thirds.
Major discussion point
AI’s Impact on Productivity and Employment
Topics
Economic | Future of work | Development
No evidence of long-term jobless growth, but job displacement will occur requiring proper transition management
Explanation
While aggregate employment levels should remain stable, individual professions will change and some jobs will be eliminated, making transition management and reskilling crucial for affected workers.
Evidence
EU employment continues to grow with unemployment at historically low levels. Aging continent with shrinking labor supply requiring productivity increases to sustain prosperity.
Major discussion point
Historical Context and Future of Technological Unemployment
Topics
Economic | Future of work
Agreed with
– Erik Brynjolfsson
– Laura D’Andrea Tyson
– Jonas Prising
Agreed on
AI will not lead to long-term jobless growth but will transform job composition
Massive workforce retraining is occurring due to AI skills shortage
Explanation
Companies are extensively retraining their existing workforce because there aren’t enough graduates with necessary AI skills entering the labor market, highlighting the critical importance of skills development.
Evidence
Microdata showing companies massively retraining workforce due to insufficient graduates with AI skills. Previous experience with digital skills shortage.
Major discussion point
Worker Transition and Skills Development
Topics
Economic | Future of work | Development
Agreed with
– Jonas Prising
– Laura D’Andrea Tyson
– Elizabeth Shuler
Agreed on
Skills development and retraining are critical for successful AI transition
Jonas Prising
Speech speed
167 words per minute
Speech length
1059 words
Speech time
379 seconds
AI skills training focuses on prompt writing, document analysis, and human-AI collaboration workflows
Explanation
Companies are taking a pragmatic approach to AI training, treating it like the evolution of basic computer skills, focusing on practical abilities like prompt engineering and document analysis rather than complex technical skills.
Evidence
Training comparable to Word, Excel, PowerPoint evolution. Skills include writing good prompts, integrating and analyzing documents, and augmenting capabilities with large language models. Job candidates listing AI skills like they previously listed Microsoft Office proficiency.
Major discussion point
Worker Transition and Skills Development
Topics
Economic | Future of work
Agreed with
– Valdis Dombrovskis
– Laura D’Andrea Tyson
– Elizabeth Shuler
Agreed on
Skills development and retraining are critical for successful AI transition
Disagreed with
– Erik Brynjolfsson
– Ravi Kumar S.
Disagreed on
Impact of AI on young workers and employment patterns
Technology progress speed is often overestimated, with human preferences maintaining roles even when automation is possible
Explanation
Historical patterns show that society often overestimates the speed of technological adoption, and humans prefer maintaining human oversight even when technology could theoretically replace workers entirely.
Evidence
Examples of blockchain hype lasting one year, driverless cars still having drivers in cabins, airplanes having three pilots despite evidence that pilots cause more issues than technology.
Major discussion point
Historical Context and Future of Technological Unemployment
Topics
Economic | Future of work
Agreed with
– Erik Brynjolfsson
– Laura D’Andrea Tyson
– Valdis Dombrovskis
Agreed on
AI will not lead to long-term jobless growth but will transform job composition
Disagreed with
– Ravi Kumar S.
Disagreed on
Speed and nature of AI adoption and technological change
Large enterprises will handle reskilling, but policy focus needed on majority of smaller employers
Explanation
While large companies have the resources and competitive pressure to retrain their workforce, the majority of employers are smaller organizations that lack these capabilities, requiring policy intervention to ensure widespread reskilling.
Evidence
Large enterprises will retrain, reskill, upskill their people because they’re competing and have resources. Need to worry about all other employers which are majority of employment providers.
Major discussion point
Augmentation vs. Automation Strategies
Topics
Economic | Future of work | Development
Elizabeth Shuler
Speech speed
163 words per minute
Speech length
846 words
Speech time
311 seconds
Current economy isn’t working for workers with 40% lacking $400 for emergencies, and AI adds to existing insecurity
Explanation
Workers are already struggling with economic insecurity and inequality, making them naturally anxious about AI implementation that occurs without their input or training. This existing vulnerability amplifies concerns about technological change.
Evidence
AFL-CIO represents 15 million workers across 64 unions. 40% of US workers don’t have $400 for emergency. People working two and three jobs just to keep up. Inequality at highest levels.
Major discussion point
Distribution of Productivity Benefits and Inequality
Topics
Economic | Future of work | Human rights
Disagreed with
– Ravi Kumar S.
– Laura D’Andrea Tyson
Disagreed on
AI as equalizer versus creator of inequality
Working people need to share in productivity gains through policy changes and redistribution mechanisms
Explanation
If AI generates significant productivity improvements, workers who enable these gains should benefit through policy mechanisms that ensure fair distribution of the economic benefits rather than having gains flow only to capital owners.
Evidence
Need for policies on tax infrastructure and redistribution to ensure working people share in technology gains. Reference to productivity gains needing to benefit those who make industries happen.
Major discussion point
Distribution of Productivity Benefits and Inequality
Topics
Economic | Future of work | Human rights
Job quality matters more than quantity – need jobs that can sustain families with single employment
Explanation
The focus should not just be on creating jobs but on ensuring these jobs provide sufficient income and benefits to support families without requiring multiple employment, addressing the quality of work rather than just availability.
Evidence
Emphasis that one job should be enough to sustain a family, contrasting with current reality of people working multiple jobs.
Major discussion point
Distribution of Productivity Benefits and Inequality
Topics
Economic | Future of work | Human rights
Successful transitions require worker-centered approaches with training partnerships and worker involvement upstream
Explanation
Rather than having workers deal with technological change after implementation, they should be involved from the beginning in designing and implementing AI systems, with training partnerships that include worker input rather than top-down approaches.
Evidence
Partnership with Microsoft to get workers in labs, example of bus drivers working with developers on transportation AI. Reference to past industrial revolutions where working people helped tame technology.
Major discussion point
Worker Transition and Skills Development
Topics
Economic | Future of work | Human rights
Agreed with
– Valdis Dombrovskis
– Jonas Prising
– Laura D’Andrea Tyson
Agreed on
Skills development and retraining are critical for successful AI transition
Workers need voice in how AI is implemented, with collective bargaining as a tool for data usage decisions
Explanation
Workers should have input into how AI systems evaluate their performance and use their data, with collective bargaining providing a mechanism to negotiate terms around privacy, data security, and AI implementation in workplaces.
Evidence
WNBA players example where AI judges player decisions and game performance, raising questions about professional judgment versus AI evaluation. Collective bargaining as tool for determining data usage.
Major discussion point
Augmentation vs. Automation Strategies
Topics
Economic | Future of work | Human rights | Privacy and data protection
Audience
Speech speed
138 words per minute
Speech length
141 words
Speech time
61 seconds
Shared productivity gains model could replace traditional development assistance frameworks
Explanation
Instead of traditional foreign aid structures, a new model could emerge where productivity gains from AI in affluent societies are shared with developing nations, creating a more sustainable and measurable approach to global development.
Evidence
Official development assistance was $212 billion globally last year but has been depleted. Productivity gains from doctors, nurses, teachers, farmers in affluent societies could be shared and are measurable and evidenceable.
Major discussion point
Sectoral Applications and Global Development
Topics
Development | Economic | Future of work
Agreements
Agreement points
AI will not lead to long-term jobless growth but will transform job composition
Speakers
– Erik Brynjolfsson
– Laura D’Andrea Tyson
– Valdis Dombrovskis
– Jonas Prising
Arguments
Past technological revolutions haven’t led to long-term unemployment but change job composition over 40-year periods
No evidence of long-term jobless growth, but job displacement will occur requiring proper transition management
Technology progress speed is often overestimated, with human preferences maintaining roles even when automation is possible
Summary
All speakers agree that while AI will displace some jobs and transform others, historical evidence and current trends suggest no permanent mass unemployment, though managing the transition period is crucial
Topics
Economic | Future of work
Augmentation approach is superior to pure automation
Speakers
– Erik Brynjolfsson
– Ravi Kumar S.
– Jonas Prising
– Elizabeth Shuler
Arguments
Companies should focus on augmenting human capabilities rather than replacing workers entirely
Business reinvention and process redesign are necessary rather than just applying AI to existing bad processes
AI skills training focuses on prompt writing, document analysis, and human-AI collaboration workflows
Successful transitions require worker-centered approaches with training partnerships and worker involvement upstream
Summary
Speakers consensus that AI should enhance human capabilities rather than replace workers, requiring fundamental redesign of workflows and processes with worker involvement
Topics
Economic | Future of work | Human rights
Skills development and retraining are critical for successful AI transition
Speakers
– Valdis Dombrovskis
– Jonas Prising
– Laura D’Andrea Tyson
– Elizabeth Shuler
Arguments
Massive workforce retraining is occurring due to AI skills shortage
AI skills training focuses on prompt writing, document analysis, and human-AI collaboration workflows
Strong aggregate demand and tight business-education partnerships are essential for successful transitions
Successful transitions require worker-centered approaches with training partnerships and worker involvement upstream
Summary
All speakers emphasize the critical importance of comprehensive retraining and skills development programs, with strong partnerships between businesses, educational institutions, and workers
Topics
Economic | Future of work | Development
Similar viewpoints
Both present research showing that AI can benefit lower-skilled workers more than higher-skilled ones when used for augmentation, though young workers in exposed occupations face employment challenges
Speakers
– Erik Brynjolfsson
– Ravi Kumar S.
Arguments
Young workers (22-26) in AI-exposed occupations show 13-16% employment decline, while augmentation users see employment growth
AI commoditizes expertise and can be an equalizer by removing entry barriers, with bottom 50 percentile workers seeing 36% productivity gains
Topics
Economic | Future of work
Both emphasize concerns about inequality and the need for policy interventions to ensure workers benefit from productivity gains rather than having benefits flow only to capital owners
Speakers
– Laura D’Andrea Tyson
– Elizabeth Shuler
Arguments
Productivity gains historically go to capital rather than labor, creating polarization and declining labor share of income
Working people need to share in productivity gains through policy changes and redistribution mechanisms
Topics
Economic | Future of work | Human rights
Both see significant potential for AI applications in traditional sectors like agriculture and healthcare, particularly for development and productivity improvements beyond knowledge work
Speakers
– Ravi Kumar S.
– Laura D’Andrea Tyson
Arguments
Higher productivity potential exists in agriculture, manufacturing, healthcare, and construction sectors
Edge AI applications in agriculture, healthcare, and tourism can help developing nations increase productivity
Topics
Development | Economic | Future of work
Unexpected consensus
AI as an equalizer rather than a source of inequality
Speakers
– Ravi Kumar S.
– Jonas Prising
Arguments
AI commoditizes expertise and can be an equalizer by removing entry barriers, with bottom 50 percentile workers seeing 36% productivity gains
AI skills training focuses on prompt writing, document analysis, and human-AI collaboration workflows
Explanation
Surprisingly, business leaders argued that AI could reduce rather than increase inequality by democratizing access to expertise and disproportionately benefiting lower-skilled workers, contrasting with common fears about AI exacerbating inequality
Topics
Economic | Future of work
Need for worker involvement in AI development and implementation
Speakers
– Elizabeth Shuler
– Ravi Kumar S.
– Jonas Prising
Arguments
Workers need voice in how AI is implemented, with collective bargaining as a tool for data usage decisions
Business reinvention and process redesign are necessary rather than just applying AI to existing bad processes
Large enterprises will handle reskilling, but policy focus needed on majority of smaller employers
Explanation
Unexpected alignment between labor union leadership and business executives on the importance of worker participation in AI implementation, suggesting shared recognition that top-down approaches are insufficient
Topics
Economic | Future of work | Human rights
Overall assessment
Summary
Strong consensus emerged around three main areas: AI will transform rather than eliminate jobs long-term, augmentation approaches are superior to pure automation, and comprehensive skills development is essential. Speakers also showed surprising agreement on AI’s potential as an equalizer and the need for worker involvement in implementation.
Consensus level
High level of consensus with significant implications for policy and business strategy. The agreement across diverse stakeholders (academics, business leaders, policymakers, and labor representatives) suggests a mature understanding of AI’s challenges and opportunities, providing a foundation for coordinated action on managing the AI transition while ensuring shared prosperity.
Differences
Different viewpoints
Impact of AI on young workers and employment patterns
Speakers
– Erik Brynjolfsson
– Ravi Kumar S.
– Jonas Prising
Arguments
Young workers (22-26) in AI-exposed occupations show 13-16% employment decline, while augmentation users see employment growth
Entry-level workers adapt more easily as they learn new AI-integrated processes from the start
AI skills training focuses on prompt writing, document analysis, and human-AI collaboration workflows
Summary
Brynjolfsson presents data showing significant employment decline among young workers in AI-exposed jobs, while Ravi Kumar argues that entry-level workers actually benefit more from AI and his company hired more graduates than ever. Prising suggests the decline may be due to normal economic cycles rather than AI specifically.
Topics
Economic | Future of work
Speed and nature of AI adoption and technological change
Speakers
– Ravi Kumar S.
– Jonas Prising
Arguments
Technology adoption requires 8-10 years for value realization, with AI being probabilistic rather than deterministic, requiring contextual engineering
Technology progress speed is often overestimated, with human preferences maintaining roles even when automation is possible
Summary
Ravi Kumar emphasizes the complexity and time required for AI implementation due to its probabilistic nature, while Prising argues that society consistently overestimates the speed of technological adoption and that human preferences will maintain many roles.
Topics
Economic | Future of work
AI as equalizer versus creator of inequality
Speakers
– Ravi Kumar S.
– Laura D’Andrea Tyson
– Elizabeth Shuler
Arguments
AI commoditizes expertise and can be an equalizer by removing entry barriers, with bottom 50 percentile workers seeing 36% productivity gains
Productivity gains historically go to capital rather than labor, creating polarization and declining labor share of income
Current economy isn’t working for workers with 40% lacking $400 for emergencies, and AI adds to existing insecurity
Summary
Ravi Kumar sees AI as democratizing expertise and benefiting lower-skilled workers more, while Tyson warns about historical patterns of productivity gains flowing to capital owners, and Shuler emphasizes that AI adds to existing worker insecurity and inequality.
Topics
Economic | Future of work | Human rights
Unexpected differences
Optimism about AI’s impact on entry-level workers despite concerning data
Speakers
– Erik Brynjolfsson
– Ravi Kumar S.
Arguments
Young workers (22-26) in AI-exposed occupations show 13-16% employment decline, while augmentation users see employment growth
Entry-level workers adapt more easily as they learn new AI-integrated processes from the start
Explanation
Despite Brynjolfsson presenting data showing significant employment decline among young workers in AI-exposed occupations, Ravi Kumar maintains optimism about entry-level workers, claiming his company hired more graduates than ever. This disagreement is unexpected because both are technology leaders who should be seeing similar trends.
Topics
Economic | Future of work
Role of collective bargaining in AI implementation
Speakers
– Elizabeth Shuler
– Other panelists
Arguments
Workers need voice in how AI is implemented, with collective bargaining as a tool for data usage decisions
Various arguments about training partnerships and business-led approaches
Explanation
Shuler’s emphasis on collective bargaining and worker rights in AI implementation received little direct engagement from other panelists, who focused more on training and business-led solutions. This represents an unexpected divide between labor and business/academic perspectives on worker agency.
Topics
Economic | Future of work | Human rights | Privacy and data protection
Overall assessment
Summary
The panel showed surprising consensus on avoiding jobless growth and the benefits of augmentation over automation, but significant disagreements emerged on the current impact on young workers, the timeline and complexity of AI adoption, and whether AI will reduce or increase inequality. The most notable divide was between optimistic business perspectives and more cautious academic/labor viewpoints.
Disagreement level
Moderate disagreement with significant implications. While speakers agreed on broad goals, their different assessments of current impacts and future trajectories could lead to very different policy recommendations and business strategies. The disconnect between data showing job losses and business optimism about hiring suggests either different measurement approaches or selective reporting that needs resolution for effective policymaking.
Partial agreements
Partial agreements
Similar viewpoints
Both present research showing that AI can benefit lower-skilled workers more than higher-skilled ones when used for augmentation, though young workers in exposed occupations face employment challenges
Speakers
– Erik Brynjolfsson
– Ravi Kumar S.
Arguments
Young workers (22-26) in AI-exposed occupations show 13-16% employment decline, while augmentation users see employment growth
AI commoditizes expertise and can be an equalizer by removing entry barriers, with bottom 50 percentile workers seeing 36% productivity gains
Topics
Economic | Future of work
Both emphasize concerns about inequality and the need for policy interventions to ensure workers benefit from productivity gains rather than having benefits flow only to capital owners
Speakers
– Laura D’Andrea Tyson
– Elizabeth Shuler
Arguments
Productivity gains historically go to capital rather than labor, creating polarization and declining labor share of income
Working people need to share in productivity gains through policy changes and redistribution mechanisms
Topics
Economic | Future of work | Human rights
Both see significant potential for AI applications in traditional sectors like agriculture and healthcare, particularly for development and productivity improvements beyond knowledge work
Speakers
– Ravi Kumar S.
– Laura D’Andrea Tyson
Arguments
Higher productivity potential exists in agriculture, manufacturing, healthcare, and construction sectors
Edge AI applications in agriculture, healthcare, and tourism can help developing nations increase productivity
Topics
Development | Economic | Future of work
Takeaways
Key takeaways
AI shows significant productivity gains (14-35% increases) but requires 8-10 years for full value realization due to its probabilistic nature requiring contextual engineering and business process reinvention
Young workers (22-26) in AI-exposed occupations are experiencing 13-16% employment decline, while workers using AI for augmentation rather than replacement see employment growth
AI can serve as an equalizer by commoditizing expertise and removing entry barriers, with less skilled workers often seeing larger productivity gains than highly skilled workers
Historical evidence shows technological revolutions don’t create long-term unemployment but do cause significant transition disruption and job composition changes over 40-year periods
The key challenge is ensuring productivity benefits are shared equitably rather than concentrated in capital, as seen in previous digital revolutions that increased inequality
Successful AI implementation requires augmentation strategies (humans and machines working together) rather than pure automation, necessitating new benchmarks and incentive structures
Worker-centered transitions with upstream involvement, training partnerships, and collective bargaining are essential for managing AI adoption successfully
Sectors like agriculture, manufacturing, healthcare, and construction offer higher productivity potential than knowledge work, where AI exposure is already high
Resolutions and action items
Develop ‘centaur benchmarks’ that measure human-AI collaboration effectiveness rather than AI-only performance
Create stronger partnerships between educational institutions (especially community colleges) and businesses to align training with actual job market needs
Implement AI skills training focused on prompt writing, document analysis, and human-AI workflow integration
Establish worker involvement upstream in AI development process, including having workers in labs during technology design
Focus policy attention on supporting smaller employers with reskilling resources, as large enterprises can handle this internally
Develop shared productivity gains frameworks that could replace traditional development assistance models
Create tax and policy incentives that favor augmentation over automation strategies
Unresolved issues
How to structurally incentivize companies to choose augmentation over automation when labor costs are high and automation is cheaper
What specific policy mechanisms should be used to redistribute productivity gains from capital to workers
How to manage the transition for workers in highly AI-exposed occupations who cannot easily move to augmentation roles
What guardrails and regulations are needed to prevent AI from ‘running over’ workers while still enabling innovation
How to address privacy and data security concerns, particularly regarding AI evaluation of worker performance and decision-making
How to scale successful transition models beyond large enterprises to the majority of smaller employers
What lessons from past failed transitions (like manufacturing job losses) should specifically inform AI transition policies
Suggested compromises
Balance AI development between pure automation and augmentation by redesigning workflows to integrate both human and machine capabilities
Combine strong aggregate demand policies with targeted reskilling programs to manage transition effects
Use collective bargaining as a mechanism to give workers voice in AI implementation while allowing companies to pursue productivity gains
Focus initial AI applications on sectors with lower current exposure (agriculture, construction, healthcare) while managing transitions in high-exposure sectors (software, call centers)
Develop edge AI applications for specific sectors rather than focusing solely on large language models, allowing for more targeted worker integration
Create training programs that treat AI skills as infrastructure (like Microsoft Office skills) while maintaining human judgment and professional expertise
Thought provoking comments
Productivity is output divided by input. And for some reason, a lot of people focus on the numerator, and other people focus on the denominator… as you increase productivity, that ratio of output to input, you can either have the numerator grow or you can have the denominator shrink or just lay flat.
Speaker
Erik Brynjolfsson
Reason
This framing recontextualizes the entire productivity debate by showing that the optimistic growth narrative and pessimistic job loss narrative are literally two sides of the same mathematical equation. It’s insightful because it reveals why these discussions often talk past each other – people are focusing on different parts of the same formula.
Impact
This mathematical framework became the conceptual foundation for the entire discussion, with multiple panelists returning to this distinction between growing the numerator (output/value creation) versus shrinking the denominator (job elimination). It established the central tension the panel would grapple with throughout.
We found that the augmenters, those who are learning new things, had growing employment. So they both had more output and more employment, so more shared prosperity. Sadly, that was a minority of the folks, but if we can get more people doing that, maybe that’s a potential path.
Speaker
Erik Brynjolfsson
Reason
This empirical finding is crucial because it provides concrete evidence that the augmentation path can lead to both productivity gains AND job growth – solving the core dilemma. However, the honest admission that this represents only a minority creates urgency around the question of how to scale this approach.
Impact
This finding became a central reference point for multiple panelists, particularly Ravi Kumar and Jonas Prising, who built upon it with their own examples. It shifted the discussion from whether augmentation works to how to make it the dominant approach rather than the exception.
This is not about applying technology to old stuff we already had… We have to reinvent the business, reinvent the process, reinvent the flow, and that reinvention and reimagination will drive productivity.
Speaker
Ravi Kumar S.
Reason
This challenges the common assumption that AI adoption is simply about making existing processes more efficient. Instead, it argues for fundamental business model transformation, which is a much more complex but potentially more rewarding approach.
Impact
This comment elevated the discussion beyond simple automation versus augmentation to consider systemic organizational change. It influenced later comments about the difficulty of implementation and why productivity gains have been slower than expected, while also explaining why the transition is so challenging for workers.
If you look at the technological revolution, say the digital revolution… you see very significant polarization of the labor market… The productivity surplus is going to capital. It’s going to capital.
Speaker
Laura D’Andrea Tyson
Reason
This introduces a critical historical perspective that challenges the optimistic narrative. By pointing to concrete evidence from the digital revolution showing that productivity gains don’t automatically translate to shared prosperity, it forces the discussion to confront distributional issues.
Impact
This comment fundamentally shifted the conversation from focusing primarily on job quantity to job quality and benefit distribution. It prompted Liz Shuler’s later comments about inequality and influenced the discussion toward policy interventions and structural changes needed to ensure shared prosperity.
For why does a company decide to use the technology to augment skills versus to automate skills? If you have a high percentage of labor costs in your overall cost structure… your incentive is to automate it away, I believe.
Speaker
Laura D’Andrea Tyson
Reason
This cuts to the heart of the implementation challenge by identifying the structural economic incentives that drive companies toward automation rather than augmentation, regardless of what might be better for society or even long-term business success.
Impact
This question forced the panel to grapple with the gap between what’s theoretically possible (augmentation leading to shared prosperity) and what’s economically incentivized (automation for cost reduction). It led to Erik’s later discussion of ‘The Turing Trap’ and the need to redesign incentive structures.
We represent all of the people that you’re talking about… We are seeing just as a backdrop, the economy, at least in the US and around the world, isn’t working for working people now… Now, put AI on top of that, the insecurity that we’re all experiencing… Of course, they’re gonna be anxious.
Speaker
Elizabeth Shuler
Reason
This grounds the entire theoretical discussion in current economic reality, pointing out that workers are already struggling with inequality and insecurity before AI even enters the picture. It reframes AI not as a standalone challenge but as an additional stressor on an already strained system.
Impact
This comment shifted the tone and focus of the discussion significantly, moving from primarily technical and economic considerations to human and social ones. It led to more discussion about worker voice, collective bargaining, and the need for worker-centered transitions, fundamentally changing the frame from ‘how do we implement AI?’ to ‘how do we implement AI in a way that serves working people?’
Can anyone give me an example of a transition that really went well? I think we should learn from our past… when we lost manufacturing, hollowed out the industrial Midwest, and left workers behind. That did not go well. Why are we not learning the lessons from these previous transitions?
Speaker
Elizabeth Shuler
Reason
This is a powerful challenge to the assumption that technological transitions naturally work out well over time. By demanding concrete examples of successful transitions and highlighting a major failure, it forces the panel to confront the gap between economic theory and historical reality.
Impact
This comment served as a reality check that influenced the closing remarks of several panelists. It shifted the discussion from optimistic assumptions about natural adaptation to urgent questions about how to actively manage transitions better than in the past.
Overall assessment
These key comments fundamentally shaped the discussion by creating a progression from technical framing to human-centered concerns. Brynjolfsson’s mathematical framing established the conceptual foundation, while his augmentation findings provided a potential solution path. Kumar’s business reinvention perspective added complexity about implementation challenges. Tyson’s historical analysis and structural incentive questions introduced critical skepticism about automatic positive outcomes. Finally, Shuler’s worker-centered interventions grounded the entire discussion in current economic reality and historical failures, forcing a shift from theoretical possibilities to practical implementation challenges. Together, these comments moved the conversation from ‘can AI boost productivity without eliminating jobs?’ to ‘how do we restructure incentives, policies, and business practices to ensure AI-driven productivity gains are shared rather than concentrated?’ The discussion evolved from technical optimism to nuanced recognition of the structural changes needed for equitable outcomes.
Follow-up questions
How can we better understand what drives companies to choose augmentation versus automation strategies when implementing AI?
Speaker
Laura D’Andrea Tyson
Explanation
This is crucial for policy design as companies with high labor costs in low-skill categories have natural incentives to automate rather than augment, which could lead to job displacement rather than job enhancement
What are specific examples of technological transitions that went well historically, and what lessons can we learn from them?
Speaker
Elizabeth Shuler
Explanation
Understanding successful transitions is essential for designing better policies and approaches for the current AI transition, especially given past failures like the hollowing out of the industrial Midwest
How can we develop better data and statistics to track the changing nature of work due to AI?
Speaker
Erik Brynjolfsson
Explanation
Better visibility into how the world is changing is needed to make informed policy decisions and track the real impacts of AI on employment
How can we scale reskilling and upskilling beyond large enterprises to smaller employers who lack resources?
Speaker
Jonas Prising
Explanation
While large enterprises can retrain their workforce, the majority of employment providers are smaller organizations that may lack the resources for comprehensive reskilling programs
How can we design tax and policy incentives to encourage augmentation over automation?
Speaker
Erik Brynjolfsson
Explanation
Current incentive structures may be steering companies toward automation rather than augmentation, and policy changes could help redirect this toward more beneficial outcomes for workers
How can we develop and implement ‘centaur benchmarks’ that measure human-AI collaboration rather than AI replacement?
Speaker
Erik Brynjolfsson
Explanation
Current AI benchmarks focus on how well AI can replace humans rather than how well humans and AI can work together, which could be steering technology development in the wrong direction
How can we create frameworks for sharing productivity gains globally, particularly for development assistance?
Speaker
Simon O’Connell (Audience)
Explanation
There’s potential to move beyond traditional development assistance models to ones based on measurable shared productivity gains from AI implementation
How can we ensure data privacy and security while implementing AI in workplaces, particularly regarding worker evaluation and decision-making?
Speaker
Elizabeth Shuler
Explanation
Workers are concerned about how AI systems evaluate their performance and decision-making, and collective bargaining may be needed to determine appropriate use of worker data
What specific training partnerships and guardrails are needed to ensure worker-centered AI transitions?
Speaker
Elizabeth Shuler
Explanation
Moving beyond top-down training approaches to include workers in the design and implementation of AI systems requires specific frameworks and protective measures
How can we better measure and ensure job quality, not just job quantity, in the AI transition?
Speaker
Elizabeth Shuler
Explanation
The focus on job creation numbers may miss important questions about whether new jobs can sustain families and provide adequate income
Disclaimer: This is not an official session record. DiploAI generates these resources from audiovisual recordings, and they are presented as-is, including potential errors. Due to logistical challenges, such as discrepancies in audio/video or transcripts, names may be misspelled. We strive for accuracy to the best of our ability.
Related event
World Economic Forum Annual Meeting 2026 at Davos
19 Jan 2026 08:00h - 23 Jan 2026 18:00h
Davos, Switzerland