Artificial Intelligence in the Office and the Factory Transcript

TRACEY MORANT ADAMS: Good morning, everyone. I am Doctor Tracey Morant Adams, senior executive vice president and chief corporate social responsibility officer for Renasant Bank. And I'm honored to join you and to open today's program, Artificial Intelligence in the Office and the Factory. I want to thank the Chicago Fed's Economic Mobility Project for bringing us together for this timely and very important conversation. As leaders, we recognize that artificial intelligence is not just another innovation. It is a profound shift in how work is organized, how people build careers, and how communities prepare for the future.

Whether we sit in boardrooms, classrooms, in factories, or offices, we are all participants in a new economy that is being created in real time. At Renasant Bank, we've seen firsthand how technology can both expand and challenge opportunity. Our work in community development and corporate social responsibility focuses on ensuring that change, no matter how rapid, the change, is also inclusive. Because when innovation happens faster than preparation, the gap between potential and access widens.

That is why conversations like today's are so essential. They allow us to ask the hard but necessary questions. How do we prepare workers for roles that do not yet exist? How do we make sure that the benefits of all AI adoption reach not just firms and developers, but also employees and communities? And how do we build systems of training, mentorship, and representation that ensure no one is left behind as workplaces evolve?

The research presentation will hear today from Doctor Gustavo de Souza provides valuable insight into those dynamics, showing how AI is being adopted across both office and industrial settings, reshaping not only what we do, but also how we define skill and value. And I'm especially looking forward to hearing from our panelists, leaders from academia, industry and advocacy, who will help us think through how these findings translate late into practice. From hiring and workforce training to innovation and equity, these perspectives will help us bridge research with real world AI impact.

As a banker, a thought leader, and someone deeply committed to community advancement, I believe this moment calls for both courage and collaboration. Technology will continue to evolve. And our responsibility is to ensure that progress remains people centered.

Thank you again to the Chicago Fed for convening this presentation and dialogue. It is now my pleasure to turn the program over to Doctor Gustavo de Souza.

GUSTAVO DE SOUZA: Thanks, Tracey, I'll just share my screen. And here it is. First, I have to make the disclaimer that this talk and this paper represents my opinions and not those of the Federal Reserve Bank of Chicago or the Federal Reserve System.

So this paper and this talk is about trying to understand how AI is already affecting the labor market. If you are like myself, in 2022, you got access to ChatGPT. And your mind blew away about what that thing can do. And you probably start asking questions like, will this technology still my job? Because I'm an economist, I asked perhaps a bit broader question of, what is the effect of this in the labor market?

[INAUDIBLE] after studying a little bit more about artificial intelligence, I start understanding that AI is actually way broader than generative AI. AI, or the field of AI, what it does is use large amounts of data to identify patterns, make predictions, and from these predictions, generate actions. Generative AI, like ChatGPT, is one application of that, but also have AI being used in self-driving cars, social media, facial recognition, and many other applications.

One application of AI is usually forgotten, is AI used in the production setting. And Jeff Bezos, discussing that, said in 2017, that I is quietly but meaningfully improving core operations in Amazon. And because AI used in production is going to be an important point of my talk, I want to give you a concrete view of what I'm talking about. So I'll give you one example.

Consider, for instance, this photo here that has a motor on it. If you look inside this circle, you're going to see a sensor. What this sensor does is collect data every minute on vibration, temperature, and sound produced by that motor. It then sent all that data to a software that, using AI, is going to predict when that machine is about to break. So then, before the machine breaks, the software sends a note to the maintenance workers at that company that they should do a particular type of maintenance to avoid the machine of breaking.

And this type of AI is called predictive maintenance AI. And scientists and engineers and economists, studying this particular application, have shown that, when they are installed in machines and factory floors, they allow machines to be running for longer because machines now break less often. And because machines are running for longer, the companies start hiring more workers to operate those machines.

So now you can understand the difficulty in answering the question of how AI is affecting the labor market because in one hand, we have AI, like the predictive maintenance software that I described to you before, that increase employment of workers. But perhaps, we also have other software types or other applications of AI that have a different effect, perhaps decreases employment.

For instance, Amazon is said to have reduced it its HR staff by 15% by using an HR AI. The same thing happened with another company that used an accounting AI. So then, we have AI potentially increasing employment in some areas, and decreasing employment in other areas. What I want to show you in this short talk is that when I look at how AI is affecting factory employment, I'm going to show it to you, the AI is increasing employment of production workers by making machines more productive and easier to operate. But then, when I look at workers performing administrative roles, like HR workers or accountants, there, I see I decreasing employment because it is automating some of the tasks done by those workers. And I got to this conclusion, the following way. And this is what I'm going to present to you and why.

So first, I'm going to collect a new data on AI development that gives me a lot of information on how companies are actually deploying AI tools. Then I'm going to match that to labor market outcomes of different occupations. And then, I'm going to break occupations down with occupations that are more exposed to AI, and occupations that are less exposed AI. And then, I show that overall, AI is increasing employment. But that really depends on which occupation you look, and the labor market characteristics because AI is increasing employment for production workers, but decreasing for workers working at the office.

Now, the challenge of answering this question about AI is really a measurement challenge because AI has many, many uses. So to understand how AI is affecting the labor market, we need really to measure it really well of how AI is being deployed. Now, to solve this challenge, I'm going to study the case of Brazil. And Brazil is very interesting to study this particular question because almost every commercial AI used in Brazil has a registration describing how that AI is used.

The registration looks just like this one here. This has important information for me on how that machine-- or how that software is being used. It has the title of the software, which tends to be really descriptive. And together with that, it has this topic called-- this variable called application domains, which is the areas within the company that are supposed to be using this AI software, and what exactly this AI software is supposed to do.

So then, what I do is the following. I go and take, create a database that has all the software registrations that have ever been created in Brazil. And then, I'm going to look at the software registration that use AI technology. And then, from that, I get a list for each AI software of what that software can do, and what it is supposed to do in each company that it is deployed.

Together with that, I'm going to do the same thing, but for workers, and for occupations. I construct a database that, for each occupation, has a list of the tasks that occupation performs. So for instance, if one of these occupations is in an economist, task one could be, construct code data. Task two could be, make a presentation, just like this one I'm making right now. And I construct that for every occupation that there is out there.

And then, using these two data sets, I construct this measure, which are called AI exposure, which is just the overlap between the share of tasks that each occupation performs, and the tasks, or the applications that the software do. So if there is a lot of overlap, then means that the occupation is very exposed to AI.

So for instance, that happens with instrumentation technicians. Because they operate a lot of machines, and many of these machines have AI software installed on them, they also have administrative assistants, which do a lot of data management or compiling data or giving information to different parts of the company. And a lot of what AI software does is really related to that. So those occupations, for instance, are going to be more exposed to AI. And then, we also have occupations that are less exposed to AI, in which this overlap between what they do and what the AI software does is not that big. For instance, janitors or construction helpers.

Now, what I'm going to do is, I'm going to look at the labor market outcomes of these different types of occupations, the ones that are more exposed and the ones that are less exposed. And this is what I'm plotting here in this figure in the left. I'm plotting the employment of-- in red-- occupations that are less exposed to AI, and in blue, occupations that are more exposed to AI.

What I want you to notice is that up to around 2013, employment in these two occupational groups grew by almost the same amount, or both of them declining by the same rate. But after 2013, the occupations that have more exposure to AI is where you see employment growing faster, differently, the occupations that are less exposed to AI. So this is the first message of this talk, that occupations are more exposed to AI actually faced an increase in employment. But after you should ask, is well, what has happened in 2013 that could explain this trend break.

And this figure here on the right gives you that answer. So this figure plots the number of different firms that own at least one AI technology. What I want you to notice from this figure is that after 2010, you see a very fast increase in the number of AI software, the number of firms with at least one AI software. This is what-- this is the period that people studying AI history, and some of them are in the panel today, called the AI boom, that scientists made a set of discovery, related to neural networks or and convoluted neural networks and deep neural networks, that allowed AI to be applied in many different tasks. And that's why we observed this boom, which on figure on the left shows, has already affected the labor market.

However, this average effect, they mask a lot of heterogeneity. And this figure is showing you a little bit of that. Here, I'm breaking down the effect of AI by different employment groups. If this number is larger, moving to the right, it means that the effect of AI is positive, increases employment. If this number is negative, means that the effect of AI is to decrease employment.

Right now, I'm plotting here, the effect of AI on production workers. There'll be manufacturing workers, agriculture workers, maintenance workers, and so on and so forth. What I want you to notice is that for all these occupations, AI is actually increasing employment. Where AI decreases employment, and by a lot, is among administrative workers, HR workers, accounting workers.

Then when I look at managers or service workers, this is where I don't find much of an effect. And then, you could ask, well, why is AI affecting administrative workers and production workers? And this figure helps us understand that. These figures break down the share of AI software in three groups, management applications, like HR, that gave examples to you before, production applications, like the predictive maintenance software that I showed you before, but also many others like manufacturing execution systems, or quality control. And the third one is academic applications like math or physics.

What I want you to notice from this figure is the first two bars. See that they are almost the same size? What that means is that AI is used as often in management applications as it is in production related applications. And this is why both type of workers are affected. Production workers, or workers in the factory, but also workers in the office.

What I showed you today, and just to conclude, is that in production roles, AI seems to be an increase in employment. A way to understand these results is that AI makes machines more productive because, remember, AI is just a software. Software is used, installed in computers and machines. So AI makes these machines that have this AI software installed on them more productive and easier to operate. And this is why employment goes up among production workers.

Among office workers, it looks like, that AI is simply replacing tasks that they previously were done, were doing. And this is why the employment goes down. The important implications of that is that the effect of AI really depends on the workforce composition. Economies with a large share of office workers are going to lose a lot more employment than gain from AI. And that, perhaps, is what happens in some areas in the US. And what the government should do is create policies that helps workers transition from occupations that are being replaced by AI to occupations that use AI, like those related to production.

And just a final thought before I let Kirsten take, these results paint a more positive view of AI, which is very different than the way that we usually think about AI because when we think about AI, the first thing that comes to mind is a world just like "Terminator," in which AI is going to take out our jobs and perhaps destroy us and kill us after that. But these results, what I'm actually showing you in a more realistic view of AI, according to that, is that AI makes machines more productive and simple to operate, which increases employment.

So if I had to compare these results with a movie, it wouldn't be terminator. It would be something like "Ironman." And that's what I have for today. Thanks a lot, and Kristen, you can take from here.

KRISTEN BROADY: Thank you so much, Doctor de Souza. That was a really interesting presentation. My name is Kristen Broady. I am the director of the Economic Mobility Project here at the Federal Reserve Bank of Chicago. I have the pleasure of introducing our panelists. First, doctor Ajay Agrawal is one of the most-- foremost voices on economics of artificial intelligence and the intersection of technology, entrepreneurship, and public policy. At the University of Toronto's Rotman School of Management, he holds the Geoffrey Taber Chair in Entrepreneurship and Innovation and is the founder of the globally recognized Creative Destruction Lab.

Doctor J. Edward Colgate is a pioneering figure in haptics and human robot interaction, recognized globally for sharing how people physically engage with machines. A long time faculty member at Northwestern University's McCormick School of Engineering, he holds the Walter P. Murphy professorship and directs the NSF Engineering Research Center Human Augmentation Via Dexterity or HAND.

With over 15 years of experience in technology, design and implementation, and support of business transformation, and in her role as head of global technology and innovation at SGPS, Nicole Mangarella is a recognized subject matter expert in leveraging digital solutions to improve business outcomes. Cassidy Reid is an accomplished professional and automation and digital transformation, who currently serves as the founder of Women in Automation, which focuses on empowering women in automation. With extensive experience, including the role of head of advisory of AMER at Tequila Automation, Cassidy led strategic AI initiatives and formed partnerships to engage clients intelligent automation programs.

And now, I have to give a similar caveat to what Doctor de Souza said. The information that you'll hear in this panel discussion does not represent the views of the Federal Reserve Bank of Chicago or the Federal Reserve System. So let's jump in.

I want to start with this question. We sometimes hear that AI will have more of an impact on office jobs than on manufacturing or factory jobs. But based on current evidence, and based on what we heard from Gustavo, do you see one group being affected more than the other? Or are they both being reshaped in different ways? And Ajay, I'll start with you.

AJAY AGRAWAL: OK, thanks very much, Kristen, can you hear me all right?

KRISTEN BROADY: Yes, I can.

AJAY AGRAWAL: Yeah, so I think Gustavo presented data suggesting that people are-- the effect on employment can be very different. His data, which comes from software companies, or the impact of software produced in Brazil, seem to have a negative effect on workers-- office workers on average and a positive effect on-- of work for workers in a factory setting.

And I think it very much depends on the extent to which, as he mentioned-- all AIs are prediction. They're all computational statistics that do prediction. And so, depending on the nature of the job, if the nature of the job that a person is doing is largely prediction, then the AIs will do it better, faster, cheaper. And it will substitute for human labor. If the AI-- if what people are doing is a complement to prediction, so for example, in his factory worker setting. The main beneficiaries were the lower wage people in factories. And those people were getting the benefit of, effectively, an intelligent AI, giving them advice and answering questions with regards to, for example, how machinery works and how to make repairs. And they were better able to do their job. So it enhanced their capability. The prediction was a complement, rather than a substitute.

And so, I think, right across the spectrum of the labor force, we will find these effects where, when the primary component of what a worker does is prediction, AIs will likely have an effect of reducing the demand for that type of human labor, and increasing the demand for where people are complements. And also, in the physical setting of the factory, just the physical instantiation of a human body moving around the factory and doing things is outside the capability of what the software alone can do.

KRISTEN BROADY: Nicole, what are your thoughts here?

NICOLE MANGARELLA: I think, for me, it has to do with the aptitude for risk and the consequences of that risk. So in a factory environment, where you have physical, automated robots that need to understand space between people, weight, size, maybe the pitch of the floor, the risk for something going wrong and injuring somebody or causing damage, means that you need more security, need more safety mechanisms. And you need a bit more testing in place before you actually go through, and you deploy that.

And so, there's some things in a factory environment that, technically, you probably could program a robot to do. But the risk and the consequences of that going wrong prevent groups from really going after it and deploying it, outside of maybe pilot cases, really, to protect worker safety. In the office environment, you do have tasks. And I'm really glad that Gustavo used the word tasks when he talked about automation, instead of roles being automated because that's a really important distinction.

You have tasks where the tolerance for error is a bit higher. Maybe it's scheduling meetings on a calendar. If something gets scheduled wrong, somebody's going to catch that. And they're going to be able to recover. And so, I think there's certain tasks within the office that are really good candidates for the use of AI, with the right guardrails, and with the right governance in place.

KRISTEN BROADY: So Ed, I want to come to you here because you've been building robots. So what do you think about that?

ED COLGATE: Yeah, I think Nicole makes some really good points. It's actually interesting. If you look at the history of robots, robots used to all be encaged, now, all sorts of safeties because they were very dangerous to be around. And these days, more and more robots are really safe to be around people.

But now with AI, there's this issue of hallucination. And are we certain how the robot's going to behave? So there will certainly be some issues there. But I think the real issue with robots-- robots driven by generative AI, things like humanoids, is that there's still many, many years, maybe even decades of work before those things are going to have the competence to really add a great deal of value in the factory.

That said, I absolutely think that AI is going to have a massive impact on factory jobs. It'll do so in conjunction with other major changes, for instance, reshoring of manufacturing to the US. So it's going to be a very complicated picture. I think one thing is for sure, and that is we're not going to see a resurgence of manufacturing in the US without the benefit of AI. And it's a really critical tool in the years ahead.

AI embodied in robots, I think that's going to take a longer time to come, as compared to things like AI embodied in prediction tools that Ajay is talking about, or that Gustavo talked about, predictive maintenance and things like that.

KRISTEN BROADY: So, Cassidy, I want to hear from you on this. And I also want to pose this next question to you. So what kinds of tools are having the most noticeable impact on administrative and office jobs? And how does that compare to how AI is being used in factories?

CASSIDY REID: Yeah, I mean, great question. I think, in office environments, the fastest moving wave is generative AI, Copilot's workflow assistants and conversational data agents. And these tools are really amplifying productivity by collapsing research, documentation, and reporting cycles. In contrast, factories, logistics, energy production are all seeing AI embedded in the invisible infrastructure of what they call, the predictive maintenance, digital twins, quality inspections, and adaptive control systems. So office AI enhances that cognition-- cognition. And it changes how we think and communicate, where industrial AI enhances precision. It changes how we monitor and optimize physical systems.

So the difference is really, in my opinion, visibility. In the offices, AI sits in front of you. And on the production floor. It quietly runs behind you.

KRISTEN BROADY: Nicole, Ajay, how do you feel about that? What are your thoughts?

NICOLE MANGARELLA: I agree. AI is just a tool as part of a wider technology ecosystem. So I think what Cassidy points out about workflow tools, specifically about data tools that either help to bring together data, or to help make data usable within those AI systems, are really important. Where we see companies being really effective with their use of AI is when they think about it in conjunction with their overall tech ecosystem as well, not just, oh, here's an AI tool in a silo, test it out for just this use case. It really is part of a wider tech stack, if you want to get real gains out of it.

AJAY AGRAWAL: I'll just add one thing here that was interesting from Gustavo's-- the evidence he reported, which is you can imagine-- so the thing we about computers on average, is that computers increased the wage distribution. In other words, computers helped everybody. But it helped highly educated people more than it helped less educated people. And that increased wage dispersion.

So it meant highly skilled people's wages increased faster than people with lower skills. And what Gustavo presented, at least in his data, was the opposite, that I was doing the opposite. It was compressing wages-- that AI was disproportionately benefiting lower wage workers. So in other words, by having an easy to use, for example, question answer system, or that it made machinery easier to use or in the office, it made tools easier to use. It helped people write their emails, that it was compressing wages.

And so, what we should keep our eye on, as this technology develops, is whether it's going to continue down this trend that Gustavo noted, of compressing wages, or go back to what most digital technologies have done and increased dispersion of wage. And that's, I think, it's still early in the game to make any definitive conclusions on that.

KRISTEN BROADY: So you made this point. And I want to stick with you on this because some studies suggest that AI, by simplifying these tasks, would allow employees with less prior experience to perform them. But in practical terms, what does that mean for hiring, training, and career progression? Because I guess if these entry level tasks can be done by AI, how do people learn the skills that they're going to need later in their career?

AJAY AGRAWAL: I'll take a first stab at that, which is, I think the skills that will be important later in their careers are going to be different than the skills that are important today, when we work without machine intelligence. And so, a very good mental model I find useful is thinking about accountants. That accountants-- I teach at a business school-- 40 years ago, when students came to the business school, they would spend the majority of their time studying accounting-- accounting. It was a dominant subject. And within that, arithmetic, addition, subtraction, multiplication Because that is what accountants spend 80% of their time doing, arithmetic.

Then, along came spreadsheets. And spreadsheets were able to perform better than any accountant, whether they were mediocre or excellent at arithmetic. The machine was better than everyone. So that skill that they used to spend 80% of their time on, became irrelevant. And now, you might say, well, if the junior accountants never learned how to do arithmetic in their head or long form arithmetic, then how are they ever going to learn the skills they need to be a more senior accountant? It's because the skills changed. What we needed from accountants changed.

And so, I anticipate a similar type of thing happening right across the board, that everyone will start using various elements of machine intelligence for the things they're doing. And the parts that people do will morph. And the young people, as they come up, will learn whatever those-- whatever the new emphasis is.

KRISTEN BROADY: I guess that's interesting. One of the jobs that I focus on in my own research is accountants because you do have to have a college degree to do that job. Frey and Osborne, they gave that job a automation risk score of 94%. But new research by Tomlinson, from Microsoft, they're looking at AI.

And so, I guess I want to think about AI versus automation, and the impact there on the skills that people are learning, and that they'll need to learn later. So I want Ed and Nicole and Cassidy to jump in on this one. What are your thoughts about how people learn skills and how AI is impacting that?

ED COLGATE: Well, yeah, I'll jump in here. I think, first of all, that Ajay makes a really important point, and the skills that we need are going to change as a result of the AI. A lot of people have a concern for the opposite of what Gustavo's data showed, which is that AI gets good enough to actually do those entry level jobs. And then, the question is, how do people get trained? How do you suddenly become a mid-career type person?

And I don't what will happen there. It's hard to predict. But I do think we've seen plenty of examples through history of fields where more experience and more training is necessary even to reach the entry level. Probably many of us work in those fields. I mean, I think there was a time when if you want to be a successful in business start in the mail room. But now you start with an MBA.

And so, our response to society is, in effect, been to provide that training. And that means people start their careers later. But very often, they have more upward mobility throughout those careers. And so, will that be the consequence? Maybe AI will be used to provide that training, and to help people start their start their productive careers at a higher level than would have been the case in the past.

So I think it's going to be very, very subtle. It's a really hard thing to predict because of what Ajay said, the jobs are just going to change.

KRISTEN BROADY: Nicole, Cassidy?

CASSIDY REID: Yeah, I'd say, I completely agree with what Ed and Ajay were saying. I think there's a little bit more concern there, to be honest with you. I think that it'll look different. Maybe there will be a bit of a pinch until, in the next five years or so, until we figure out how to adjust accordingly. Because AI is happening so rapidly, we're not able to bring up the training in the amount of time.

So it's really compressing the experience curve. And what used to take five years of repetition can be mastered in months. So we really need to make sure that the training is matching that because as of right now, it is making it very difficult for entry level positions.

NICOLE MANGARELLA: I would agree with that. And I would also point out that it does really depend on the industry and the use case. So we did a study with work tech. And we found that about 74% of senior management was using AI. They felt it made them more productive. They were releasing benefits of it. But only about 40% of junior staff were doing the same.

And when we probed a little bit deeper, we found that the concern from junior members of the team was really around policy of what they were allowed to use, and how they were allowed to use it. And this was specific to generative AI. So it's a little bit more, let's say, user directed than AI that's built into enterprise software platforms. That is very purpose specific.

But we've also seen studies where, when AI is really deployed effectively to younger workers, it does help them onboard and learn faster. I think there was a study in the Journal of Economics earlier this year that showed that in customer service roles, junior employees in that role were able to become as effective and as productive as more senior members in shorter amounts of time, when they were giving tools that helped them to learn faster, to be able to ask more questions. And they were able to get that coaching without taking the time of more senior managers.

So overall, productivity benefits did actually show up in those cases. So it's sort of a bit of both. How do you give younger team members the confidence and the capabilities and the right sort of walled garden to experiment with? But then, also how do you make sure that you get those gains for more senior members of the team too?

KRISTEN BROADY: So I guess, in thinking about this, we've talked about entry level jobs, versus more mid-level experience. And I guess, I wonder, is AI impacting tasks or roles or jobs in the same way that automation reshaped manufacturing? Where, is it happening faster or slower? Or like, what are we seeing here? Ajay why I start with you on this one?

AJAY AGRAWAL: Sure, so just to begin your question by pulling on a couple threads that we already started, when Nicole mentioned the junior worker, and she said, for example, in the call support-- customer support, that it disproportionately benefited the juniors. Let's just-- like, that sounds very great. Like, it sounds positive. Who wouldn't want something that accelerates the capabilities of the juniors?

It's important to remember the downside of that. That means that it reduces the wage premium for skill. So that-- in other words, it's great for the junior person. For the person who has more skill because they've been working at it for the past five years, the value of that skill has just been eroded. And so, it means more people can compete, with less experience, for a higher skilled job.

So that's going to have an effect on wages. And also, Cassidy's point where she said, hey, not so fast, we can't train people as quickly as this is changing-- in economics, we use a term that is arguably a real euphemism. We call it adjustment costs. Like adjustment costs are when there's a technology change and people need to adjust. And so, it sounds very neutral. But of course, that's people's lives. And a lot of people are going to be struggling to quickly tool up and learn these things.

And so, if I take those things, to your question, any jobs that are exposed to this effect, where Junior people get upskilled very fast, as opposed to higher skilled people are given extra superpowers that make them even more productive at their high skilled jobs. Those two separate effects. That first one, I think is going to create a lot of angst for people who are further along in their careers, older people, basically.

Older people are going-- if they are in categories where junior peoples are super accelerated in their capabilities because the AIs augment them, we could see quite a significant impact the demographic that is, let's call it, 50 and up or up, or maybe even 40 and up.

KRISTEN BROADY: Nicole, thoughts?

NICOLE MANGARELLA: Yeah, I think it's a great point. And it's a true observation. As soon as you show that you can onboard somebody to be really productive in two weeks, people look at that and say, oh, well, that's great. As soon as I need people, I can easily find them. I think it will be interesting to see which companies are able to adopt it successfully because the change management aspect is a huge challenge with any technology. And it is usually the success factor that will determine if it's something that actually gets the benefits for those younger workers as well.

So I think that communication, how do you help workers to feel like this is going to augment their capabilities? How do you give them that sense of purpose that what they're doing is value add for them and for the company? And those are huge challenges for businesses today.

CASSIDY REID: I want to piggyback off of what Nicole just said because I could not agree more. I was just the chairwoman of the Chief AI Officer Exchange here in Austin, Texas a couple of weeks ago. And you had the chief of Wayfair, Airbnb, Takeda, just all these massive organizations. And they all talked about how technology is not the issue. People are the issue for making sure integrating these technologies.

So someone was very clever. And they said that you always need to consider the three P's, it's people, process, and politics when it comes to AI. And I thought truer words have never been said. So I just-- I loved exactly what Nicole was saying, just wanted to echo off of that.

KRISTEN BROADY: I like that. So I want to talk about the benefits of AI. And I'm going to start with you on this one. Who is benefiting from AI the most so far? And what policies might help ensure that productivity gains are shared more broadly?

ED COLGATE: I think we could probably all agree NVIDIA is benefiting the most so far and maybe OpenAI. So I mean, it definitely helps to, whatever it is, own the means of production. But I don't think that's really where you're going.

I don't have enough expertise to say. My general sense is, based on Gustavo's work, that workers, especially inexperienced workers, are benefiting. And I don't think I've seen a lot of evidence that the employers are benefiting as much. I think that these gains are really having a big impact on the bottom line. Maybe others know differently. But certainly, in areas like in factory automation, there have been benefits, things like predictive maintenance. But the broader benefits that come from, again, generative AI, robotics, those are still lagging.

So my general sense is that at this moment, it may be the worker who has this new tool that helps them upskill more rapidly, or just augments their work more effectively, who is seeing the greatest benefit. And that's a lovely thing. Whether that will all maintain over time is, I think, more difficult to say.

Specifically with regard to robotics. I'll just point out that I think that the changes that we're looking at are not happening that rapidly. When it really does, when you're talking about embodying AI and a machine, it's a much more complicated, much more slow moving type of change. And so, we probably come back to this point. But I do think that means that we have time as a society to think about how to react and how to ensure that the benefits, as the technology does come, are more broadly shared.

KRISTEN BROADY: I feel like, depending on who is hearing that, that could be a good thing or a bad thing. Like, some people want AI to be better, to be more efficient, to do more right now. And others are fearful of that. So, Cassidy, what do you think? Who's benefiting from AI the most right now?

CASSIDY REID: Yeah, that's the million dollar question. I think there was just a study that came out from MIT in September called the Gen AI-- Gen AI Divide, where they revealed that 95% of generative AI business efforts are failing, with only 5% actually achieving meaningful revenue growth, which is shocking because we keep hearing AI is the silver bullet. But there's so much that goes on behind it. Between process re-engineering change management, culture shift, culture towards more adoption and enablement. Right now, I'd say it's technically beneficial to the workers.

I think it's very dependent on which industry and what type of role and tasks you're looking at. But right now, we're not seeing as much-- as many benefits from a corporate side as we thought we would be.

KRISTEN BROADY: Nicole, who's benefiting in your mind?

NICOLE MANGARELLA: Yeah, just to piggyback off of Cassidy's point, that MIT study, I thought the point that was interesting was they pointed out that, I think, in the high 60% of cases where companies were using partners were way more successful than instances where companies were trying to build their own. And I think, to Ed's point, and he meant it as a joke, but I do think that the large tech companies are really benefiting because you have all of this activity in the media that says, you have to start to use this. And you have to test it out. And they are experimenting with it, and who's in a better position to help them understand how this tool can work in the rest of their ecosystem than a tech company that does this day in and day out?

And so, I do think, even when I work with our partners, every call, I learn something different that we haven't seen in the past. And so, I do think there's a lot of benefits from the tech side of it. That being said, I would love to see if there's any data on benefits for small and midsize businesses. I think the thing that has always struck me about the transformative power of technology is its ability to really augment human capabilities in a way that makes a team of 5, 10 people seem like a team of 30, 40, 50 people.

And so, I'd really be interested to see if there's any data, or if there will be forthcoming data on the impact for smaller to mid-sized businesses, versus the corporations that seem to be throwing increasingly larger budgets at the tools.

KRISTEN BROADY: So I feel like I have to come to you on this because I'm thinking about HAND, which I've been talking about for, I don't know, it seems like two years now. And I guess, what you're doing is making AI where automation or robotics, you can say it better than me-- but trying to make it more accessible. So can you follow up on that?

ED COLGATE: Yeah, I was really enjoying listening to Nicole just now because it speaks to my heart. What we're doing in this research center is, in effect, trying to democratize access to robotics, and especially thinking about those small and mid-sized firms because the data are very clear on this. If you look at, survey American manufacturers, where do you find robots? You find them in the really large firms. It's the Amazon distribution centers, the Automotive OEM Manufacturing. But you don't find them further down the supply chain.

Those small firms that could benefit, that need to drive productivity in order to compete on a global landscape, they don't, they don't use robots. And why? It's pretty obvious it's not the cost of the robot. It's everything else. It's the integration. It's programming the robot. It's integrating with the vision system. It's changing your entire production system, so that the robot actually works.

And also, they often live in these lower volume, higher mixed type settings. So it's not like you can just do it once and be done and keep building the same widget again and again for a couple of years. You have to keep changing. And so, the economics just don't make sense.

So what we're hoping to do in our center is change that picture by, first of all, making robots that have more physical capability, very versatile end effectors hands. But critically, it's that AI integration. It's like, the idea that the robot really should have vision are already integrated when it comes out of the box.

It should have skills. It should how to use tools when it comes out of the box. And it should have an interface that lets you anybody who's a domain expert, but not a robotics expert actually use it.

Just like there was a time in my life-- I'm a little older than you guys-- when it took a lot of expertise to use a computer. And what happened? It wasn't just that computers got cheaper. It was that the interfaces evolved the GUI, the mouse, things like this and the out of the box things, like the spreadsheet that we've mentioned. Then it changed computing, democratized it. And we absolutely believe that this sort of thing can happen in robotics. And in doing so, it creates opportunity, especially for the smaller firms.

KRISTEN BROADY: I want to stick with training because I don't how old anybody is, but I remember when I got my first computer when I was in college. And it was a big, giant thing. It took forever for it to warm up. And most of us at the time didn't how to get it to do a lot of things. We didn't its capabilities.

So Ajay, I want to ask you, what training or credentialing approaches are most promising to help workers work with this technology? Like, what Ed is doing is trying to make it easier, trying to make these robots that when they come out of the box, they're useful. But until it is able to do that, while he's doing that, what training or credentialing is most helpful for workers?

AJAY AGRAWAL: Sure, that is industry specific. So I have seen just many different applications of AIs in different industries and people trying to learn how to use those tools. Now, we're at a period in history where this is so new, that in many cases, it is left to workers to learn the tools. And some lean in, and love it. And others are slower.

So for example, I've seen it with accountants. I was just looking at one work where there's a new tool for doing very sophisticated work with AI in terms of complex tax related issues. And some of the accountants working in this area are learning on their own and experimenting and becoming experts in their firm. And their value inside their companies is going up significantly. While others are sticking to doing what they have always done. And they are falling a little bit behind.

Same thing in engineering. I'm just looking at one of the tools called Colab. And it's for engineers collaborating with each other and building things like cars and airplanes. And the AIs do a step in that process called code review. And some of the engineers are significantly more advanced in terms of how much they're using. Nobody is telling them how to use it. It's too early.

I suspect, at some point, they'll start learning to use these tools in engineering school. But right now, the tools are changing so fast that if someone were to make a course focused on how to use this particular AI engineering tool this year, it would be obsolete by next year. So right now, it's the Wild West. People are in there. And those that have a propensity to try and learn to use these things are getting ahead of their peers. And it just is what it is.

KRISTEN BROADY: So to follow up on that, Cassidy and Nicole, I want to ask you this question. From a policy perspective, where should we be placing emphasis, upskilling, reskilling and income support or guidance? Like, what should we be doing considering what Ajay just said and how fast things are changing?

NICOLE MANGARELLA: I think that's a really tough question to answer. And I bet there's an AI engine that could give you a model of how to predict where to best invest time and resources on that. But I do think there are some underlying skills, especially if you look broader at where the AI is going to sit as part of an overall process. I think things that never really go out of style is process design, critical thinking, really good communication skills.

We've seen a lot more interest in hospitality training recently for teams because as you can automate some of the behind the scenes work, that face to face interaction with people becomes more magnified and becomes more important as well. So I do think, to Ajay's point, it's really important to understand how to use the tools and how to interact with them. But I think there's also soft skills that are wrapped around making sure that deployment of AI is successful, that can be complimentary and just as important as the technical capabilities.

KRISTEN BROADY: Yeah, I guess it's interesting that you point out hospitality. Chris Torrance, who's in management at Savannah State University, is looking at that. And particularly with hospitality, because it is so customer facing, that the I can do a lot of the work behind the scenes.

I think about when I was a cocktail waitress in undergrad, that there could be a machine that tells you how to make the cocktail, or how to make the food, the recipe or whatever. But the interaction, the greeting, the welcome, all of that is human. And I think about robots that you can follow, that can translate. But the human voice, like, that interaction, maybe AI will be able to do it at some point. But I think we're pretty far off from that.

We are almost out of time. But I do want to ask this question. If we convened this panel five years from now, which workers do you think would say that AI improved their work experience, which ones would say the opposite, and what factors do you think will be determinative? And I'm going to let you guys just jump in on this one. Think five years from now.

CASSIDY REID: All right, I think that if we're going to look five years from now, the workers who will say AI improve their jobs will be the ones who organizations treated it as a collaborator, not a cost lover. So knowledge workers

who gained autonomy through AI Copilots, technicians who gain foresight through data driven tools, these are the people that will thrive. Those that will struggle with these environments that deployed AI to people rather than with people. So it's really that kind of mindset shift. Defining factor won't be the model's capability. It'll be the organization's imagination, and whether it used AI to elevate human work or to remove it, in my opinion.

KRISTEN BROADY: Ajay, what about you? Who's going to benefit the most? Who's going to say the opposite, if we came back five years from now?

AJAY AGRAWAL: So I divide work into decision making, which AIs are ultimately what they're doing is taking actions and making decisions, as having two components prediction and judgment. And AI do the prediction part. But they don't have judgment. Only people have judgment.

So it'll be the it'll be those people who have invested and continue to develop their judgment because that will become increasingly valuable as the cost of prediction keeps falling with the advance of AI. So everywhere that people are employing their judgment, they will be benefiting from the advanced machine intelligence.

KRISTEN BROADY: With one minute left, a Nicole and Ed, who's going to benefit? Who's going to think the opposite?

ED COLGATE: Go ahead, Nicole.

NICOLE MANGARELLA: I was going to say, I agree with both what Cassidy and Ajay said. I think that people who are working in supportive work environments, where they feel like the AI is there to help them do their best work, are the ones that are going to be the happiest. The ones that-- I love, what Cassidy said about having it applied to them versus for them, I think those are going to be the ones that are the least happy five years from now.

KRISTEN BROADY: Ed, we'll give you the last word.

ED COLGATE: Well, I'll just say, when it comes to robotics, I don't think you're going to see a lot of change in five years. People may be getting really exhausted from having heard about how humanoids are going to take their jobs. But I don't think it will have done much in that regard. So I think speed is always the thing that's most difficult to deal with. And that's going to have a bigger impact and office jobs and things of that nature, information handling. But the answers earlier were just beautiful. So I will leave it right there.

KRISTEN BROADY: Well, I think that Gustavo and I and other people that are doing research in this area definitely benefited from this discussion. I hope that the audience did too. Thank you Cassidy, Nicole, Ajay, and Ed. I really appreciate all of you for joining us, for sharing your opinions. Please check out all of these people's websites because they are all doing amazing work.

I want to thank Gustavo de Souza for his work. And thank you most to our audience. Thank you all for joining us. I would ask you to check out our website, the Economic Mobility Project at the Federal Reserve Bank of Chicago. There you can find Gustavo's working paper. You can find his policy brief.

We will be sending you a survey just to see how you felt about this event, and to see how we can improve future events. So please check out our website. Thank you all for joining us. And have a wonderful rest of your day.

ED COLGATE: Thank you, Kristen.

CASSIDY REID: Thank you.

KRISTEN BROADY: Thank you.