Wharton@Work

June 2024 | 

Rethinking the Labor Market in the Wake of AI

Rethinking the Labor Market in the Wake of AI

As artificial intelligence continues to reshape industries, business leaders are focused not only on its potential to affect innovation and competitive strategy but also on its impact on the labor market. To explore this transformative shift, we turn to Wharton expert Lynn Wu, whose cutting-edge research provides valuable insights into how AI is not just altering jobs but redefining the very nature of work itself.

Wharton@Work: Let’s start with the big question: are leaders’ fears that AI is going to dramatically decrease or otherwise disrupt the labor market justified?

Lynn Wu: The answer is yes — and no. I have concrete research showing that in terms of the companies that have adopted AI and robotics, demand for workers has increased, while managerial work has decreased. That means you are hiring very different people than you hired before, and the type of manager you need is very different than before. In terms of generative AI, it’s too new to know, but I suspect the impact on labor will be similar.

W@W: Can you be more specific about the types of workers who are being impacted?

LW: If the work you do complements the technology that's happening today, like generative AI, demand for your labor is going to increase and you will probably get paid more. Examples include those who work directly with AI, like using LLMs [large language models, which understand and generate text] to make your work more productive, or designing an improvement for a business process using AI. Those jobs are going to be safe. But if the majority of your tasks involve the exact thing that AI is good at, then you may be replaced.

If, on the other hand, 60 to 70 percent of the tasks you do every day can being replaced with AI, your job class may change. Your employer may still need you, but what you're doing fundamentally will be very different than what you did before. In our research, we specifically look at three labor groups: people without high school diplomas, people with high school diplomas or associate degrees, and people with college and graduate degrees. We found that employment demand increased on both ends, but decreased in the middle.

W@W: What are the implications of your research? What happens when those workers in the middle category decrease or are eliminated?

LW: It actually creates a big problem, because low-skilled workers who haven't completed high school or gotten a GED may have more job opportunities, but they also have no way to progress. With the middle removed, the career ladder is fundamentally broken. That means workers no longer have a career trajectory, so the work of motivating them just got much harder. We could see huge numbers of workers disenchanted, with no clear job aspirations, which would create a lot of uncertainty about what labor forces will look like. That is actually one of the biggest problems with labor. It's not necessarily displacement, but we have to rebuild a career ladder — and that has implications for managers too.

This has already happened with robots, and it's going to be exacerbated more with generative AI. My prediction is that it's going to get much worse before we figure out a solution. Leaders should start thinking now about the type of managers you are going to need to oversee this new class of employees. How do you manage people with no discernable career path? We have to start thinking about the management philosophy regarding motivation for workers in almost a brand-new way.

W@W: How do you envision organizations restructuring work and the career ladder so a new middle can emerge?

LW:  Ultimately, you need to train people to use AI as a complementary tool.  That could be creating a new set of job classes that leverage the skills people already have and then marrying AI to it, but the key will be coaching people to use these tools to help them do their job as opposed to replacing their job.

For managers, those who make sure you arrive on time and leave on time and can tell you exactly what to do will no longer be needed because AI can do those tasks better. But the managers who coach others rather than monitor them will probably be on the rise. For firms that are now using robotics, we are seeing that the supervisory type of work has decreased quite a bit. It is not that we don’t need managers anymore, but we need a different type of manager. Likely, managers like performance coaches who help you become better at your job will be on the rise. Similarly, managers who know the insights about the organization or industry and also understand what AI technologies can do will also be on the rise. They will be at the best position to think strategically about how to improve business processes and reorganize work using AI.

W@W: You’re talking about disruptive change within organizations to make these kinds of changes to job classes and management skills. What kinds of capabilities will that require?

LW: You have to build a safety net for people to be able to bravely test AI technologies. If they think, “You’re asking me to use AI, so that means I’m going be fired next,” they aren’t going to work with it successfully. People need to know they have job security to boldly experiment. That means culture, in addition to industry, is going to be a factor in determining which firms are successful. Within an industry, one firm with the right culture that allows you to boldly explore is going to be rewarded much more than a firm that does not have that culture. So, in a sense, it's a complicated problem, but in another sense it’s simple because we already know a lot about culture and what works in terms of security and the permission to experiment and learn from failure.

W@W: What about older workers who have deep knowledge of their business and industry. How might employers think about leveraging them?

LW: I think the firms that figure out how to reuse their skills are going to be a lot better off than those that do not. For AI to work really well, you have to understand the industry context. Blindly applying AI to an industry will not solve its problems. AI requires humans in the loop, so using workers with deep industry experience to a new purpose that complements AI would be the right thing to do, as opposed to getting rid of them and then later figuring out that while the tasks they used to do are obsolete, the knowledge they have is very valuable and should have been deployed elsewhere. Managers and executives need to identify these workers and begin retraining them, while also understanding that certain positions will become obsolete.

In terms of top performers, we have to think about how LLMs can help them to do even higher function tasks. When it comes to this kind of type of work, you have to think more creatively. What do these top performers need to do even better than before? That’s not clear, but we need to start imagining the type of work they could do to add more value than before.

I like to use software engineering as an example because I know that industry really well. They don’t need to use LLMs to be better coders, because they are already amazing. But what they can do is use them to be a better designer of software. LLMs can free up the time they spent on coding so they can do something even more valuable, like algorithm and system design.

W@W: We’ve talked about what AI can do and the types of jobs it will transform or replace. But what can’t it do, at least right now?

LW: AI, even generative AI, is still solving a very narrow class of problems called supervised learning or self-supervised learning. What that means is you must have both an input and an output to train the AI algorithm. AI can recognize cats better than the humans can, because we have inputted a bunch of photos of cats and then labeled them as cat, which is the output. You must give lots of examples, real input and output, in order for AI to be good at it.

But the most difficult problems in the world do not have an output, so they cannot be solved yet. I don't want to diminish the huge impact on our lives that AI is having. But I do want to remind people that of all the problems we have in the world, the ones we can solve with AI are just a sliver.

Solving the next class of problems is where the most value is going to be. In the meantime, we need to think about re-organization of work. And that doesn’t necessarily require more education, as we observe middle-skilled work is diminishing.  But it does mean new training and skill acquisition so workers can use AI tools effectively.