Artificial intelligence technology is transforming the workplace, not only by automating tasks but also by reshaping how teams function and what skills are needed to succeed. As businesses adopt AI tools, focus is shifting from “Will AI replace workers?” to “How can humans and AI work together effectively?”

David Deming, Harvard economist and Kenan Institute Distinguished Fellow, studies how labor markets and education systems respond to technological change. During his recent visit to campus, Deming discussed his research exploring workplace collaboration in mixed human-AI teams and how soft skills like teamwork and decision-making are increasingly central to career success.

Below is an excerpt from our conversation. Some responses have been edited for clarity and length.

As part of this year’s annual Grand Challenge, the Kenan Institute is focusing on the nation’s skills gap. Drawing from your expertise in labor economics, how would you characterize the present skills gap in the U.S. labor market?

A skills gap is when the education system does not produce the combination of skills that matches what employers are looking for in a new hire. Employers often focus on very specific skills, like the ability to operate a certain machine or the ability to write a particular kind of code – these are commonly referred to as “hard” skills. Employers focus on hard skills because they are very direct and verifiable – you either know how to do it or you don’t. Yet I don’t think that this sort of skills gap is the most important priority that should be addressed through policy.

When a skill is specific to a particular company, the firm should bear the burden of training. It is the education system’s job to produce general skills that are valuable everywhere, for all kinds of people. General skills, sometimes called “soft” skills, are especially valuable because they survive changes – technical disruption, shifts in production processes, offshoring, firms moving and closing – all the things that can happen in a person’s career. As educators, the best way we can prepare people for these disruptions is to give them a broad tool kit that’s transferable and useful everywhere.

Technological disruption, particularly from AI, seems top of mind for just about everyone in business, education and other sectors as well. A commonly held fear is that AI will replace many jobs, especially white-collar ones. Yet your work points to a more optimistic view focused on the potential for collaboration between humans and AI. Delving into this subject, how can we apply what you call “task trade” to the AI-augmented workplace? What does effective teamwork with AI look like in practice?

I admit the phrase “task trade” is kind of kludgy, but I use it to show how principles of trade in goods relate to modern office work. One of the most important insights the economics profession has ever contributed to society is the principle of comparative advantage, which finds that if two countries have different strengths in the production of goods, they should specialize in the thing they’re good at and then trade.

You can apply the same insight to teamwork, but instead of trading goods, we’re trading job tasks. For example, if you and I are going to write an article together, and we’re going to do some data analysis, maybe you’ll do the writing because you’re a writer, and I’m a data nerd, so I’ll do the data analysis. With this division of labor, we could write something better and faster together than if we did it separately.

You can find lots of examples like that where when you work on a team, you have a specialized role and you play different roles, depending on your strengths relative to your teammates. Working well in a team is about fitting in with others so that you can realize the benefits of the gains from trade – in the modern workplace, it’s trade in tasks. Firms that figure out how to put people into teams with complementary strengths tend to benefit the most. This synergy is why we work in teams in the first place. As projects get more complex, people ought to tackle different parts and then trade insights.

Some people think AI is going to be better than people at everything. I don’t think that’s likely to be the case because there’s so many things that humans do. We’ll always be able to develop our relative strengths. I believe that AI is going to be a big deal – it’s going to be a fantastic contributor to productivity gains. But one of the lessons of comparative advantage is that there are gains from trade even when one party is better at everything. That’s a key insight, that comparative advantage is about relative strength, not absolute strength. And I don’t think AI is going to be better than us at everything.

Let’s look at what AI can do right now. I would say that when it comes to knowledge work – anything you could do on a laptop – AI is well above the human average at almost everything. But it does not exist in the physical world yet. It can’t get on a plane and close a deal with a handshake – there’s a lot of things AI can’t do.

AI is like a teammate that is good at everything but truly expert in nothing, so as a human partner to AI you want to fit in as the expert in something. This arrangement means that people will need to know their own deficiencies and use the AI to help in those areas.

Recent studies show declining test scores among students across the performance distribution since 2013, reversing decades of progress. There are many plausible explanations, yet it is impossible to ignore these trends’ coincidence with the rise of smartphones and social media. How can we be confident that AI won’t further widen educational inequalities, even as it promises to democratize access to knowledge?

I think there are two questions here: One is how much of the decline in test scores is due to smartphones. I could be convinced that smartphones are the main culprit, but I think it’s hard to single them out because the pandemic had a role as well when kids were home instead of in school. Some of the decline started happening before the pandemic, although the pandemic surely made it worse.

I don’t think AI will make it worse. I think it’s more likely that AI will change our beliefs about the value of measuring using today’s tests. One question is, How does AI affect your performance on a test? Another question is, Does AI affect whether the test should be the right metric? I think kindergarten through eighth grade, we should double down on what we know works. There’s a science of teaching people how to read. There’s a science of teaching people basic math. We know how to do this – it is a solved problem. We’re just not executing on it. We should make sure that every kid gains the basic skills they need to be a functioning member of society, and that likely involves banning phones in schools while limiting access to computer tools, AI and other digital technologies to young children. Just as you wouldn’t try to teach calculus in third grade, you shouldn’t try to introduce AI tools at that age either. I’m open to the idea that we could integrate AI into education in some creative ways, but I don’t think putting kids in front of screens for long periods of time, especially young kids, is good for them. As kids get older, it becomes a harder question because the educational space is more open-ended.

If you were to design high school and college course curricula today, what concepts and lessons would you emphasize to prepare students for a future of AI collaboration?

Now more than ever, being a good teammate is very important to being an effective, valuable worker. So, I would integrate technology, including AI technology, to make education much more interactive and with a lot more group-based work.

I still believe the magic of learning is in the teacher-student interaction. This is basically because learning is hard, so when you have a teacher whom you believe in and who believes in you and who keeps you on task, that’s extremely valuable. I would use technology to be the architecture around the student-teacher relationship so that we can save the most valuable time for that interaction. I would use AI to make the essential part of the learning experience better. This could take the form of, for instance, a teacher using a tool that synthesizes and summarizes how their students have used AI and what they know and don’t know what questions they have asked. You could imagine building tools that give teachers a rich understanding of what their students know and how they can help them. I think the motivational side of learning – what gets people excited to learn – is something we ignore at our peril. It’s all about human relationships.

---

• Keyword(s):
• artificial intelligence 10
• skills gap 14