AI Could Actually Help Rebuild the Middle Class, Says MIT Economist

CHRISTIANE AMANPOUR, HOST, “AMANPOUR AND COMPANY”: Next to the future of work here in America, where artificial intelligence is taking hold and fears of unemployment are growing. Elon Musk called it the most disruptive force in history. And 75 percent of Americans, adults, think AI will lead to job losses, according to a recent Gallup poll, that is. But, MIT Economics Professor David Autor says this fear is actually misplaced. He joins Walter Isaacson to discuss the opportunities AI will bring.

(BEGIN VIDEOTAPE)

WALTER ISAACSON, CEO AND PRESIDENT, THE ASPEN INSTITUTE: Thank you, Christiane, and Professor David Autor, welcome to the show.

DAVID AUTOR, FORD PROFESSOR OF ECONOMICS, MIT: Thank you very much for having me.

ISAACSON: You’re a labor economist. You’ve looked at how technology affects jobs. And you have a new piece out about artificial intelligence, saying that AI could actually help rebuild the middle class. But, before we get to that, let’s do a little bit of a walk through history. For a long time, whether it’s back in the Luddites in the early 1800s, they felt technology would destroy jobs. The followers of Ned Ludd smashed the looms of England. Has there ever been a case where technology decreases the total number of jobs?

AUTOR: It doesn’t normally decrease them, but it does change them a lot. And the Luddites were not off base. Their artisanal skills were devalued by power looms. And the transition from the artisanal era, where people made things by hand with tremendous expertise from start to finish, to the early industrial era was wrenching for labor. It displaced a lot of skilled, valuable skills, people who were tailors and blacksmiths and wheel rights and so on. And a lot — what we ended up with initially was a lot of unmarried women and children working indentured servitude, in dangerous factories, using very few skills, and getting paid very little. And the first five decades of the Industrial Revolution were not a good time for labor. Wages didn’t rise, even though productivity rose. And why was that? Because expertise, what was needed, was not scarce. You just needed people, physical bodies, who could tend (ph) machines. Now, that changed over time as industry advanced. The machines became more complex until the industrial era eventually gave rise to a period of what I’m going to call demand for mass expertise, people who could do those skilled tasks on assembly lines, but also in the offices. Right? You can think of early 20th century offices being like an assembly line for information. And although that set of skills was quite different, not like the artisans, they weren’t making cars from end to end, one person at a time, but the ability to operate a lathe or to install a wheel, or to proofread and typeset a document, those were valuable skills. They were made very productive and efficient by automation, by the Industrial Revolution, by this new way of organizing work. And that led to a lot of economic growth, both for consumers and for workers that powered us from really the late 18th century all the way up through the 1980s.

ISAACSON: So, in other words, productivity growth, new technology, ends up helping the economy, creating a whole new set of jobs, but it ends up being wrenching and leaving people behind. Is that what you’re saying?

AUTOR: Absolutely. And when it creates valuable jobs is when it rewards human expertise. What is expertise? Expertise is the specific know-how to do something valuable. So, just to give you an example, think of the job of air traffic controller and crossing guard. These are basically the same job, actually. It’s a job to prevent people and machines from colliding with one another or machines and machines colliding. And yet, air traffic controllers get paid more than four times what crossing guards get paid. And the difference is not social value. If we had to spend a lot of money to prevent our children from being run over on the way to school, we would spend that money. Right? It’s a question of expertise. It takes several years of flight traffic control school and then hundreds to thousands of hours of apprenticeship to become a credentialed air traffic controller. To become a crossing guard requires no training or credentialing in almost any state. And therefore, the people who can do it are abundant. And so, it pays low wages. And that was true of early factory work as well. So, it’s never been a question in the United States, certainly, of the quantity of jobs, but it has been about the quality, and quality means using human expertise.

ISAACSON: You wrote a seminal paper about 10 years ago called “Why are there still so many jobs? And it really came out of a period in the 1960s when everybody said automation would totally put us out of work. What was your point in that story, and what did we get wrong about leaving people behind?

AUTOR: The main point of that article is that we had entered a different industrial era with the computer revolution, and that actually competed with the expertise of workers in factories and offices who were carrying out these literate, numerous tasks. They were skilled tasks, but they followed well understood rules and procedures. And so, that actually caused this pushing — workers to be pushed out of middle skilled jobs, out of production jobs, out of office jobs, administrative support, and clerical jobs, and it created a bifurcated labor market. On the one hand, if you’re a professional or a technical or a managerial worker, you’re a decision maker. And computing is a great input into decision making. It gives you data. It gives you analysis. It gives you all the information you need. You still have to do the hard work of deciding, how do I care for this cancer patient? How do I design a building that people want to live in? How do I architect a piece of software? How do I contract and re-engineer a house, right? So, those are hard decisions. Computing is really valuable. It makes that work better. It makes people more valuable in doing that work. For those people, however, who are not fortunate enough to have a college education, which is more than half of the workforce, many of them as they were moved out of middle skilled jobs, they ended up in services that use relatively generic expertise, food service, cleaning, security, entertainment, recreation, home care. And again, those jobs are socially valuable, but because they don’t require specialized skills and expertise, they pay poorly. And so, the computer revolution, it didn’t reduce employment. We have high unemployment to population ratios. What it did was it bifurcated the labor force and kind of cut out the middle rungs of the ladder that weakened the middle class, reduced economic mobility, and created a big divide between more educated and less educated workers. And that’s really what we’d be contending with for the last four decades. From approximately 1980 to approximately 2020, we’ve been really feeling the effects of this polarization of employment.

ISAACSON: So now, let’s go to the era of artificial intelligence, the era of AI. We’ve gone through 50 years since the advent of the personal computer. And as you’ve explained, it’s hollowed out sort of the middle class, the middle worker in favor of those with high-end expertise. Will AI change that?

AUTOR: It has the potential to change that if we use it well. So, let me say, what is AI? What makes it even different from traditional computing? So, traditional computing followed rules. It follows what we call inductive logic. It just does the steps until it gets to an answer. Artificial intelligence is the opposite. It learns inductively and it learns from examples. It learns from looking at unstructured data and drawing out patterns and recognizing regularities that are useful for making decisions, for predicting what’s going to come next. And in fact, it’s an irony. It’s actually the opposite of traditional computing. If I told you the world’s frontier computer technology can’t do math and can’t keep facts straight, you’d say that sounds doesn’t sound like a very advanced technology, but that’s what AI is. Right? It’s really good at learning from example and extrapolating from example. And so, that makes it potentially a very good decision support tool, because it recognizes patterns and regularities like we do when we’re making a judgment about how to care for a patient or how to build a building, or how to do research or even how to teach.

ISAACSON: Well, wait. So, how is that going to help the middle skilled workforce?

AUTOR: Sure. Let me give you an example that I think is motivating. It’s actually, it has nothing to do with AI specifically. Let’s take the job of the nurse practitioner. So, nurse practitioners are registered nurses who have an additional master’s degree in training, and they do things that nurses were not allowed to do some decades ago and doctors were exclusively allowed to do, which is to diagnose, to prescribe medications, and to treat. And they are essentially a kind of a middle class of medical professional. Now, there are several hundred thousand in the United States. It’s a well-paid job, better than registered nurses. Now, it came about, not because of technology, it came about because of social movement, that nurses, primarily women, recognized they’re underused and they fought like hell against the American Medical Association to carve out a new field and a credential and a scope of practice. But, at this point, they’re very heavily supported by technology, electronic medical records, diagnostic software, prescription software, and that enables them to do a broader scope of work. And so, not only has this created a valuable job, it creates a valuable patient service. You don’t have to wait as long to see a doctor. And it’s not as expensive to do so. And so, it broadens the availability of care. And it’s not hard to imagine a future where people with additional medical training or even nurse practitioners could do a broader set of activities without having to bring in the most expensive professional in the room. And that matters, because most of these elite professions that we’re speaking of, they require a bachelor’s degree, plus a master’s degree and a PhD or a JD or MD, or an MBA, and nurse practitioners are also highly credential, but they’re not at that same level. And this example is just an example. It could be true. You can imagine a contractor who has better tools to scope out what are the viable kitchen designs. What are the — what are certifiable engineering designs so that the building will stand etc.? You can — yeah. So, I could — or even in law, right, people who are not – – do not have as much — many years of legal experience could potentially still do more valuable work. So, they — so, the good scenario, right, as I mentioned earlier, six out of 10 U.S. workers do not have a four-year college degree. Most of them are found in these low paid services that aren’t using specialized skills if more of those workers with additional supporting training could do medical care, could do legal services, could do design. And so, we will know if we’re succeeding with this technology if we enable people without four-year degrees to do more valuable decision making work, to open up the field of expertise, such that it’s not to say to eliminate the — I’m not saying we’re going to get rid of doctors and lawyers and computer programmers, but now enable more people to do that work at some level.

ISAACSON: So, what you’re saying is that somebody with a high school diplomas, but not a college degree, who probably lost out a bit in the information and computer revolution, they can be empowered to do things that now take experts to do,

AUTOR: Right. Again, with the right training, right? You wouldn’t just say, hey, I’ve got this tool for you. Go, take care of this patient, and do it, and let me say, perform some procedure, insert a catheter or something. That will be a terrible idea. Right? Something is going to go wrong and the patient is going to — there’ll be an emergency and the person won’t know what to do. But, it’s — but, it could be. It’s quite plausible that I would say, hey, you have a two-year medical certificate in X-ray technology, or you’re a physical therapist, and so on, and here is — you can do a broader range of procedures now that you have the judgment and you have the foundational knowledge, you have a better tool that allows you to go further with that knowledge.

ISAACSON: Previous technology revolutions throughout history generally hurt those with less skilled — less skills, less education. This one perhaps will disrupt jobs of the most educated. Which jobs are the most threatened?

AUTOR: I think the jobs that have the most opportunity for being substantially automated are ones that are kind of mid-level decision-making tasks and managerial work, for example, but simultaneously, we could also see greater assay (ph) for more people to do that type of work. So, there is a number of experiments where we see what computer — what AI does is it kind of levels the productivity differences between more and less experienced workers. We see that in writing tasks. We see that in customer support tasks, like technical customer support tasks. We see that in even a bunch of consulting and analytic tasks that often — the tool complements judgments, enables people to do better work. And if that’s true, then potentially it can lower the barriers to entry. Now, there is a counterargument to this or there is probably many, but one of them, he will say, well, OK, let’s say it makes your nurse practitioner five percent better or 20 percent better, but it makes the best doctor 100 percent or 1,000 percent better. Doesn’t that make the nurse practitioner no longer competitive? Right? And I would say the answer to that is, no. And the reason is, because doctors have capacity constraints. Right? If the best doctor in the world gets 100 times better, I’m still never going to see that doctor. Right? That’s not relevant to me. So, many, many services are — they cannot be dominated by one expert. Right? There is too much — you need too much one on one, whether that’s a legal case, whether that’s education, whether that’s medicine, whether that’s design, right, whether that’s our research. And so, you’re going to have to — a lot of people will have to be involved. Healthcare is the best example. Right? There is infinite demand for labor there, and that’s not going to go away.

ISAACSON: But, you’re talking about democratizing expertise.

AUTOR: Yeah.

ISAACSON: Couldn’t it happen, though, that that just makes some of the experts redundant at a certain point, instead of a great teacher? I’ll have Khan Academy’s Khanmigo as a personal tutor. And likewise, even for most legal work or medical work, those can be replaced by great AI in five or 10 years.

AUTOR: In some cases, it’s going to create more competition. Right? It’s definitely going to create more competition at the top. And in a way, that’s good. Right? The problem we face is a lot of our inequality is driven by very, very high wages for highly educated workers. I’m not saying they’re not working hard. They haven’t earned that money. But, that scarcity actually is a problem for the rest of us. Right? So, if you and I were professionals, it’s great. If we pay for healthcare and it’s expensive, we say, well, great, I’m being rewarded well as a professional. Good for me. But, if I’m an auto worker or I work at Walmart, I still pay the same price for healthcare and for education. I’m not on the upside the equation, only the downside. Right? So, if we could actually — if it was possible to make some of those services less expensive, more accessible, it’s true. We made — the premium salary paid to the mostly professionals may go down a bit. But, if that creates a lot more jobs for others that enables people without as much formal training to still do really good work, I don’t mean no training, I just mean the right level. And it lowers the prices of those services for others, makes education more affordable, more accessible, actually more interesting, if it makes healthcare more available to more folks, if it means that — software is actually less expensive to create. So, you can create customized applications for your business or for your home. Right? There is a lot of benefit to that. So, I’m not arguing that everybody always wins. In all these cases of technological change, they’ve always created winners and losers. Right? The artisans lost out. It took 50 years for the workers, the industrial-era workers to start to benefit. Computerization benefited professionals a great deal. It really was not good for middle skilled workers. It was not good for office workers. It was not good for production workers. It just made a lot of their skills redundant and increased aggregate wealth. But, the distributional consequences were pretty crappy. Most of the time, technology is good for the elite, and not so good for everybody else. This is a case where the technology may compete a little more with the elite and enable more people to do valuable work. So, I’m willing to take that trade if it’s offered.

ISAACSON: Professor David Autor, thank you so much for joining us.

AUTOR: Thank you so much for inviting me.