Episode Transcript

CARDIFF GARCIA: Hey everyone, Cardiff here. So much commentary and analysis and chatter out there about what Artificial Intelligence will eventually do. Maybe it’ll do your job, or at least some jobs. I’m pretty sure eventually it’ll be able to do several parts of my job. But today’s conversation is about what AI will not do — what it cannot do, by definition. 

Adam Ozimek is an economist. He’s also my colleague at the Economic Innovation Group, EIG, and he’s got a new post out today about the parts of the economy that rely on the human touch — including jobs and tasks throughout the economy that, in theory at least, could have been automated away a long time ago. And yet we prefer humans to do those jobs, to do those tasks.

What are the implications of this preference for the human touch? What are the implications for the economy, for society? That is the topic of today’s chat, so let’s get right into it. Adam’s actually here with me in the studio. Adam. Hello.

ADAM OZIMEK: Cardiff. Excited to be here.

CARDIFF: That’s your podcast voice. People should know that you and I talk about this stuff all the time behind the scenes, and now we’re essentially bringing our conversations live to the public.

ADAM: That’s right. Heard a lot about this podcast. Very excited to be here.

CARDIFF: (LAUGHS) You’ve been on the show before.

Where do you want to start? Do you want to start with the story of the player piano? That seems to be the story that inspired you to write this. So why don’t you tell the audience what the player piano is and why it’s meaningful?

ADAM: A player piano is essentially a machine that — initially, you would push it up against a piano, and it was sort of like pneumatic tubes and pumps and levers, and it would press piano keys. And then eventually they started building them into pianos themselves. And essentially what you do is you load a roll of paper in, and the paper contains the music coded onto it, and this plays piano music.

And the only thing a human has to do is manually pump them, because they were invented in the 1890s. And so this is before widespread electricity, so it had to be driven by a manual power source, and that was the last role that was left to humans.

What I found fascinating about it is that here you have something that humans can do, which is play piano, and it’s been automated for almost 130 years, and yet today there are still so many people who play the piano. You can go into a nice restaurant, a nice bar, a hotel lobby, and there’ll be someone there providing live piano music.

And you can look at that person and say that what they’re doing right now has been automated for 130 years, and yet a human is doing it. And so I think that’s an interesting way to approach the AI question: what is it that could have been automated — sometimes for a very long time — but hasn’t been, and why?

CARDIFF: Even staying with the example of the player piano, you had this line in the piece that was: a lot of people just prefer a piano player, not a player piano. Why do you think that is?

ADAM: That is what I call the human touch.

 I think people want to see a human do it. They want to hear a human do it. They want to know that it’s not pre-programmed. They like the uncertainty about what’s coming next. I think that sort of describes the human touch for piano players. But there’s obviously human touch in a wide variety of occupations and tasks, and it differs from occupation to task.

Why is it exactly that we prefer a human to be doing that? I think that’s what it is with humans. And it’s certainly very hard to find a player piano these days.

CARDIFF: Yeah. Although you obviously do have, for example, in hotels and restaurants, just music playing in the background. Sometimes it’s piano music, not being played on a piano by a human or by a player piano — just somebody who loads up a Spotify playlist of piano music.

But you’re a lot more likely to see an actual human being playing the piano than you are some kind of automated piano where the keys are just moving, like it’s a haunted house or something. 

ADAM: Yeah, the recorded music is sort of another face of automation that I think is worth thinking about, because like you said, you can go into a wide variety of contexts today and hear prerecorded music playing, and that in a sense represents automation.

It represents what, in some circumstances, might have been a live human playing the music, but instead it’s prerecorded. And so that illustrates the way that the automation of music is widespread in society, but still leaves humans with so much to do when it comes to live music.

There are so many circumstances in which we still prefer to see a live person play the music, and that’s another important example of the human touch.

CARDIFF: When movies went from silent to talkies, it also meant that you didn’t need an orchestra at the movie theater with you. I’d sort of forgotten that that’s how it worked. But there was resistance. It was almost like — I don’t know if you’d call it union or organized activity of the time — but people hated it. They didn’t want us to shift in that direction.

ADAM: It was union-backed. There was a union — I forget the name of the union — but

CARDIFF: The Music Defense League.

ADAM: Well, that’s the organization that the union created. That was the anti-automation front group that was funded in order to oppose what they called canned music.

And this is in the early 1900s. There’s a lot of really great advertisements and comics and cartoons that they put together depicting robots taking human jobs — the job in this case being the musicians who would play along with a movie.

And certainly that’s true. They were correct in the sense that the job of ‘musician who plays along with movies’ is long gone. It’s long automated away. There’s no one playing along when you go to the movie theater these days.

But zoom out. Are there jobs for musicians still? Yes, there’s more jobs than ever. We put together some census data for this and showed that there were over 200,000 people working as musicians or composers today in the U.S. The data goes back to at least 1850. So there are more musicians today than ever, even though in so many contexts it’s true that roles that previously involved live music have been automated away.

CARDIFF: What is the relationship between automation and the human touch, and the way that those things actually supplement or complement each other as opposed to vying for primacy?

ADAM: I think automation allows us to hear and experience music in far more contexts than we used to. You can listen to millions of songs on Spotify anywhere you go, at any time of day. So it provides consumers with a huge amount of access — previously unthinkable — at very low cost.

But still humans demand to see live music. And I’m not even talking about Taylor Swift — that’s one example I bring up in the paper. She’s charging astronomical amounts to see her live and selling out massive arenas all over the world. There’s obviously huge demand for superstars.

But it’s not just superstars. If you go into any dive bar across the country tonight, you could hear live bands playing — sometimes really terrible live bands.

CARDIFF: Some good ones too! (CHUCKLES)

ADAM: Sometimes good ones. Not to slag off all the live bands playing in bars tonight. Some of them are great.

And so I think the demand for the human touch is not just about superstars. It’s not just about the ultra-high end of quality. It’s fairly widespread, at least when it comes to music.

CARDIFF: Yeah, there’s other obvious analogies throughout the arts, especially. I go to the theater a lot — I live in New York City — and I really enjoy that.

Obviously, we’re now moving to a place where you could almost have CGI automate away acting, and of course there are actors in films and things like that. But there’s something uniquely intense about sitting with thousands of other people watching a performance that only you and the other people in that theater can ever experience.

Even though, obviously, there are plenty of kinds of storytelling and acting and great performances that you can see in movies, which I also love.

ADAM: Yeah. Personally, I hate the theater. 

CARDIFF: (LAUGHS)

ADAM: You’d have to pay me to go see it, so I can’t quite get inside that—

CARDIFF: Do you like live performances yourself, though? Because I like concerts as well. I love concerts. Do you like concerts, or do you just not like big arenas — lots of people crowded into a place?

ADAM: No, I like live music. I’m not going to crowd into a big arena anymore at this age. 

CARDIFF: Why?

ADAM: (CHUCKLES) We don’t need to get into where my demand for the human touch ends.

CARDIFF: Why? It makes no sense. Okay

ADAM: A big, loud crowd —

CARDIFF: It’s the best!

ADAM: I’d rather see someone playing a small local venue, doing some good music. But there are certainly bands I would crowd in for.

CARDIFF: There are different kinds of human touch, is your point.

ADAM: Right. Yeah.

CARDIFF: It’s interesting too, because this is a really controversial thing right now within the film industry. People using AI, people using a lot of CGI and other things — will you ever be able to replicate a phenomenal performance using just an AI actor?

I think you can still tell, the technology isn’t quite there yet. I don’t see any reason why the technology would never be there. I think it will eventually get there. And yet, I have a feeling that people will still want to see movies where you know, and it can be shown, that the acting is being driven by actual, in-the-flesh human beings. You know what I mean?

ADAM: Yeah, I think that’s true. Also, CGI has been able to replace sets and scenes and action sequences for a very long time. And you have tons of professional artists getting together trying to make the best digitized version of reality that they can.

And yet in movies, there’s still widespread use of real scenes, real sets, practical effects. I think there’s an uncanny valley there. And if today’s best CGI artists can’t fully overcome it — obviously in some contexts they do — but they can’t fully overcome it, I think that tells us something about our preference for the real.

I also think that people have connections to the real artists themselves, to the real actors themselves. There’s a reason that superstar actors make so much money, because people want to see them, even though there are a million unknown actors knocking down the door to play all of these roles. Why is it that people want to see the same sort of people in them again and again? I’d say that’s an aspect of the human touch. It’s going to be difficult to replace them with digital copies of themselves when you can’t even obviously replace them with some other actor.

CARDIFF: Do you think the economy overall will shift more and more in the direction of becoming performance-based?

This is a hypothesis of mine, and I’ll be interested to see if it turns out to be right. I think of examples from sports and even from chess or other games. If you think about sports — I love basketball — you could design a bunch of machines that could shoot from 60 feet out and make it 90 percent of the time. But I’d rather go see spectacular feats of athleticism from actual real-life players.

You and I are talking on a Monday. Last night I went to see the Knicks versus the Lakers. It was fantastic. I love seeing live sports. Same thing with chess, which I think about all the time. Even a mediocre chess machine — one of the offshoots of the original… I forgot the name.

ADAM: Deep Blue, right?

CARDIFF: Deep Blue, yeah. Those guys. Those machines can beat the best human chess player in the world with no problem. But there are tons of people who want to see the best human chess players in the world, to the point that there are controversies if one chess player accuses another of cheating, of having some weird device in his pocket or something like that.

There are these controversies, and it’s precisely because you don’t want the cheating. You want to see who can perform this amazing feat — this amazing creative or, in the case of chess, strategic feat. It’s amazing to watch. And people love that. They actually take the machines out of it because it’s very performance-based, very human-based. That has persisted even though machines can beat every human chess player.

ADAM: Yeah, I think that’s another example of the human touch. To put it in economics terms, I would conjecture that the human touch is a normal good. What that means is that the more people make, the more of it they consume, the more of it they demand.

I think this is true in general. Higher-income people tend to spend more of their money going to concerts, events — they go out and do things. Over time, as we as a society have gotten richer, we do more of these things. A great example is restaurants. People eat out now more than they ever have, even though that takes a lot of human hands to do it.

CARDIFF: Cooks, waiters, you need the overhead and all that stuff. Somebody needs to create the place.

ADAM: Right. And the idea that we’re always going to go to the lowest-cost option really misses out on the way that demand for goods and services works, especially when the human touch is involved.

CARDIFF: People might not know this, but you’re also a part-owner of a restaurant/bar/arcade/bowling alley. When you hire for roles at those places, I have to assume you’re looking for people who can provide that human touch. That’s actually a key component of what you want, right?

I know the bars I like are the ones that have really good bartenders — not just because they make the best drinks.

ADAM: Yeah, absolutely. When you go to a place, you want the staff to provide a personable experience. You read Google reviews for any restaurant, and if they aren’t doing that, people notice and they don’t like it. They leave bad reviews and they stop coming.

Other places are well known for their level of service quality. It’s part of what you go for. So I think that’s a great example of the human touch. Occasionally, it happens, but you can’t just replace all your staff with a machine that serves beer, and you go up and pour it yourself. You can do that — that machine does exist — but there are still a lot of bartenders and millions of waiters and waitresses.

CARDIFF: You ever been to one of those bars?

ADAM: I haven’t. We had one around here for a few years, but it went out of business.

CARDIFF: Did it have actual bartenders or was it just a wall of beers you could grab? How did it work?

ADAM: I think the idea is you just go up and pour your own beer. But it turns out the human doing that role was providing valuable service.

CARDIFF: Something different. Something harder to put your hands on. That’s another thing to bring up — it’s hard to empirically show what this is.  It exists in this other place.

We can track it in certain ways. We can say there are so many composers and musicians right now, so many bartenders. But this is a really hard thing to identify. Where does a human touch begin, and where do other more impersonal forces of economics — supply and demand and stuff like that — take over?

ADAM: It varies so much depending on the context. The reason a live band still plays at a dive bar is very different from the reason a bartender still works there, which is different from the reason a sales engineer is selling robotic parts to clients.

The human touch is important across the economy, but what exactly it is varies a lot, so it’s hard to put a number on that.

CARDIFF: What are some other jobs and occupations where you think this is a key component — where AI might shift around some of the way the job is done, but the human touch still provides such value that people will still want other human beings to do those jobs? 

ADAM: We’ve already talked about musicians. Artists broadly, I think. There’s a reason that when a painting is proven to be a counterfeit, its value drops astronomically — even though the replica might be so well done that to almost everyone it passes off as the real thing. For whatever reason, humans value that it was actually done by the real artist.

So the arts are one area. In the more nuts-and-bolts parts of the economy, I think sales is  an area where the human touch is valued. I don’t have data on this, but in general, the nicer the thing you’re buying, the more likely it is you’ll be dealing with a better salesperson.

You go into a fancy car dealership, a fancy watch store, a place to buy a nice suit — the luxury end of the economy is some of the most hands-on, human-touch-centric parts of the economy. So if AI delivers productivity growth and leaves us all much wealthier, we’re going to have demand for those sorts of things.

CARDIFF: Here’s a trickier one, I think, which is in the realm of caregiving.

Not just childcare, elder care, but even healthcare more broadly. When you go to see a doctor, you go to see a nurse. I think sometimes people don’t want faster productivity growth in those sectors precisely because they like the human experience. They’d rather be able to have their doctor explain to them something for a little while, right?

At the same time, if these services get very expensive, if there are blockages that make it too hard or even impossible for enough healthcare providers to take those jobs because of administrative barriers or legislative barriers, whatever it is — these things exist all over the economy — and so those services go up a lot in price. Some automation might help people out because you also need to be able to afford healthcare or childcare or elder care. 

I don’t exactly know what I think about that just yet. I think there are some places where automation would really be helpful, right? In terms of like diagnosis, it seems clear that that’s a direction we’re headed in.

Doesn’t mean that doctors are gonna go away, but maybe even being a doctor shifts in the direction of being able to provide a human touch while also using these better tools, algorithms, machine learning, whatever, to help them do their jobs. What do you think? 

ADAM: I think we can start by dividing the jobs into the tasks where the human touch is very valuable and the tasks where the human touch is not as valuable.

And I think you can picture in personal care areas where a lot of automation could free up human capacity. So, for example, if you’re thinking about seniors who need care, perhaps there’s automated care that helps lift them outta their chair that’s always there watching — sort of just looking for problems, making sure they’re okay.

That does a lot of the simple diagnostics — sit up, stand up, walk around, that sort of thing. And then the places where the human touch is more valuable: talking to them and asking, “Is everything okay? Are you feeling any pain?” That sort of thing. Those are areas where now the humans can spend more time focusing on.

I think a great example of where you can see this sort of divide of a singular task into the human touch part and the automatable part can be seen, believe it or not, at the Olive Garden.

CARDIFF: At the Olive Garden?

ADAM: At the Olive Garden.

CARDIFF: What’s going on at the Olive Garden these days? (CHUCKLES)

ADAM: At the Olive Garden, they still have wait staff and the hosts.

So people still come see you. They sit you down at the table. But then, there’s a tablet on the table where you can order from it, if you want, or you can wait for the wait staff, and you can pay, and you can leave. So—

CARDIFF: Does this example come to you because you went there recently and you sampled it for yourself?

ADAM: I saw it firsthand. 

CARDIFF: Alright. 

ADAM: I saw it firsthand at the Olive Garden.

CARDIFF: They still have unlimited breadsticks?

ADAM: They do still have unlimited breadsticks, yes.

CARDIFF: All right. Hey, not bad, I gotta go back.

ADAM: And so if you think of the staff, the job of a waiter, you could divide it into two parts. The part where they come and they order the food, and they say, “What would you like? How are you doing?” And provide that human touch, provide that service, and maybe answer questions, maybe offer advice. 

But then, the other part of the task, which could be thought of as bringing the check and paying for it, you can do that very quickly.

And so, this frees up the wait staff to spend more time seeing more tables, seeing more customers, focusing on the part of the job where the human touch is very valuable and less on the part of the job where it’s not valuable: the part where you take the check, and you pay.

CARDIFF: Are you disturbed at all by instances of people using AI as companions? Because there’s been a lot of commentary about this: the idea that people now are finding friendships, or in some cases even romantic interests, with AI, with virtual machines. It goes back to the movie Her with Joaquin Phoenix and the AI played by Scarlett Johansson.

People worry about this a lot. There are instances where this kind of thing’s happening, people are disturbed by it. I gotta say, I also am a little disturbed by it, but this seems to be like something that might play a role in the caregiving aspect in the future. And this is something that I — again, I’m very open to where the future is headed, right? But man, some of this stuff is a little bit disturbing.

ADAM: Well, let me take a dodge on whether I think that would be good or not, and instead suggest that the way that the human touch works is to divide the task into the things where it doesn’t matter, and that leaves people with more time to focus on where it does.

So, instead of the staff at the nursing home having to deliver the food and help the patient get up and down and do all the dirty tasks that don’t necessitate the human touch, they can instead spend their time on the more personable part of it. On the part where the relationship matters, and spending more time building that relationship.

I think there’s a lot of context where people would be better off having more one-on-one time or small-group time with a person whose job it is to provide a service, but it’s not economical. Nursing homes would be one example. Classrooms are another example.

If AI or automation is helping to administer the tests and do some of the more rote instructional things in the classroom, that frees up the teachers to spend more time on one-on-one tutoring. So I think we should realize that in a lot of these contexts, time is very constrained by money and human hours right now. And it would be nice to loosen those constraints in ways that can lead to more of the human touch.

CARDIFF: What you’re pointing to though, as you shimmied your way out of the harder part of that question, was more of a roadmap, right?

You’re pointing towards a more ideal outcome here and maybe that’s where we’ll end up. But let me give you an example of something I came across recently that made me stop and think about this. It was an article by Martin Sandbu, who’s also an economist, writes at the Financial Times. Great writer, former colleague of mine, and he was writing about AI companionship. 

And one of the points he made was that if an AI is designed to be a companion and is customized for you, to be your companion — customized in some sense, either to flatter you or to make you feel better — what it really is something closer to an actual mirror.

There’s nothing upstairs, so to speak, right? It doesn’t have a presence of its own. Whereas like human connection, real human connection, is defined precisely by distinctiveness between two people, right? And yeah, that involves a lot of friction. Sometimes that involves a lot of frustration.

Sometimes it involves terrible things, difficult things, heartbreak and so forth, but you cannot have real human connection without the possibility of those things being a part of it. It has to be two different people. Otherwise, it’s just another version of loving yourself essentially. That seems not awesome.

It seems not ideal, and we’re talking about a future scenario where ideally the tasks are split between the human side is still handled by humans — the human touch, right? 

I’m worried about a scenario where what we end up with is a facsimile of the human touch being provided in a way that’s quite lamentable by AI, by machines or something like that, and I don’t have a good way to prevent that.

I like your version of it. I prefer your version of it. I hope that’s where we end up, but that doesn’t mean that we’re gonna end up in that place. You know what I mean?

ADAM: I would say that you cannot find an example of a technology—

CARDIFF: I like how nervous you got there. Your economist brain was like, “What is this dude asking me, man?” (CHUCKLES)

ADAM: Yeah. Since I’m not gonna be allowed to dodge it, let me say that there’s no example of a technology or major form of media or communication media that did not result in some sort of behavior that we wish someone wouldn’t have done with it. I mean, we’re young enough or, I guess, young enough and old enough to have been around when the internet sort of came of age.

CARDIFF: Correct

ADAM: And we can remember all the social panics around what people were doing on the internet, what they were doing with the internet. The egregious message boards and chat rooms, and what they’re learning. 

And this is true of books as well, right? There was Tales of Young Werther (sic) that — I think it was Goethe — they said it was responsible for spreading suicides. And there were all these people who read this book about a young man who killed himself, and then they were killing themselves. And that’s why they call it the Werther Effect today. This idea that news of suicide leads to more suicides. And so, you can’t look at that and say, “Well, gee, I wish we hadn’t invented books. We could have prevented that, right?” 

It’s just when you have these human-augmenting, human-informing, human-connecting technologies, it’s an inevitability that there’s going to be things done with it that we don’t like. And so we just have to bite that bullet, and that’s what moving forward looks like.

CARDIFF: I wanna be clear about something: I’m pro-technology. I think the ability of faster productivity growth to lead to rising living standards to make people’s lives better over time, it leads to the world we have now, which is so much better than the world of the past, and I’m so deeply appreciative of that. 

But if you’re gonna bring up the example of the internet. Well, yeah, in the long run it’s gonna be shown to have been quite a wonderful, magnificent thing, but also a lot of the things people warned about on the internet, those proved true also, right? 

The damaging side effects. What it seems to be, at least for now, the hit to people’s ability to concentrate. The spread of things online that polarize people, stuff like that, you know what I mean? People getting misinformation, those things actually do exist. And if we don’t pay attention to them, if we just assume that the future will unfold in a positive way, because technology always leads in that direction, that’s not really true, right? 

You have to be preemptive about this stuff. You have to be thinking in advance about the potential harms so that you can try to steer things in a better direction, right?

ADAM: Sure. I just think that, on net, this is what progress looks like and that’s what the future coming true—

CARDIFF: Messy, ugly sometimes. 

ADAM: Yeah, absolutely.

CARDIFF: But awesome also.

ADAM: How many people have been killed by cars? How many people have been killed by the automobile? Like every form of technology that allows something great, allows something bad to be done with it. And I think there’s diminishing returns to how much you can plan away those things.

There’s nothing we could have done to prevent books from being created in such a way that bad people did not use them in bad ways. There’s nothing could be done to prevent the internet from being used by bad and weird people in bad and weird ways. 

I’m not objecting to people trying to think these issues through, but when you look at people advocating, for example, we need to pause AI research for six months so we can figure this out. I think it’s foolish. I think it’s a fool’s errand and I think it is motivated by this idea that we can just control a big technological wave that’s really not in the cards.

CARDIFF: Yeah. And they can pry your Ford F-150 from your cold, dead hands, right? 

ADAM: (CHUCKLES) Are they coming for that? I didn’t know they— 

CARDIFF: (CHUCKLES) You brought up cars! You brought up cars. 

ADAM: I thought you meant the six-month pause people. And I was like, “I don’t—”

CARDIFF: Oh no, no, no.

ADAM: If they’re going for that too, they’re in big trouble.

CARDIFF: I’m with you on all these things. I agree with you on all these things. A pause seems like the wrong way to go, but it also seems to me like the wrong way to go to simply put aside the possibility for these things to be misused instead of trying to like confront that head on.

ADAM: I think accepting that it will be misused and that that’s inevitable is a way of confronting it head-on rather than to say, “Well, those things won’t happen.” Or to say that, “Well, we’ll simply pause until we figure out how to stop that.” To me, that’s not confronting head-on.

Confronting head-on is accepting technological progress. It’s coming, it’s big, and people are going to do bad things with it. It’s a big world.

CARDIFF: How are you gonna prepare yourself for the possible disruption from artificial intelligence, should it come to pass? 

For economists, people like that, people are saying, “Well, learn to code used to be the thing to do. And now the coders are the ones that seem at risk of being automated away, but the fact that they learned to code in the first place suggests that they’re pretty talented people. So they’ll be able to adapt.”

What is your version of adapting? You’re a smart dude.

ADAM: Well, like I said, I think the artist will be okay.

And I think what I do as an economist is a little bit of like being an artist, believe it or not. 

CARDIFF: (LAUGHS)

ADAM: I mean, if I think of something that I want automated away that can be handed off to a smart machine and still do my work, I do that all the time except the smart machines are our RAs, our junior staffers. I hand them off stuff to help my work all the time. 

So I’m already taking things that I could do and handing them off. I think a lot of what I do as being artistic.

CARDIFF: Learn to art!

ADAM: Learn to art, yeah.

CARDIFF: That’s the new ‘learn to code.’ 

Let’s talk about the part of your post where you discuss what happens if AI really does end up disrupting a whole lot of jobs and how that shifts society, how that shifts the economy. What happens to them?

ADAM: So I think that this is just as important as understanding the limits of automation is understanding what automation can do when it comes to work. And I think that it’s difficult to rule out a world where automation is disruptive to the labor market. It’s not my base case, but I think it is possible.

And so what can we do about it? And I think the human touch is very relevant here, because there’s two very different kinds of worlds…

CARDIFF: Yeah?

ADAM: …that are possible with automation. One is where it replaces humans in all contexts. And there’s nothing we do that’s more valuable because machines can do it all. That’s a very different world from a world in which we are automated, except in the areas where the human touch remains valuable. 

And so I think that that is more likely to be true. And so there will be demand for humans to do some things. There will be growing demand for humans to do things as income growth leads us to demand more in-person services and things like that. 

So what I would suggest for policy, if we end up in this world where it’s disruptive, there’s really two things you can do. One is you can redistribute income, right? If what’s happening is that a larger share of the gains are going to capital or a small share of labor, then we can redistribute.

That can be politically challenging, but it’s not a technically challenging policy issue. The harder one is: what do we do about work? What is it? What do we do if there’s widespread joblessness? And if we think that work has value, right? Some sort of innate value to how we see ourselves, how we function. In a very sociological sense, if work matters, then what do we do about that? 

And it’s very important in that context that the human touch is something that predicts there will still be things for humans to do. The question simply becomes, are they compensated enough? And that’s a policy challenge I think we can do something about. My colleague, Ben Glasner, and I suggest using a wage subsidy already for parts of the labor market where we think compensation is too low. So you could just do that in a much more widespread way.

CARDIFF: So there’s a few components to that. And earlier, you brought up the idea that, especially as people’s incomes go up, they tend to have more and more demand for products, goods, experiences where they can have the human touch.

This especially exists in the luxury market. So let’s say you do have a world in which there are huge gains to the people who own the capital that makes these AIs. And so you have a segment of the population that does fantastically well, and simultaneously a lot of people who are doing white-collar jobs, what you might call “thinky” jobs. 

A lot of their tasks shift. A lot of their tasks, in some cases, are just automated away entirely. Maybe they lose their jobs, and now you have this sectoral adjustment where this big part of the population gets super rich, but they’re gonna spend that money, they’re gonna spend their new surplus, and the stuff they demand then will rise as a big, new, prominent economic sector.

You do have the issue, in addition to the compensation angle of people who used to have very prestigious jobs as software developers or engineers now having to shift into jobs where they’re essentially serving the demand, the wants of the new elites, right? That kind of thing.

I actually think that could be a little messy. It could be messy in a lot of ways. I think politically it could be very messy. There’s gonna be calls for slowing down automation. And for the first time in a long time, it’s people with actual power right now who are gonna be resisting it. You see what I mean?

People may not wanna provide the human touch, especially if they trained as kids or as young adults to do something else, right? They thought they were gonna be the people who were gonna be buying the stuff that had the human touch. Now they’re the ones who have to provide it. And that’s an adjustment.

It’s an adjustment, by the way, that I should say, we have asked blue-collar workers to do many times over the years in manufacturing and other sectors, right? Now it’s gonna be white-collar sectors. I think there’s gonna be some resistance if that’s the world that we end up with. What do you think?

ADAM: I think if we have enough productivity growth to displace hordes of white-collar workers, so many that not only do their relative wages fall, but there’s widespread white-collar unemployment, you cannot have that without also having rapid productivity GDP.

That’s a truly disruptive level of innovation, technological change, and growth. You can paper over a lot of things with that kind of GDP. That gives you fiscal space to do all sorts of redistribution, to take income taxes to zero, right? To implement a $40/hour wage subsidy.

So if you can find any job, we’ll pay you $40/hour to do it, and the government will foot a percentage of the bill. Yes, large disruption would be disruptive, but that kind of large disruption would also give us a lot of fiscal space. And I think that those two policy challenges of redistribution and then subsidizing work, they’re not as insurmountable as the policy challenge of all of a sudden, there’s nothing for humans to do. 

CARDIFF: You’re pretty optimistic for the future, I think.

ADAM: I guess. I mean, there’s a reason that I’m laying out why we might need income redistribution and a wage subsidy.

I’m planning for what kind— whether if it could be economically disruptive, a true optimist would just say, “We’ll be fine. Don’t worry about it.”

CARDIFF: Yeah. And you’ve got me, the agnostic who says, “Prepare for all outcomes.” Not saying, “Get in the bunker.” I’m saying, “Just be vigilant. Be ready to retrain if need be.” 

Do you expect you’ll have to do significant retraining, or do you think you’re already well prepared for this shift towards a more artistic version of yourself, where you can use the automation?

Do you think you’ll have to do the equivalent of, I don’t wanna say going back to school, but doing some of your own deliberate retraining and something basic? And something new that emerges? And you’ll have to start, if not quite at ground level, at something like that.

ADAM: It varies so much person to person, role to role. For me specifically, if I never wanted to touch AI, I would just have Jiaxin and Sarah do the AI for me.

CARDIFF: Those are our research assistants, to be clear.

ADAM: Yeah. If I didn’t want to touch data ever again — forget AI, if AI went away tomorrow — if I didn’t want to touch data ever again, I wouldn’t have to. There’s people to help me.

There’s the junior staffers, they’re great at it. They’re really great at putting data together and running analyses that I suggest. I think, in my job, it’s not a have-to question. At the very least, if I want to be lazy, that’s the approach I’ll take. But in my job, I do feel like I need to get my hands dirty with the data. I need to do exploratory data analysis. 

I like to just take the raw observations, run my own regressions, and look at the shape of the relationships and the data. And so that part I think would prevent me from being my best at my job if I totally unloaded it onto either automation or junior staffers, but also something I just don’t see automation replacing.

I think that’s part of the art, but maybe I’m delusional. Maybe that is possible.

CARDIFF: No, that is at least sensible. It’s reasonable. 

I will say that if you see the part of your job that’s hard to automate as asking smart questions and arriving at interesting conclusions based on the results of the data analysis and so forth, you might just have more competition for that because the people who, right now, are doing the hardnosed gritty data work, if they see that, they’ll be able to automate that away as well. 

Well, they’ll join you up there in the lofty space of asking the questions and trying to think of novel thoughts based on the conclusions, and trying to say interesting things. You see what I mean? 

ADAM: That’s true. I would say that’s a continuation of a trend because, not that I was there to experience it, but the bars to data analysis were a lot higher in the 1960s, 70s, 80s, 90s, and all we’ve seen over time is that the bar to entry just keeps getting lower and lower.

First, it was great, you don’t have to use punch cards anymore, right? You don’t need access to the timeshare supercomputer. And then we’ve all got our personal computers. That makes it a lot more competitive if your skills are data analysis and writing about it. And then, widespread improvements in free statistical analysis software has made it easier.

And so this would represent a continuation of that trend. I don’t think anything brand new.

CARDIFF: Last question. You read a lot about the history of automation. You like history, you like economic history, you like the history of economic ideas. Why don’t you leave our listeners with one book you’d recommend to really ground themselves well in these themes, even if it’s about a niche sector of the economy, even if it’s about something specific, but can open a window into these kinds of topics?

ADAM: So, one of my favorite authors to read on automation is James Bessen. He’s an economist who has just, I think, some of the most insightful work on what happens when there’s rapid productivity growth, what happens when things get automated, and he really focuses you on what happens to demand when the price of goods falls and what happens at the firm level.

And I think that his framework for thinking about productivity growth and automation is really powerful. So I would suggest reading some of his papers.

CARDIFF: All right. Grab something from James Bessen. 

Adam Ozimek! Long may you persist in human form and not just whatever the AIs decide is the best representation of you.

ADAM: Except for those tasks which can be automated efficiently.

And same to you.

CARDIFF: Thanks very much for being here.