Tyler Cowen asked when AI's leisure dividend will arrive, since so far people seem to be working more. Garicano answers that it may already be here, just hidden, and recounts what a Chinese CEO told him and his co-authors over dinner. The last line is hers, as he reports it.
You lose certain bits of the job. Certain things get unbundled. Other things don't get unbundled. Other things, I mean, we talk about Jean and Jan Hui and me talk about strong bundles and weak bundles. Weak bundles, the job kind of breaks into pieces and some parts are done by the AI. When the job is a strong bundle, the AI kind of comes into your bundle and helps you, like in a relationship manager, an account manager for a bank who gets to prepare the meetings better, but still needs to have the cognitive task bundled with seeing the clients. And I think that part doesn't quite require staying as close to the frontier as you and me are staying. I think that part requires knowing that the LLM can help you prepare the client meeting. And you ask the client, the LLM, please help me prepare this client meeting. Here are the budgets and the balance sheet statements of the client. Help me see which accounts, which products of the bank I could offer. And I don't think that needs really to know what the actual frontier model is and how does it differ from the previous ones.
Now, as you well know, virtually all new technologies and productivity improvements, they lead to more leisure over time. And in a good way, I don't mean unemployment, actual enjoyed leisure. If I think of the current day, there's some people, like the two of us, we choose to work more to learn about AI. And then there's other people, they feel they have to work more to learn about AI, to not lose their jobs. And so far, there's no leisure dividend. People basically are working more. When is the leisure dividend going to arrive?
I'm not sure that's the case. Let me give you two data points. So one is a Chinese CEO I had dinner with, my two co-authors are from Hong Kong University. And we had dinner with this CEO who told us she sees every one of her IT people more productive. She looks at the productivity of the overall IT department. Nothing happens. And quote, she said to us, I think they are smoking the productivity gains in the corridor. The second is from the recent NBR Summer Institute. There was a paper in health which gave doctors scribes and they all adopted. I mean, you know, adoption is slow. People are... Well, the automated scribes went up to 100% of adoption really, very quickly. You see doctors scheduling more meetings and doing more stuff. And eventually, and there was a recent New York Times article from a doctor exactly arguing this. She was saying, you know, I have more time for my yoga classes. So she was actually using the scribe, which transcribes her meetings and which stops a lot of the bureaucratic work that is nasty for a doctor. My wife is one, so I know how much they hate that. And she was actually doing more yoga. So, and this actually shows up in this paper to some extent. So I do think that what we see with this usage, this shadow usage of AI in the job, which is people have their own computer and they actually ask questions that maybe their corporate AI doesn't allow them to ask, has a lot of welfare gains in the form of leisure and lower work stress in the job because you just ask the LM, you post it there and you just spend some time browsing the web. So I am not as sure that the leisure is not expanding, but definitely not mine. I've never worked as much in grad school.
But say in the short run, when taxes change, almost always the substitution effect overwhelms the income effect, right? Tax cut, people work more. If taxes go up, people work less. Of course, we see this in Western Europe. So aren't these AI productivity improvements a bit like tax changes? You're more productive. In the aggregate, if doctors are more productive, they can earn more. They'll work harder. I'm not complaining about that fact, but maybe we should expect that instead of a leisure dividend.
Both things can happen. I mean, we are seeing Jevon's effects in many of our own utilities and in many of our own decisions to work. It's definitely the case for you and I, as you said. I was just for the amusement of our readers, I guess. Many of you have a whoop. I was really very happy with my whoop until January this year. I was 53 years old in actual age versus 59. Sorry, in physical age. Since January, it's been steadily declining. I'm now 57. I grew four years in six months. And that's because I am not sleeping a lot and I am kind of running things into the night and going to bed things. Thinking, oh, I should have done this other thing. Why don't I try this? So, yes, it definitely happens to me. So, why wouldn't it happen to those doctors?
Is there a plausible scenario whereby Spain, in particular, and I don't mean the EU, I mean Spain, achieves AI sovereignty? Now, you've written about this on your sub-stack, Silicon Continent. What's your view right now? And keep in mind, the whole stack in some way has to be under your control, right?