Week 1 of my new Economics of AI class

For those interested in the “Economics of AI” syllabus and class I’ve been developing, I thought I’d share how it’s going. This week was the first week.

My first week in this new "Economics of AI" class went well I thought. But the first week tends to always go well. I thought I'd share it nonetheless just so people could see what I'm doing. First, the coin flip. Will this be a paywalled substack post or will it not be? I had Cosmos run a python script where he flipped a coin 9 times, and 6 times it came up tails, which means that the substack will not be paywalled.

First, here is the syllabus. And here is one of my lectures handwritten. I’m not going to make slides this semester I don’t think. I just feel like if I make slides, I’ll use ChatGPT to make them, and they’ll suck. So for now I’m hand writing the lectures like the old days. If you want to see what I’m doing and how I’m going about this “economics of AI” class, that’s at least a broad sketch. But what did we do this first week?

The first week I needed to do a review of the relevant economics. As Acemoglu's work is sort of central, I am focusing mainly on macroeconomics and labor economics. And to teach the macro, I went back to a 4th edition of Frank and Bernanke from 2009. I used it for years to teach, so I know it best, and felt that the applications I had in mind were ultimately coming from that anyway.

The nice thing about using a very old textbook, though, is that the empirical examples are outdated. Since part of the class is to have them use generative AI to teach themselves things, particularly python, I decided that the first assignment will be to extend all the data to 2024. I am just not entirely sure yet how to make that assignment, but I'm going to work on it tomorrow. I'm hoping there is a way that I can combine an assignment that helps them set up python and use some API to pull time series data on basic things like GDP, per capita income, and various things we are discussing. I'm sure there is, I mean, but whether I can do it is my point in a simple assignment. It only occurred to me as I was prepping, which is why I'm hesitating.

There are a handful of definitions of economics that float around. Instructors of principles classes typically use one of a few definitions that they know by heart. The one I used to use was by Lionel Robbins which basically went something like “economics is the study of the allocation of scarce resources with people of unlimited wants” or something like that. I have been opting for a different definition lately for my history of economic thought class, which I’m carrying over here to the AI class. It is from Paul Samuelson, who was kind of channeling I think Adam Smith. He said economics is a science that asks and answers three questions:

1. What will we make (i.e., production)

2. How will we make it (i.e., technology, factors of production)

3. How will we share it (i.e., markets, prices, income, income inequality and so on)

You can see Robbins’ definition in there, but his brings in the utility maximizing actor (“people with unlimited wants”) whereas my definition really doesn’t at all. The sharing could happen via utility maximizing agents, for instance, but that’s the market economy. You could have a tribal economy in which there are no markets. Anyway, point is, that’s the starting point.

We then dove into the measurement of GDP. I spin a story in which I tell them that there was early on not as much of a distinction between macroeconomics and microeconomics as there is now. Adam Smith was focused on “the wealth of nations”, meaning prosperity measured maybe as something more like aggregate output. His invisible hand explanation focused on profit maximizing firms investing in capital, increasing output and leading to hiring more workers in the future, creating a virtuous cycle of economic growth.

I’m not really sure this is exactly accurate, but I take some liberties to keep the story moving. But I basically position the labor theory of value as noncontroversial and just say that to Smith, labor had to logically be the main source of value because if you stripped all economic activity back to the very beginning, before the agricultural age when humans had no property and no capital (his example of this being North American native Americans), then things were worth ultimately whatever it took to make it using labor inputs, specifically the exertion of time. Therefore if you could make an arrowhead in 2 hours or a beaver pelt in 1 hour, then the arrow head was “worth” 2 beaver pelts.

So then I say to at least some of these early classical economists, they took it for granted that labor was the source of value in all production, even as they focused on production itself as the North Star. The problem was it was really a failed project for Smith and then others too to actually make something out of that. Once we move into the commercial stage of history, where time use is not the sole determinant of things we make, but rather machines and capital are too, then two things tend to happen. First, economists would argue the machines and the capital were both produced at earlier stages using labor, so they too are endowed with labor. Second, it still became next to impossible to quantify anything reliably with this theory of labor as the source of value in aggregate output. Ricardo would have all these tables trying to count up the amount of labor through sequences of production, but it just seemed like it went nowhere.

I basically spend maybe 10 minutes on this, trying to just tell a story of where we came from. So then I say that others came along later with other theories of value, namely utility, and some would even go on to say that the real economic output that societies were producing weren’t cars and cotton and so forth, but utility. That utility was the wealth of nations. I then draw this picture.

And then there’s other people who also tend to emphasize the circulation of money, as opposed to production only, from which we get theories of recessions or what they called “gluts”. Malthus being the first one whose theory of gluts is quite similar to Keynes’ theory developed later.

So I don’t spend much time on this, as I said. I just am trying to really lay down the framework where artificial intelligence will become relevant. As AI will affect (1) the production processes, (2) through technology and the use of labor and capital inputs, and (3) it could very well also affect the distribution of that aggregate output. And to talk this way, I think they need to have a refresher of understanding what GDP is. So that’s when I introduce them to GDP.

It’s fairly fast, but not superficial I don’t think. I emphasize that the aggregate output is measured. Smith had tried to tinker with measurements in place of the labor counting method, though I do get confused a little about Smith on this. But I basically say Smith had tried to root the value in something stable, choosing gold prices, but then cannot give an account of where the gold prices come from. I basically skip over everything at that point and just say “point is, all this stuff gets folded in over time into what is now called neoclassical economics. Neoclassical economics excises the labor theory of value to some degree, but what it keeps, it merges with this utility approach, which is where we get the supply and demand model of prices.” It’s not exactly wrong, but it’s definitely a major simplification.

But the point is I’m trying to get us to measurement of GDP, without losing sight of aggregate output and production. And that’s when I just remind them of that one definition of gross domestic product: market value of final goods and services produced in that period in that country. I find this is helpful actually because then I can just keep harping on the fact that the “wealth of nations” concept is not the money exchange, but rather it is the production of new goods and services which are distributed (third question) via money exchange. I think people tend to see all buying as the economic activity, and so to them, buying a car — used or new — is really the same thing. And it is the same thing to the person, but it’s not the same thing if the focus is on production. So I say all that.

Anyway, I go through a standard macro lecture about measuring GDP is my point. I then describe the different ways that it’s measured. I discuss the production method that I just mentioned. But then I note that everything produced is allocated to someone via spending and discuss the four groups: household consumption, firm investment, government spending and net exports. And then I say everything produced is purchased and becomes income for either owners of capital or labor. And I note that labor share of income are thought to be around 60% of GDP and capital the rest.

And that’s it. That takes us through GDP. That’s at that point all I wanted to show them. We then go into labor supply and labor demand. And all we did this week was just document five trends and I started to support them with figures from Goldin and Katz’s The Race Between Education and Technology. I pause and tell them about Claudia Goldin winning the Nobel Prize and winning it for her work in economic history and labor as it related to gender. Remember, many of these students have not had upper level econ classes, and some are not Econ majors. It was a popular class and drew broadly, and I only required that they’d had a basic principles class, and we have both a principles class for majors, and we have one for non-majors on other campuses. So I use this as an opportunity to just say that Goldin’s work could be considered fitting into this material on production, even though at first glance you might say what does gender have to do with production? I tell them about their Power of the Pill paper — which mind you is a paper about technology, specifically oral contraception technology. I explain to the students that Goldin and Katz in that paper try to empirically test whether or not the availability of oral contraception helped women work. I don’t get into the subsequent work by Martha Bailey, Caitlyn Myers and others, but I just kind of collapse it all and say that that work has suggested that the pill allowed women to increase schooling, delay age at first marriage, and enter the work force by timing their fertility to suit their goals. The females, in particular, seemed to nod their head and understand. I say basically “that’s production. That’s production because if they’re working to produce final goods and services, then it increases GDP”.

Of course, they were already working though — they just were often producing non-market goods in the household. It’s hard to bring up any one thing, though, without feeling pulled to say ten other things, so I try not to. I try to just keep their eyes on the ball — production. Making stuff. Women working, women are making stuff, it’s increasing technically GDP as a result.

But back to the trends. I note five trends:

1. rising real wages over the 20th and early 21st century

2. a slow down in wage growth around mid 1970s

3. income inequality

4. increased employment

5. stubbornly high unemployment in parts of western europe

This is where I just pull out the excellent Chapter 2 of Goldin and Katz’s book and we just go through pictures. That chapter, and other chapters, are just replete with excellent time series and histogram plots showing income inequality in a variety of really intuitive ways. I put it on the document camera and I show them rising incomes over the 20th century — something it really seemed like they’d never seen before to be honest. And then I note the decrease in inequality over this period too. Goldin and Katz typically will show this using a log specification of log(95/20) percentiles in the income distribution, which becomes log(95) - log(20). So when it’s, let’s say 0.45, then it’s e^0.45 equal to around 1.57 and means that the people in the 95th percentile are making roughly 57% more than the people at the 20th percentile.

I wish I had them here, but I don’t. I had to come home yesterday because I’m sick, but I’ll post them so you can see them later. Point is, going through the rest of chapter 2 of GK on Tuesday is the goal because what I want them to see is three things: that real earnings rose over the 20th century, that income inequality fell until around 1975 then rose, and that income inequality has grown because of stalled growth in earnings for the median and 20th percentiles and so on, while the 95th percentile has just continued growing like everyone had been growing before.

And that’s really it for this week. It’s just laying basic facts of history down for them, helping them think in terms of making stuff, and linking that to distributional questions, but then showing that using GK. Once I get all that laid down, then I can move into discussing more economic history with labor productivity driving economic growth, and discuss the Industrial Revolution. I’ll be discussing parts of Allen’s book on the Industrial Revolution, emphasizing the connection between discovery and the location of economic activity in the Industrial Revolution as connected to wages, capital and energy prices. But I think I will also then have them read their first paper — it’ll be this one by Acemoglu and Johnson. I think I’ll have them read this for us to discuss on Thursday next week. It’s running example is textile workers and power looms. It’s about how the technological advances did not cause broad prosperity, but rather the structure of income changed with more income going to owners of capital and worker wages staying flat if not falling.

I actually think this is holding together well so far. And what’s good is that with these 2 lectures I just discussed going back over basic macro and basic labor supply and demand (I skipped that but I’ll teach them that on Tuesday), I think we are really setting ourselves up to read the first AI paper which is this Acemoglu and Johnson paper. It flows really well, pedagogically, with how we go through Goldin and Katz, though I have to be careful that I am not going too fast. Still, point is, they need to understand the broader context about machines and labor and economic output to understand automation, to then understand 20th century computerization of work, skill biased technological change, to then constantly be interacting with the “economics of AI”.

Then we will probably start the Melanie Mitchell book on AI in the 3rd week, but I’m just not exactly sure how I want to do it. Anyway, I’ll figure it out. Very soon, though, we will be going into Acemoglu’s “The Simple Macroeconomics of AI” and Restrepo’s article on automation. It's think it’s just going to be a bit of a leaky boat, or maybe like building the airplane while we’re in the sky. Whichever those two metaphors work. I have a plan, though, and I’m going to drag us through it.