Welcome back to the Economics of AI. The topic of our second module is how to describe technological change in our economic models. So what are the big picture themes of technological change in the age of AI? We'll start with materials that some of you will view more as a refresher, talking about production functions into properties, returns to scale, et cetera. If you are well-versed in production theory, you may want to just skim over the first three videos and start with the one on modeling technology. But just to be sure, you will need all the materials that I am covering in this module for the rest of the course. After that we'll cover specifically how to capture technological progress in economic models. I'll go through our most common models of progress. My main message will be that the way that we economists typically describe technological progress, for example, an increase in some technology parameter A, is really problematic. In the past, it may have been a decent description from a high level perspective, but our models have a really hard time capturing the main problems of technological change that we are concerned about in the age of AI. So what are these main problems? The first one is that all our traditional factors of production, like capital and labor, are increasingly being displaced by information goods, we could say by digital factors, and these follow fundamentally different laws than what we are used to. In traditional economics, when I use, say, a shovel, you can't use it at the same time. We say that the shovel is a rival or rivalrous good, or a rival factor if it's an input to the production process. That's no longer true with information goods. We'll first observe that the state of technology itself is one of those nonrival goods. You and I can both use the same technology without subtracting from each other's efficacy of using it. This is not new to AI. It has been true forever, even when we went from the Stone Age to the Bronze Age. However, back in the Bronze Age, nobody was thinking of patenting smelting to combine copper and tin and produce new bronze weapons or armor. They went the more direct way to convert their new technology into material riches. Instead of applying for a patent, they just went over to slaughter the neighboring tribe and take all their possessions. In the age of AI, we would consider that a little savage, but we have developed a system of intellectual property rights that has, in some ways, similar material implications: to extract material riches from others based on technological superiority. What I will emphasize is that our modern economy is increasingly built around new forms of information goods and digital factors that are nonrival. Software, big data, blueprints and designs. And they all behave in fundamentally different ways. Whenever we deal with information goods, there are fixed costs and extremely low marginal costs, and this combination gives rise to increasing returns to scale and natural monopolies. So these new information goods make up a larger and larger part of our economy, and their owners correspondingly reap large rewards. The flip side of all of this is the diminishing role of traditional factors, especially the diminishing role of labor. We'll have an entire module devoted to labor in two weeks, but let me just tell you that one of the most striking economic developments in recent decades has been the decline in the labor share of output, meaning that the fraction of total output that is earned by workers has gone down. Technological progress has been biased against labor in that sense. And even if you look at the distribution of income amongst workers, most of the benefits of progress have gone to the highly educated, you may say to the superstars of the economy. Most of our classic models of technological progress have a hard time formally capturing that technological progress can make some parts of society worse off. But I will show you that this isn't because there are any economic laws that would say so. It's quite to the contrary. Our economic laws explicitly say that progress can make some people worse off. It's rather for basic reasons of algebraic convenience. For the past several decades, we economists have been using functional forms and economic models that are very easy to deal with but that have blinded us to the likelihood that progress will make people worse off. It was simply the trade off between convenience versus realism. We'll talk about how to best capture the decline of labor in economic models to better reflect the reality that we have been facing in recent decades and that we may continue to face going forward in the age of AI. A third theme that we will briefly cover in this module is the role of production networks. This has been an exciting area of research over the past half decade as more and more data on input-output linkages has become available. And I want to provide you with a quick primer of it, so you know how to capture technology at a more granular level. But now without further ado, let's start with the economics of how to model production processes.
