AI and the Returns to Education in Poor Countries

An impending education gap

Most people think AI will transform the global economy, but few are asking how it will actually reach and uplift the world’s poorest people. I believe solving that question is essential for the future of development.

Today, I'm wondering about the way that AI affects the incentives in poor countries to go to school. About 25% of the gap between rich and poor countries can be directly attributed to differences in human capital, and that doesn't even include the ways that it feeds into productivity through better institutions or agglomeration effects. In countries where educational attainment is already low, will AI eliminate or complement the incentives to go to school?

Two Effects

I see two primary effects of AI on the returns to education in poor countries. First, there will be an effect where it depresses the returns, and then the second could increase them.

Until now, advances in information technology have increased the returns to education in these countries. As the world became more connected, businesses realized they could outsource call centers to India. As these connections grew (and as Tim Ferriss spread the word), people started hiring virtual assistants around the world. By one estimate there are 40 million worldwide. These assistants do more than just answer calls. They enter data, coordinate calendars, and write essays for your development economics class (you know who you are).

But these jobs are exactly the ones that AI is most likely to replace. Indeed, they already are. Africans who made a living writing essays for college students have seen a steep drop in income. Online freelance websites are seeing a 21% decline in posts for jobs that can already be done by ChatGPT. Even I wrote about how ChatGPT was doing data entry tasks that I normally would have to hire someone to do.

As AI replaces these jobs, the returns to education fall. Marginal students decide education is overrated, and enrollment falls.

But there's a countervailing effect. To understand that effect, we have to ask a few questions. What is the value of education in a world with AI? If it gets smart enough to match PhD-level researchers, is it even worth investing in education?

For most people, the value of education is not in the stuff you learn in school. Few are making a living applying the FOIL method from Algebra 1. Yet businesses are willing to pay educated people more. That's because education is bringing some sort of productivity benefit. But, given that you don't use the stuff you learn, what is that productivity benefit?

The value of education is in training students graduate from simple tasks to complex tasks. What's the difference? Simple tasks have little variation in the daily challenges. Complex tasks have a lot of variation. I like to think of the difference as building an iPhone vs coding an iPhone app. I cannot build an iPhone. But you could sit me in on the iPhone assembly line and tell me to solder a chip to the board. It would be hard at first, but it's the same task with every phone and there's little variation in what I need to do, it's a simple task. Coding an app, on the other hand, has significant variation in daily tasks. I once worked as a software developer, and my daily work shifted nearly every week. One day I would be adding a button to the app that triggered an action, the next week I would be digging through legacy code to fix bugs. All of it was software development, but because the work varied so much, it was a more complex task.

One study, appropriately titled Why Are There Returns to Schooling, demonstrated the benefit of education to tasks with two cases. First, in America, when contraception is simple (such as a birth control pill) it is equally as effective for high school and college graduates, but if it requires more complex practices (such as the rhythm method), it is nearly as effective as the pill for college graduates but completely ineffective for high school grads. Second, in India, the Green Revolution brought high-yield variety seeds, but they required more complex techniques than the traditional seeds. The farmers who were most likely to profit from these innovations were those with more schooling. Neither task was taught in school, but education complemented both.

As AI advances, the winners will be those who understand how to use AI to achieve increasingly complex tasks. Indeed, this divide has already begun. We're already using AI to do simple tasks. Analyze this memo. Write a Python script that does a specific function. But complex tasks require the human and AI to work together. And by work together, I mean the human needs to find out how to heckle the AI until it does exactly what he wants it to do, and he needs to discern when the AI has not yet achieved it. I'm hypothesizing that users with more education will be better equipped to do this. And when they do this, they will be significantly more productive, which will bring them higher wages.

So while AI could eliminate the current high-paying jobs in developing countries, it will hopefully raise the returns to education. But there's a barrier.

AI-induced Education Gap

There's a reason the virtual assistant jobs came to developing countries. They are right in the complexity sweet spot. Talking with customers about a specific product is complex--you have to speak English, adapt to accents, manage tempers--but it's also simple--the number of problems you have to deal with is small, and if you're ever in trouble you can escalate it to someone else. You need education, but you can get by with going to high school. And there are a lot of people who are on the margin of deciding between continuing at high school and dropping out. These jobs increase the returns to education on a margin where the elasticity is really high. This is backed up by an entire literature on the things that induce kids into primary or secondary education, but I'll just give one example relevant for this discussion: when a call center was established near a village, more kids started going to school. (Ironically, the social institutions around getting jobs meant that girls benefitted more than boys.)

But the complex jobs that AI will enable will increase the returns at a much higher level of education. It's not impossible to expand higher education in developing countries. When the expected returns to IT increased, India expanded IT education; and when the expected returns to nursing increased, the Philippines created more nurses. But since so few in developing countries are even positioned to graduate from college, the elasticity on this margin is much smaller.

So one force is acting to decrease education on a margin that's highly elastic, and the other is pushing to increase education on a margin that's inelastic. On net, that looks like the demand-side pressures will decrease education in developing countries.

Could we do anything with AI to counter this downward pressure? One way is to use it to shift the supply curve by decreasing the costs of education. That's an issue I'll address in a future post.