Recently, I’ve been incorporating generative artificial intelligence into my day-to-day work pretty often. Specifically, I like to ask ChatGPT to help me write Excel functions. I’m not super familiar with all the syntax of Excel, so this can save me a lot of time when I’m trying to work on an extremely large dataset.
As a busy policy analyst who finds excel functions tedious to write sometimes, this is an amazing way to increase my efficiency. I can outsource some of the things I am inefficient at and spend more time on more valuable things.
Of course, there’s a catch. Writing excel functions may be a small part of my job, but it might be most of the work for another person’s job. That person might not be so thrilled that their extremely valuable skill can now largely be automated.
Computers are better than humans at a lot of things. Often we assume that computers only excel at things with certain outcomes like mathematical calculations, but are incapable of performing creative tasks.
I’d argue that this is not actually true, and that computers have actually been able to perform extremely creative tasks for decades now. For example, Stockfish and Google’s AlphaGo have been playing Chess and Go respectively at levels that no humans can match for quite some time. Chess and Go are far too computationally complex for these algorithms to “solve” the games, and some would argue that they find unique and creative ways to approach their games.
Instead of computers being incapable of creativity, I’d argue that computers have lacked flexibility. The algorithms that play complex strategy games are immensely powerful and creative, but they could never be used for anything other than the games they were designed for.
What makes the new era of AI models so fascinating is that they are extremely flexible in what they can do. Writing excel functions used to be difficult for a computer because it is context dependent. Now AI models can take a description written by a human and output working code for nearly any problem.
As we begin to think about how AI might fundamentally change the labor force as it continues to develop, we need to understand from a theoretical perspective how the labor force is currently constructed.
A macroeconomic model for a labor force with AI is presented in a new paper by Anton Korinek from the University of Virginia. Currently, economists describe our economic output using a simple production function
Y = A · F(K, L)
In this function, Y is the total output, A represents the level of technology, K stands for capital and L stands for labor. Essentially, our economic output is a function of capital and labor that is scaled by our level of technology. In a world with extremely sophisticated AI models, Korinek suggests that we be exposed to a new production function:
Y = A · F(K, L+M)
Here, M stands for machines, which represents AI and robots that can replace labor. Korinek assumes that at some point of technological advancement, machines will serve as a perfect substitute for human labor.
Whether or not AI can actually become a perfect substitute for human labor is still an open question. In particular, I wonder if AI will be able to replicate the social dynamics that are extremely important in some jobs like teaching. Regardless, policymakers are going to have to grapple with a new reality where labor can in many cases be easily replaced.
While the efficiency gains from AI can lead to increased productivity, the reduction in employment is going to require a change in the way our economy works. Currently, most people get some slice of the total output of our economy as a reward for contributing to its creation, often in the form of wages for labor.
An AI-driven economy will theoretically be able to generate far more output, but we need to come up with a new way of distributing that assuming that most people won’t be directly contributing to its creation in the same way. Finding a way to do this that is equitable and efficient is going to be an essential challenge of an AI-driven economy.