🤖 Are we overestimating AI's potential economic oomph?

The guesswork continues, but I remain optimistic

Super-Short Summary: A new study by economist Daron Acemoglu looks at the economic impact of generative AI. Using data from recent experiments, Acemoglu estimates that GenAI could lead to “nontrivial but modest” increases in productivity and GDP over the next decade. While these gains are much lower than some other forecasts, there are additional channels for faster productivity growth not included in Acemoglu's calculations, such as AI creating productive new tasks and revolutionizing scientific progress. Stay positive, people!

As you read this essay, keep in mind the following bit of analysis (I will return to it later): Replacing the least effective five to 10 percent of American teachers with merely average-performing teachers could significantly improve the United States' international education ranking. Such a change could raise the US from its current mid-teens position to the mid-single digits. This performance improvement suggests annual economic growth would be 0.8 percent higher. 

Now to AI. It’s still early days for generative AI such as large language models. And it remains to be seen just how much GenAI will affect worker productivity, a big theme of this newsletter.

That said, there have been several promising experiments. One study you might have heard of looked at the impact of GenAI on a software company’s call center. Stanford and MIT researchers trained an AI tool on data from over 5,000 customer service agents. The AI assistant then monitored customer chats and provided real-time response suggestions to agents, who could choose to use or ignore them. The study found that access to AI assistance increased agent productivity by 14 percent, with the most significant gains observed among less-experienced and less-skilled workers, possibly because the AI helped to transfer knowledge from more experienced agents to novices, accelerating their learning curve.

On such improvements, a better economy is built. Still, the study is more proof-of-concept than proof the US economy is on the verge of a step-change higher in productivity and economic growth, much less suggesting the Technological Singularity is nigh. That 2023 Stanford-MIT study is one of several experiments making an appearance in a new NBER working paper by MIT economist Daron Acemoglu, who writes frequently on the economic impact of AI. Those experiments document “nontrivial productivity gains from generative AI, largely driven by improvements for less productive or lower-performing workers,” Acemoglu writes in “The Simple Macroeconomics of AI.” As the economist sees things, “nontrivial but modest” productivity gains are pretty much what we should expect from GenAI going forward.