AI & productivity: the economic effects

If the numbers go up and to the right that has to be good, right?

For now we are the puppetmaster, and AI is the puppet that makes everything better. It’s unclear for how long this metaphor will last.

Years ago, I worked as a financial analyst at Citigroup. I managed a $120 million capital budget for a years-long project involving the relocation of its data centers from the northeastern United States to a more geographically dispersed, and less risky¹, set of data centers across North America. I had nothing to do with the development of the data centers or the selection of sites for these data centers. Rather, I sat in an office in New York City, and collected and aggregated information from across the bank’s internal operations teams, and compared the project budget to the actual project spend. What I quickly realized was that forecasting is more art than science. Sure, there are trends, and these trends can be extrapolated. You can extrapolate linearly or non-linearly. You can tart up your extrapolations with things like moving averages and other pseudo-quantitative reasoning, but at the end of the day, forecasts are only as good as their comparison to actual results.

I think about this a lot when I read forecasts about the future, as it pertains to artificial intelligence and its effect on the world. One of the bullish claims about artificial intelligence is that it will generate a sustained period of higher economic productivity. As the future is inherently unknowable, these kinds of forecasts always remind me of my time at Citigroup. Is the forecast accurate? Well, if you consider these assumptions and those variables and you squint your eyes, sure, you can see a scenario in which this forecast is accurate. But that’s a very hedged statement: any one of these variables could be off, by an order of magnitude, and, well, the future could look very different from what has been forecast.

One of the more common AI-related forecasts being bruited about is that AI will usher in an era of sustained higher economic productivity. Higher economic productivity is good! It would solve many of our fiscal issues. So let’s think a bit about what this means. We may well be a few years away from a period of much higher economic productivity, and the second-order effects of that could be fairly profound.

A sustained increase in productivity has a significant effect on a country’s Gross Domestic Product (GDP). GDP is the total value of goods and services produced in a country over a specified period, usually a year. It’s a primary indicator of a country’s economic health and growth. Here’s how increased productivity may affect GDP:

And there are reports! Here’s a report from Goldman Sachs, titled Generative AI could raise global GDP by 7%:

Breakthroughs in generative artificial intelligence have the potential to bring about sweeping changes to the global economy, according to Goldman Sachs Research. As tools using advances in natural language processing work their way into businesses and society, they could drive a 7% (or almost $7 trillion) increase in global GDP and lift productivity growth by 1.5 percentage points over a 10-year period.

“Despite significant uncertainty around the potential for generative AI, its ability to generate content that is indistinguishable from human-created output and to break down communication barriers between humans and machines reflects a major advancement with potentially large macroeconomic effects,” Goldman Sachs economists Joseph Briggs and Devesh Kodnani write in a report.

Goldman even provides us with this handy graph, which shows stuff moving up to the right:

Look, I get it. Goldman Sachs is in the business of selling counsel to corporations looking to issue equity or debt, or to sell themselves, or pieces of themselves, to other buyers. So Goldman has to tout the economic benefits of AI: there will be a boon in wealth creation over the coming decade so you (and your board) had better get on board with the AI hype train! And, of course, let our bankers lead the way.

Yes, I’m being cynical here. There is something to the claim that AI increases productivity. Just look at all the programmers who tell us how amazing Github Copilot is. Here’s research from Github about how productive users of…Github Copilot are:

Look, these are all important things, and they may even be true. And, if they are true, then, yes, AI is making people much more productive. And if people are sustainably more productive, then, sure, over time, GDP grows, we all become wealthier, and everyone’s happy.

And—there’s a lot of reason to think that these forecasts are directionally correct. I don’t know whether GDP will grow over the next decade by a percentage point or more than it would have in the absence of AI. I know that a lot of prognosticators cite suspiciously specific forecasts that have the patina of quantitative certainty. I’m less interested in the certainty of forecasts and more interested in the directional arrows of progress: I want to develop an intuition for general trends. I care little about the details. The future is too high variance for forecasts to be of much use. What is more useful to know is this: AI technology is rapidly improving, and it is making many people much more productive at their jobs. Over time this will affect the economy in ways which will surprise us, and which will redound to our benefit.

1. Citigroup initiated this project in the aftermath of 9/11, at the behest of its regulators, who believed that its concentration of data centers in the NYC metropolitan area was unacceptably risky. A natural question which arises, in 2023, about this, is: why didn’t Citi sign cloud computing agreements with Amazon, Google, etc.? For one thing, cloud computing wasn’t really a thing around 2007-09 when I worked on this project. For another, banks operate under a set of onerous regulations which significantly constrain how they can conduct their operations. I don’t know whether those regulations have changed since I worked at Citi. Banks may be more amenable to cloud computing now.