Artificial Intelligence

AI is showing "very positive" signs of eventually boosting GDP and productivity

Goldman Sachs Research predicted last year that generative AI could boost GDP and raise labor productivity growth over the coming decade. Since publishing that outlook, investment in generative AI has boomed, but it will take time for the technology to filter into the overall economy.

“Until we’ve seen more significant uptake in the actual application of AI, in the regular work production process, I don’t think that we’re going to see as big of an impact on productivity,” says Joseph Briggs, who co-leads the Global Economics team in Goldman Sachs Research. Briggs wrote last year’s AI report with Goldman Sachs economist Devesh Kodnani.

“That being said, the early signals of future productivity gains look very, very positive,” he adds.

While adoption of generative AI is lagging investment in the technology, Goldman Sachs Research sees potential for AI to automate many work tasks. It’s expected to start having a measurable impact on US GDP in 2027 and begin affecting growth in other economies around the world in the years that follow.

We spoke with Briggs about how the team’s forecast has held up over the past year, which businesses are adopting generative AI, and the technology’s impact on the labor market.

Based on what you’re seeing from the data, how are we tracking against your productivity prediction?

We haven’t seen much of an impact on productivity growth so far. But the reason being is, even though we still see a lot of potential for AI to automate a lot of the things that workers do on a day-to-day basis, thereby saving a lot of time and generating large productivity gains, the adoption rates are just fairly limited right now. The key step, of course, in automating tasks is that people have to start using it. And so until we’ve seen more significant uptake in the actual application of AI, in the regular work production process, I don’t think that we’re going to see as big of an impact on productivity.

That being said, the early signals of future productivity gains look very, very positive. Some of the academic literature and economic studies that have looked at the increase in productivity that we’ve seen following AI adoption, in a few specific cases, supports our view that large productivity gains are possible. The average increase in productivity is about 25%. Case studies of companies that have adopted AI imply similarly large efficiency gains. And so, you know, there’s a lot of reasons to be optimistic. It will just take a little bit more time to see these productivity gains realized.

Did you change your forecast for the coming decade based on the actual data?

No, it really hasn’t changed because our forecasts don’t assume any AI boost at all before 2027. And the very small increases in adoption that we’ve seen in the one year since we wrote our initial report, I think, are consistent with our view that over the next three years AI is probably not going to be a main driver of labor productivity and potential GDP growth. Even though we do still think that it’s going to be a significant driver of productivity and GDP growth over a much longer horizon.

What explains the divergence between the strong investment in generative AI that you’re seeing compared with the slow rates of adoption?

For AI to be deployed on a widespread basis, there’s a lot of things that need to happen. First you need to have models that are powerful enough and trained appropriately so they can actually be useful in everyday work product. Then you need to have the capability to facilitate and answer all the queries that people are going to be posing to AI models, when they do use them every day multiple times a day when they’re engaged in regular work. Having both these things requires a big increase in investment in semiconductors which in turn requires a big increase in investment in network capacity.

And ultimately, that’s going to require an increase in electricity and collective power investment to support the increase in demand that facilitating queries will require.

We are seeing clear signs that investment is increasing. Revenues of semiconductor manufacturers are up about 50% since early 2023. If you look at forecast revisions for AI hardware providers, they imply about a $250 billion increase since a year ago. And so there’s a lot of signs that the investment laying the groundwork for future use of AI is occurring.

The adoption and usage will occur when these pieces are in place, and companies actually start using AI on an everyday basis. For the most part that hasn’t happened yet. We see about 5% of companies reporting that they do use generative AI today in regular production, but this is a fairly small share relative to the overall number of companies that we think will ultimately benefit.

How is that 5% putting generative AI to use?

There are a few specific use cases that are emerging if you look across the industries that are using generative AI. First and foremost, we see adoption rates higher in areas like information services, finance and insurance. The motion picture and sound recording industry, for instance, is another area where adoption is far above the economy wide average.

If we’re talking about the things that people say that they’re using it for, marketing, automation, chatbot, speech text, and data analysis are all areas that stand out as ways that companies are applying AI right now. This is kind of the low hanging fruit where AI is most applicable, at least in its current form. Ultimately, we think that a broader set of tasks are going to be automated by generative AI. But that probably requires a build out of an application layer to support the broader automation we see possible.

What about the other 95%? What’s holding them back?

There are a number of factors that are slowing down the pace of AI adoption. A lot of executives can see the economic potential of generative AI, but even those that see benefits report a lack of knowledge about AI, concerns about privacy and security, and concerns about overinvesting in an early version of the technology as barriers. I think that a lot these reasons broadly reflect that companies want to make sure that they get generative AI right, and companies are therefore taking deliberate approach to AI adoption.

These views generally align with what we’ve seen in some of the business surveys, where CEOs are asked about their intention to use generative AI. Very few say that they expect it’s going to significantly impact their business over the next one to three years. Most say that they expect to see a significant impact over the three-to-10-year horizon.

What’s been the net impact on jobs? How do you expect that to change over time?

Given that we’ve seen very little adoption, it’s not surprising that we haven’t seen much of an impact on the labor market. If we look at things like the unemployment rate between occupations that are highly exposed to AI automation, and those that are less, they basically tracked each other one-for-one for the last year or two. There have been some layoff announcements attributed to generative AI, but for the most part it seems like a very, very small share – less than 20,000 of all layoffs generated in the economy, which comes down to less than 0.1% of total job separations. So AI hasn’t resulted in any significant job loss yet.

In fact, if we look at the labor demand that is generated it’s probably driven a net increase in employment. There has been a notable pickup in job postings mentioning AI as a desirable skill. This is especially true in the information technology sector. And so, it’s very well possible, and probably even likely, that the net impact on the labor market has been positive thus far. This is kind of in line with our expectations over the long run, where we do expect that generative AI won’t lead to a large amount of job loss. We generally think that it’s going to create opportunities either in AI adjacent sectors or occupations or in sectors where labor has a comparative advantage.

 

This article is being provided for educational purposes only. The information contained in this article does not constitute a recommendation from any Goldman Sachs entity to the recipient, and Goldman Sachs is not providing any financial, economic, legal, investment, accounting, or tax advice through this article or to its recipient. Neither Goldman Sachs nor any of its affiliates makes any representation or warranty, express or implied, as to the accuracy or completeness of the statements or any information contained in this article and any liability therefore (including in respect of direct, indirect, or consequential loss or damage) is expressly disclaimed.