Why AI May Not Lead to GDP Growth

This article is written by Denis Kalyshkin, the US VC with 12+ year experience. Please subscribe to him on LinkedIn and contact via dk@askvc.org.

When people talk about AI, they usually argue that it will increase labor productivity, which will ultimately lead to GDP growth. And indeed, higher labor productivity is necessary for GDP growth. What AI proponents won't tell you, however, is that this won't happen automatically. It is painstaking work. Moreover, under certain circumstances, GDP could actually fall by tens of percent.

Today, I want to show you several cases in which AI may fail to lead to higher productivity and GDP growth:

1. AI is used the wrong way. There is little point in using a productivity-enhancing tool on a part of the process that is not the bottleneck, because the productivity loss will simply occur elsewhere. Eliyahu Goldratt illustrated this very clearly in his book The Goal, using the example of a factory that bought an expensive robot but installed it in the wrong part of the production chain. The robot performed its job extremely quickly, but a bottleneck developed further down the line. As a result, the robot sat idle 99% of the time. There is automation, but productivity does not increase. This is probably the simplest part of the equation to solve. By the way, something similar had already happened in the 1970s–1980s. Read about the productivity paradox. The technology was there, but productivity did not increase.

   
   

2. Customers don't have the money to buy your product — or a crisis of overproduction. Technologists often forget that creating a great product is only part of the job. You also need people to buy it. And if your customers don't have the money, then even if they desperately need your product, they won't buy it. If large numbers of people lose their jobs and cannot find new ones for 6–12 months, naturally they have less money. Consequently, they buy less. Now imagine that these people never find jobs again. They will never buy anything from you. You may have heard about crises of overproduction from history. For example, during the Great Depression, U.S. GDP fell by 26% between 1929 and 1933, while unemployment peaked at 25%.

   
   

3. Customers have money, but they save it. In fact, optimism and confidence about the future are among the most important factors for GDP growth. If people know that they will have an income tomorrow, they are more willing to spend money and take out loans. But if they are frightened, they save money for the future and cut back on non-essential spending. This is precisely why central banks around the world make every effort to make their policies predictable and to give markets confidence. In this regard, I think LLM vendors are currently doing a very poor job. Forecasts showing terrifying levels of unemployment may help sell products by playing on fear. But they also undermine people's confidence in the future. This is one of the reasons I don't like hype, although I understand how it works.

   
   

I want to end this post with the words of King Solomon, spoken several thousand years ago:

“What has been will be again, what has been done will be done again; there is nothing new under the sun.”

When applied to AI, we often hear that we are at a point in history that has never existed before. I want to remind you that by 1820 around 70% of the US population was engaged in agriculture rather than intellectual work. Today, only about 27% of the global population works in agriculture (and only 1% in the US), while the global population has grown from 1.1 billion to 8.2 billion people. Admittedly, humanity had to endure two devastating world wars along the way. I hope that this time we will be able to learn from the mistakes of the past.