7 Reasons Why AI Isn’t As Game-changing As You Fear

Artificial intelligence is changing some things and challenging our ideas about others.

For instance, the number of internet searches carried out through AI services is rising at a rapid clip: we could soon have an internet used by humans only indirectly (through AI). Numerous university professors are returning to strict in-class assessment methods to keep students from submitting work generated with ChatGPT, and several writers are asking what their job will be (or whether it will even survive) when it seems ChatGPT could write a novel in a few hours.

However, as much potential as AI has, it won’t change everything overnight. Let’s consider its impact on economic growth. The standard economics textbook says ideas make the economy grow. And ideas depend on the amount of resources invested in research and development (R&D). Put the other way around: more research and development generates more ideas, and more ideas generate faster gross domestic product (GDP) growth.

More resources devoted to R&D includes all efforts aimed at generating new ideas. Investing in the creation of a new medicine is R&D, but so is walking commercial districts to analyze what new business might succeed in the area. In short, R&D is thinking about how to make better use of what we already have.

If we want to increase economic growth, we have to increase the resources invested in R&D. Suppose we want the rate of idea growth to increase tenfold. We would need to multiply the resources devoted to R&D by ten. Before AI, only humans could do the work of creating new ideas, so we would have to multiply by ten the number of people dedicated to R&D. In the US there are 1.5 million people working in R&D. We would have to raise that figure to 15 million people. Switching 13.5 million people into new jobs sounds almost impossible, because in practice it is.

AI offers an alternative. Instead of needing 13.5 million new R&D employees, I can put 13.5 million new ChatGPT agents to work. That is plausible. Millions of AI agents increase the rate of idea growth; the faster growth of ideas boosts productivity; and that faster productivity growth in turn generates economic growth and therefore the capacity to improve AI, so we can have more and better AI agents, in a chain that produces explosive economic growth.

Now, for the nuance. 

AI requires physical infrastructure. The graphics processing units (GPUs) AI depends on consume a lot of electricity and need physical space. You can’t add millions of AI agents with a single click. Even if we could, at least for now, we need humans interacting with AI to pose the relevant questions and extract the answers.

Duplication across companies slows down innovationEvery pharmaceutical company will want to invent a medicine that cures cancer. Since profits are at stake, they won’t share their results with competitors. This means many resources will be wasted. Duplication doesn’t eliminate idea creation, but it does slow it down.

Starting in 1971, chip makers increased their R&D investment seventy-eightfold, but productivity barely doubled. If that’s true across sectors, it implies that every time we reach a higher productivity level, it becomes harder to climb to the next one. Maybe in the first year of using 13.5 million ChatGPT agents ideas would multiply thirtyfold (an arbitrary figure), but that level would decline in subsequent years. Innovation is increasingly complicated.

The rate of productivity growth is inversely proportional to the level of productivity. What does that mean? Think of two taquerías (taco shops). Taquería C makes 100 tacos a day and taquería H makes 2,000 tacos a day. If both implement a technology that lets them produce 50 more tacos. Taquería C now makes 150 a day and H makes 2,050. So, the same improvement raised C’s productivity by 50 percent but H’s by only 2.5 percent. Think about society. The leap from the time when there were only landlines at home to the time when everyone had a cell phone was huge; the leap from a brick-style cell phone to a smartphone with apps was big, but not as big as the previous one, because the core function (being able to reach someone wherever they are) stayed the same. 

Ideas only lead to change if somebody wants to use itAnother wrinkle is behavioral. Knowing what to do and doing it are different things. For new ideas to turn into higher productivity, they have to be implemented. If I have the formula for a new cancer drug but never produce the medicine, then the idea generates no improvement for society. It may sound redundant, but the simplest ideas are often crucial. Think of all the people who know that to have a better life they should exercise daily. They know it, and yet they don’t do it.

Implementation is hardImplementing new ideas requires change. Humans don’t get along well with what’s new. History shows this: many people opposed photography, the bicycle, and even reading. In many countries, like Mexico, we still see many people who want Uber to disappear. Another case is Zoom. Before the COVID pandemic we already had the tools to work remotely, but the practice wasn’t widespread because we didn’t want it. That is, the idea already existed, but not its implementation. The lockdown forced us to implement it. Having thousands of new ideas won’t make us use them.

Economists think of a productivity increase as “producing more with the same inputs.” But “producing the same with fewer inputs” is also a productivity increase. It means we chose to take the productivity gain as time rather than as output. If we decide to use new ideas to keep producing the same amount, but with less effort (if the factory owner or the writer or the lab director simply works fewer hours for the same result), the quantity of goods and services in the economy would stay the same and, by definition, there would be no economic growth).

We might ask why anyone would choose to keep producing the same amount with fewer inputs instead of holding inputs constant and producing more. Because preferences, not only possibilities, guide behavior. A baker might prefer to spend more time with friends rather than sell more bread. Another reason is satiation. If everyone already has the best phone available, why produce more of that same model? This choice, while possible, is unlikely. We tend to want more, and humanity is nowhere near satisfying the needs of all its members, so it’s hard to argue that further production is unnecessary.

Which brings us to humility. As F. A. Hayek wrote, “The curious task of economics is to demonstrate to men how little they really know about what they imagine they can design.” That is a useful antidote to AI triumphalism. It would be a shame to waste good ideas, and AI will surely help us find many. It will improve particular processes, open narrow but valuable avenues of discovery, and make some tasks easier or cheaper. But it is far from generating a civilizational revolution.