Will AI save or sink the US economy?

The debate among experts over the potential costs or benefits of Artificial Intelligence (AI) in the US economy is not going to be resolved anytime soon. So I decided to ask one expert with a nearly inhuman reputation for objectivity. Since the answer will depend on AI's ability to improve labor productivity, and therefore economic growth, I asked the generative AI chatbot ChatGPT-4 itself: "How is AI going to improve labor productivity?"

GPT-4 listed several possibilities, including:

• Automation: AI-powered systems can automate repetitive tasks, freeing up human workers to focus on more complex and creative endeavors.
• Predictive Analysis: AI can help businesses make more informed decisions, optimize resource allocation, and enable personalized experiences for customers.
• Innovation and Creativity: AI can augment human creativity and innovation by assisting in idea generation, design optimization, and problem-solving.

It would seem that researchers and policymakers are counting on artificial intelligence tools to reverse the global productivity slowdown identified by many as the key element to foster higher future growth. But how confident can we be about that prospect? Not very, it turns out.

What do economists mean when they speak of labor productivity? Roughly speaking, productivity can be defined as either the amount of output each worker generates or as the amount of output per hour worked. Using the latter definition, and according to data compiled for the United States, the average quarterly (annualized) growth of labor productivity fell from 2.8 percent in the decade 2000.I-2009.IV to just 1.2 percent from 2010.I to 2019.IV. But measuring the extent of productivity gains associated with the generative chatbot's view is elusive. It's true that a substantial boost in productivity and technological progress involves innovation and creativity. But can AI augment human creativity by "assisting in idea generation," as ChatGPT-4 claims? Evidence from different fields of science suggests there may be something to the claim. But consider also that some of the most significant human discoveries in the 19th and 20th centuries came from the marriage of creativity, luck, and coincidence.

Antibiotics

The most serendipitous and one of the most important and life-saving innovations of the 20th century was the discovery of penicillin by Alexander Fleming in 1928. The bacteriologist was working with strains of Staphylococcus, a microorganism that causes several human diseases, including deadly sepsis, an infection of the blood. He noticed that one of his plates showed segments completely clear of bacteria—an invading fungus appeared to be growing in those areas.

Fleming isolated the fungus and discovered it produced a substance deadly for certain bacteria. The discovery marked the birth of antibiotics, without which human beings had perished for centuries in massive outbreaks of illnesses, such as the Black Death. Suddenly, masses of people were no longer dying or no longer having to be isolated during an outbreak of a bacterial infection. Moreover, antibiotics could be used to improve crop yields and other food production.

Today, AI is being used to find new antibiotics, but we are still unsure how these efforts will evolve. The boom era of antibiotic discovery, from the 1940s to the 1970s, resulted from human ingenuity alone. And, one of the biggest health challenges we now face is finding alternatives to antibiotics since these substances eventually lead to antimicrobial resistance (AMR).

The X-ray

Serendipity plays a vital role in revolutionary discoveries, but it takes a scientific mind to recognize something with great potential. The discovery of the X-ray by Wilhelm Roentgen in 1895 began with an accidental observation that subsequently required careful analysis. Alone in a lab, Roentgen used a vacuum tube to get charged electrons to pass through the air. His tube lit up a screen on the other side of the lab at high electrical charges, an unexpected result. During the next seven weeks, Roentgen studied the phenomenon intensely, coming to the conclusion that he had discovered a new type of radiation that would later be deployed in several fields, from security to medicine.

The microwave

Microwaves can be used to cook food, analyze weather patterns, and detect the intensity of hurricanes. They are also the background noise of the birth of the universe. Here is the story from NASA's website:

"In 1965, using long, L-band microwaves, Arno Penzias and Robert Wilson, scientists at Bell Labs, made an incredible discovery quite by accident: they detected background noise using a special low-noise antenna. The strange thing about the noise was that it was coming from every direction and did not seem to vary in intensity much at all. If this static were from something on our planet, such as radio transmissions from a nearby airport control tower, it would come only from one direction, not everywhere. The Bell Lab scientists soon realized that they had serendipitously discovered the cosmic microwave background radiation. This radiation, which fills the entire universe, is a clue to its beginning, known as the Big Bang."

What these innovations have in common is the combination of serendipity with the scientifically trained human mind: While scientists were intentionally looking for one thing, something else was found and taken seriously. Serendipity comes from observing and making connections between events that may have been previously regarded as disconnected. The discoveries had major consequences for technologically driven productivity and economic growth.

By contrast, large language models (LLMs), as exemplified in the responses generated by GPT-4 to the query about productivity, are not driven by serendipity because there is no observer to make informed connections. LLMs are complex probability models that are very good at interpolating and extrapolating words in a sentence without ascribing meaning to the linguistic forms that are strung together. Hence, LLMs will not innovate independently, though they might facilitate some innovations in the form of sophisticated internet search engines/deep learning algorithms—the latter are more elaborate data search tools.

Will this amount to a boom with enough punch to reverse the protracted productivity decline? It's not impossible, but it would require AI in its various forms to facilitate a discovery as far-reaching as penicillin or as mind-boggling as the microwave.