🤖 Yep, ChatGPT and generative AI are going to have a big impact on jobs

Also: 5 Quick Questions for … analyst Brink Lindsey on the challenges of economic growth

“Any sufficiently advanced technology is indistinguishable from magic.” - Arthur C. Clarke

---

The Essay

Midjourney

🤖 Yep, ChatGPT and generative AI are going to have a big impact on jobs

If you’ve been wondering if generative AI — large language models with text-, code-, and image-generating capabilities such as ChatGPT and Midjourney — will take all the jobs of knowledge workers, you’re asking the wrong question. That’s not how economists think about the issue, as can be seen in the fascinating new paper “GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models” by Tyna Eloundou, Sam Manning, Pamela Mishkin, and Daniel Rock.

Two things to keep in mind: First, when looking at the impact of technology on jobs, economists think of jobs as bundles of tasks. Some of these tasks are more susceptible to automation than others. Second, when employing that task-based model, there’s a tension between technology that displaces workers from existing job tasks by automating them and lowering demand for workers, and technology that creates new tasks for human capabilities and thus reinstating demand for workers.

In the above paper, the researchers calculate the exposure of different jobs tasks — using “human annotators and GPT-4 itself as a classifier” — to these LLMs. (This is important: “We define exposure as a proxy for potential economic impact without distinguishing between labor-augmenting or labor-displacing effects.” Certainly distinguishing between those impacts would be the subject of future research.) And these are the headline findings:

Our findings indicate that approximately 80% of the U.S. workforce could have at least 10% of their work tasks affected by the introduction of GPTs, while around 19% of workers may see at least 50% of their tasks impacted. The influence spans all wage levels, with higher-income jobs potentially facing greater exposure. … [We] discover that roles heavily reliant on science and critical thinking skills show a negative correlation with exposure, while programming and writing skills are positively associated with LLM exposure. … We analyze exposure by industry and discover that information processing industries (4-digit NAICS) exhibit high exposure, while manufacturing, agriculture, and mining demonstrate lower exposure. The connection between productivity growth in the past decade and overall GPT exposure appears weak, suggesting a potential optimistic case that future productivity gains from LLMs may not exacerbate possible cost disease effects. … Our analysis indicates that the impacts of LLMs like GPT-4, are likely to be pervasive. While LLMs have consistently improved in capabilities over time, their growing economic effect is expected to persist and increase even if we halt the development of new capabilities today. … Our research serves to measure what is technically feasible now, but necessarily will miss the evolving impact potential of the LLMs over time.

“GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models”

So, LLMs won’t take all the jobs but they will have a significant impact on lots of them. And that analysis is based on current capabilities, which seem highly like to evolve. This doesn’t surprise me. My powerful first experience with ChatGPT made me so excited about its economic potential that I initially assumed “GPT” was an acronym for “general purpose technology” rather than “generative pre-trained transformer.”  That’s how important the technology seemed to me. In the world of economics, GPTs are the straws that stir the drink, tech advances with significant, long-term impacts across a wide range of business sectors. Classic examples of GPTs include the steam engine, electrification, and the computer.

“GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models”

So, yeah, GPTs are pretty important. In the 2022 paper “Additive Growth,” New York University economist Thomas Philippon finds that total factor productivity growth — that portion of productivity growth at broadly represents tech progress — tends to increase in regular increments, linearly rather than exponentially. The evidence, according to Philippon, is that “new ideas add to our stock of knowledge; they do not multiply it.” But GPTs are different. They cause breaks and upward shifts in the steady, incremental pace of progress that Philippon describes. In one of my Five Quick Questions chats last year, I asked Philippon if we underestimate the importance of GPTs? His response: “Perhaps a bit in the sense that they look even more striking in the linear model. But to be honest many people have been arguing that GPTs are important.”

Again, the exact scale and nature of the impact of generative AI models is unknown, both on the labor market and productivity growth. But at this point it might be easier to underestimate those impacts than overestimate them.

---

5QQ

5 Quick Questions for … analyst Brink Lindsey on the challenges of economic growth

Brink Lindsey is a vice president at the Niskanen Center and author of The Permanent Problem, a newsletter about the challenges of capitalist mass affluence, here on Substack. He has written a number of fantastic pieces that Faster, Please! subscribers will enjoy reading. Here are a few to get you started: “Technological Progress vs. Diminishing Returns”; “Is Dynamism Doomed?”; “The Anti-Promethean Backlash”; “Loss Aversion (by Any Other Name) and the Decline of Dynamism”; and “The Age of Stasis.”

1/ How does the "self-undermining dynamic" of economic growth lead to lower productivity?

The self-defeating characteristic of economic growth arises out of what Tyler Cowen has called the exhaustion of low-hanging fruit. Economic growth during the 20th century benefited from a number of one-off changes that took decades to unfold — in particular, the mobilization of women into paid employment and the steady upskilling of the work force through big investments in secondary and tertiary education. These growth boosters were basically played out by the end of the century, which means that productivity growth needs to surge to make up for the difference. Unfortunately, the low-hanging fruit problem arises here, too: There’s good evidence that scientific progress and technological innovation are getting harder as well.