The AI Situation Could be Helpful for Productivity

We have jumped into the fourth quarter of 2023—time has flown—and we just had another exciting episode on Behind the Markets. Professor Daniel Rock, an Assistant Professor of Operations, Information, and Decisions at the Wharton School of the University of Pennsylvania is a returning guest to the show, and he brought with him Anton Korinek, a Professor at the University of Virginia in the Department of Economics.

The topic was Artificial Intelligence (AI), viewed more through an economic than a technical lens.

The AI Situation Could be Helpful for Productivity, says Professor Jeremy Siegel

One of the interesting points that occurred at the start of the episode was that Professor Jeremy Siegel, WisdomTree’s Senior Economist, made a brief point about AI from a productivity perspective, and I’d note that my ears perked up because Professor Siegel has been discussing the ‘productivity paradox’—meaning that the economic environment has had a lot of positives to it in recent years even if productivity growth has been nonexistent—and said that the AI situation could be helpful. Notably, he was thinking that maybe productivity growth could possibly hit 1.5% or even 2.0%.

From current levels and thinking in terms of the relevant scale of this data point, that would be a massively positive economic development if it occurred.

The Economy has Surprised more to Strength than Weakness so Far.

The discussion began with both Daniel and Anton talking about their views of the strength of the U.S. economy. Daniel’s comments focused on the 10-Year—he thinks that a yield of 6% could be possibly, based on what he is seeing. Anton was noting a positive supply situation in the economic picture, and that part of the strength we have been seeing could very well be continuing recovery from Covid.

Neither is expecting to see data indicative of recession in the near term, especially not a severe recession.

Productivity: A Bit of Historical Context, Bringing us into the Present Day

Many discussions regarding productivity cite a variety of historical periods. It’s important to understand, for example, that those measuring productivity growth will see that the period from the end of World War Two through the 1970’s was particularly strong. From the 1970’s forward to the present day, the data largely trended lower, even if there were a few upward moves in the 1990’s.

It was thought that the upward move in productivity in the 1990’s was less about the Internet or ‘.com’ craze and more about the general computerization of the economy. A tricky part of any discussion of a new technology, be it the computer, the Internet and now AI is the dissemination.

If one thinks about electricity, as an example, we would all likely agree that it was a big positive for the global economy, but the usage was not simply based on it being released of invented. An entire infrastructure had to be built, and people had to shift their habits and learn how to extract the maximum benefits. Building infrastructure, shifting habits, gaining skills—these are not things that happen instantly.

Right now, with AI, we are in a position where interesting new technologies have just come out and we stand on the precipice about to see a lot more. During this discussion, it was estimated that we are a few weeks away from seeing Google DeepMind’s new model, called Gemini, and we’ll have to then evaluate whether it is simply slightly better than GPT-4 or if it is bringing entirely new capabilities to the table. Even if we couldn’t cite exactly what Gemini would bring, we do know that computing power is doubling roughly ever six months. There are many people who did not realize initially how strong these large language models could be with their capabilities, but we can note that if something is doubling at that type of a rate it becomes very difficult to predict exactly what will occur, especially over a longer-term period.

Productivity Expectations?

While we already heard from Professor Siegel earlier in the episode, it was interesting to hear both Anton and Daniel approach this question.

Daniel noted that, if we did get from current, very low levels to 1.5%, this would be a fantastic outcome.

Anton was more optimistic; in that he believes that it is possible to go above 2.0%. Now, it was interesting to hear him discuss aspects of a ‘measurement challenge.’ He noted that he believes the new AI tools—the large language models—make him roughly 20% more productive. We can accept his word at face value, but this 20% would not be automatically reflected in the statistics just because he says it—he would need to be paid commensurately, 20% above his current level.

One thing that was notable was how Daniel and Anton stratified the general labor force, thinking that there would be some members at the upper end of the capability spectrum and some at the lower end, possibly due to being new. The gain from using these tools was bigger for those on the lower end of the capability spectrum—which is great in a way in helping those employees increase their skills—but then those gains may not be immediately measurable in the broader statistics, even if accurate.

A Thought for the Future…

AI is, in our opinion, the most exciting megatrend because it has the chance to intersect with and magnify other megatrends to increase their impact. Anton and Daniel talked a bit about robotics towards the end of the discussion. The focus was on the concept of taking future versions of the large language models that we are seeing and using those to power, in a sense, different kinds of more general-purpose robots. If a robot could be leased to do a particular job, finish, and then be leased by another client to do another job, this would increase efficiency and make the use of robots much more economical. Businesses would not need to fully buy and then depreciate these expensive capital assets and robotics firms would not need to make a different robot for every use case.

That’s not to mention that getting the robots up to speed, meaning doing the actual job needed, faster, could also be a productivity enhancer.

It was an excellent episode, available here or below. We look forward to continuing to track and examine the progress on this topic as it continues to evolve.