Massive Bets On AI Suggest Early Optimism Is Smarter Than Premature Dismissal

There is a lot of grousing about why AI is stubbornly refusing to show up in the economic statistics, let alone corporate profits, given the gargantuan size of the investments required. 

But cooler - and perhaps better informed - heads are reminding that it may still be too early to seen productivity and other financial enhancements - just as it took a while for computers and the internet to kick in (to say nothing of electricity, the telephone and automobiles, as well as other transformative technologies). Which is not to say that the concerns are not without merit, but that the returns to early optimism may outweigh those to early pessimism. And that may justify the risky investments now being made. JL

Megan McArdle reports in the Washington Post:

If AI is so revolutionary, why isn’t it more visible in the statistics? Despite claims that AI is displacing workers, economic evidence is dubious. Capital expenditures on AI are so huge they contributed more to GDP growth this year than consumer spending. If you think AI will be transformative, the eye-popping compensation and hog-wild investment are a rational, if risky, gamble on unlimited upside. We might not be seeing immediate productivity improvements because humans and machines are at the early end of the learning curve. AI is still too young to create the secondary and tertiary productivity gains we saw from computers and the internet. This is why so many people are willing to bet on AI. When a technology is truly transformational, it’s safer to bet on premature optimism than to prematurely dismiss it.

A famous economist once remarked: “You can see the computer age everywhere but in the productivity statistics.”

That epigram, issued by Robert Solow in 1987, became the subject of a lot of debate among economists in the 1990s. You don’t hear those arguments so much anymore, because it’s clear computers have transformed American work.

 

A decade later, another famous economist made a similar observation about the internet — actually, a prediction: “By 2005 or so, it will become clear that the Internet’s impact on the economy has been no greater than the fax machine’s.” That was Nobel Prize winner Paul Krugman, who now has a highly profitable email newsletter. We’re now hearing similar questions about artificial intelligence.

 

If it’s so revolutionary, why don’t we see evidence that it’s replacing workers? Why hasn’t it transformed productivity statistics? How can these companies earn back the massive amounts they’re spending building data centers and training models? Isn’t all the doomcasting and wishcasting a load of hype? Aren’t we in an AI bubble that’s getting ready to pop? 

 

Capital expenditures on AI infrastructure are so huge that Renaissance Macro Research says they contributed more to gross domestic product growth this year than consumer spending. If you think AI will be transformative, and especially if you think it’s possible to get to superintelligence, the eye-popping compensation packages and hog-wild investment are arguably a rational, if risky, gamble on potentially unlimited upside. Risky investments, definitionally, sometimes don’t work out. But as Silicon Valley has repeatedly discovered, it’s very nice when they do. The dot-com bubble might have produced spectacular failures, such as Pets.com. But that era also gave us Amazon and Google.

 

But then we’re back to the original question: If AI is so revolutionary, why isn’t it more visible in the statistics? Despite the oft-repeated claims that AI is displacing entry-level workers, macroeconomic evidence of this phenomenon is dubious.

Here are a few potential solutions to this mystery.

One possibility is that AI is raising productivity, but many workers are using their extra time for leisure, rather than more work — a kind of “dark leisure” that doesn’t show up in statistical data but still represents workers doing their jobs faster. That was true of the internet. In my early blogging days, when I tracked my web traffic closely, I noticed that visits spiked as the workday started, fell off during lunch, rose again in the afternoon and plummeted when the evening commute started. I concluded that whatever my high-minded ideals, the actual product I was supplying to readers was a computer activity that looked sufficiently like work to fool inattentive managers.

I’d bet something similar is happening with AI. A friend who is a lawyer, having read some of my AI columns, recently reached out to tell me that he’d finally used the technology for work. He’d asked a chatbot to draft a document, and though the draft needed work, he estimated it had saved him two to four hours of typing. I asked him what he did with the extra time. He pleaded the fifth.  Over time, however, expectations for productivity will change. Managers now understand how long a piece of work should take when you’re plugging a search into Google rather than hunting down physical references. They’ll develop a similar understanding as AI takes over writing many emails and daily reports, and demand commensurate increases in employee performance. Some of that “dark leisure” will go away and show up in higher productivity figures.

Another reason we might not be seeing immediate, large productivity improvements is that humans and machines are at the early end of the learning curve. When I was in college, I worked as a clerk at a firm that employed two top-notch legal secretaries. Those women, both in their early 50s, had started their careers on typewriters and could type more than 100 words per minute with few mistakes. They had also acquired an effortless mastery of legal forms that allowed them to quickly translate a lawyer’s handwritten notes into a flawless brief or filing. By the early 1990s, they were both using word processors, but how much time was that actually saving them? Some definitely, but the real productivity gains came later, when lawyers could type documents directly into a computer, much faster than they could dictate or write them out by hand, and firms didn’t need to employ so many legal secretaries.Today, I use ChatGPT to kick-start much of the research I do for columns. But I have been doing this for a while, and my Google-Fu was already pretty well honed, so it’s not that much of a time-saver. I also have a pretty deep familiarity with the subjects I cover, so I usually know where to find facts or explanations, and I have a long list of experts I can quickly call or email if I have a question. Chatbots are making me a bit more productive (note to my editors: I am using that time to do more work), but the real productivity boost to my profession will come from enabling people with different skills to do new kinds of work, not slightly boosting the productivity of those who are adept with the old methods.

One further speculative possibility, intriguing but difficult to assess: When AI starts displacing workers, it may do so not just at the entry level but outside the United States, as companies insource outsourced functions to machines.

Finally, this technology is still too young to create the secondary and tertiary productivity gains we saw from computers and the internet. Young people won’t remember how transformative spreadsheets were, but the ability to easily record and analyze data on the desktop didn’t just streamline a lot of accounting jobs — it gave managers and investors tools to hunt for productivity gains throughout other departments and the rest of the economy. I can’t tell you what kinds of similar opportunities AI might create, but I certainly expect there to be some.

This is ultimately why so many people are willing to bet on AI. Their excitement may be premature, but it’s not unfounded. And when a technology is truly transformational, it’s probably safer to bet on premature optimism than to prematurely dismiss it out of hand.