A few weeks ago I posted “AI Advances Aren’t Likely to Occur Nearly as Quickly as Many Believe,” a blog focused on a research paper, “The Simple Macroeconomics of AI,” by MIT professor Daron Acemoglu, — a co-recipient of the 2024 Nobel Memorial Prize in Economic Sciences. The blog referenced a NY Times article, “What if the A.I. Boosters Are Wrong?,” which discussed Acemoglu’s skeptical views that AI will supercharge economic productivity by the end of this decade, as well as “Gen AI: Too Much Spend, Too Little Benefit?, a report by Goldman Sachs Research senior strategist Allison Nathan, which included a number of insightful interviews, including one with Professor Acemoglu.
As I wrote in the blog, Nathan started her interview by asking why his paper argued that the upside to US productivity and GDP growth from GenAI will likely prove more limited that those of other predictions, including some from Goldman market analysts. “Why are you less optimistic on AI’s potential economic impacts?,” she asked him.
Acemoglu replied that the forecast differences are primarily about the timing of AI’s economic impacts rather than about the ultimate promise of the technology. “Generative AI has the potential to fundamentally change the process of scientific discovery, research and development, innovation, new product and material testing, etc. as well as create new products and platforms.” But the economic impact of historically transformative technologies like AI take time to play out.
I’d like to now discuss another recent NYT article, “Will A.I. Be a Bust?,” which was focused on the views of Jim Covello, Goldman Sachs’ head of stock research, one of Wall Street’s leading AI skeptics. “Three months ago, he jolted markets with a research paper that challenged whether businesses would see a sufficient return on what by some estimates could be $1 trillion in A.I. spending in the coming years,” said the article.
The article mentioned Covello’s personal experience with the history of overspending during major technology transitions. “He followed the bursting of the dot-com bubble as a semiconductor analyst and was scarred by watching colleagues lose their jobs. More recently, the Goldman veteran joined an internal team that has been evaluating A.I. services for the firm to use. He said the services he reviewed were costly, cumbersome and not ‘smart enough to make employees smarter.’”
Covello was also interviewed by Allison Nathan in her GS report, “Gen AI: Too Much Spend, Too Little Benefit?”
“Tech giants and beyond are set to spend over $1tn on AI capex in coming years, with so far little to show for it. So, will this large spend ever pay off?,” asked Nathan in the report’s introduction. According to Covello, she wrote, “to earn an adequate return on the ~$1tn estimated cost of developing and running AI technology, it must be able to solve complex problems, which, he says, it isn’t built to do.”
“He points out that truly life-changing inventions like the internet enabled low-cost solutions to disrupt high-cost solutions even in its infancy, unlike costly AI tech today. … He’s also doubtful that AI will boost the valuation of companies that use the tech, as any efficiency gains would likely be competed away, and the path to actually boosting revenues is unclear, in his view. And he questions whether models trained on historical data will ever be able to replicate humans’ most valuable capabilities.”
Nathan started the interview by asking: “You haven’t bought into the current generative AI enthusiasm nearly as much as many others. Why is that?”
Covello replied that to earn a reasonable return on investments (ROI), AI applications must solve highly complex and important enterprise problems given the capital investments required in specialized chips, large data centers, and electric utilities. The crucial question is: “What $1tn problem will AI solve? Replacing low-wage jobs with tremendously costly technology is basically the polar opposite of the prior technology transitions I’ve witnessed in my thirty years of closely following the tech industry.”
Even in its early years, the internet was a relatively low-cost technology solution for email and text communications, for accessing lots of content in websites, and for enabling e-commerce applications, a new way of shopping. Then came new kinds of applications, like Uber and Airbnb, which disrupted more expensive consumer services. While we can debate whether AI will deliver on the promise many people are excited about today, “the less debatable point is that AI technology is exceptionally expensive, and to justify those costs, the technology must be able to solve complex problems, which it isn’t designed to do.”
Nathan then asked whether we can expect the costs of AI infrastructure and applications to decline dramatically as the technology evolves and matures?
Covello replied that the performance and price/performance of computer technologies like microprocessors, memory, and storage improved exponentially over several decades thanks to Moore’s Law and other major technological advances, as well as a highly competitive IT marketplace. The internet has grown very rapidly over the past few decades because its key applications, — e.g., email communications, the World Wide Web, e-commerce, — were cheaper than the alternatives they replaced from day one, not ten years down the road. In addition, the starting costs of AI infrastructure are so high that those costs would have to decline dramatically to make automating tasks with AI affordable.
“Are you just concerned about the cost of AI technology, or are you also skeptical about its ultimate transformative potential?”
“I’m skeptical about both,” replied Covello. “Many people seem to believe that AI will be the most important technological invention of their lifetime, but I don’t agree given the extent to which the internet, cell phones, and laptops have fundamentally transformed our daily lives, enabling us to do things never before possible, like make calls, compute and shop from anywhere.”
While AI has shown promise to make existing processes more efficient, — e.g., code generation, sales and marketing, customer and employee self-service, — unlike the internet, the costs of utilizing AI technology for these tasks is significantly higher than existing methods. “For example, we’ve found that AI can update historical data in our company models more quickly than doing so manually, but at six times the cost.”
“More broadly, people generally substantially overestimate what the technology is capable of today. … This is not a matter of just some tweaks being required here and there; despite its expensive price tag, the technology is nowhere near where it needs to be in order to be useful for even such basic tasks. And I struggle to believe that the technology will ever achieve the cognitive reasoning required to substantially augment or replace human interactions. Humans add the most value to complex tasks by identifying and understanding outliers and nuance in a way that it is difficult to imagine a model trained on historical data would ever be able to do.”
“If your skepticism ultimately proves correct, AI’s fundamental story would fall apart. What would that look like?
“Over-building things the world doesn’t have use for, or is not ready for, typically ends badly. The NASDAQ declined around 70% between the highs of the dot-com boom and the founding of Uber.
“That said, one of the most important lessons I've learned over the past three decades is that bubbles can take a long time to burst. That’s why I recommend remaining invested in AI infrastructure providers. If my skeptical view proves incorrect, these companies will continue to benefit. But even if I’m right, at least they will have generated substantial revenue.”
“So, what should investors watch for signs that a burst may be approaching?,” asked Nathan in conclusion?
“How long investors will remain satisfied with the mantra that ‘if you build it, they will come’ remains an open question,” said Covello. “The more time that passes without significant AI applications, the more challenging the AI story will become.” But he added that the more important area to watch is corporate profitability, because as long as corporate profits remain robust AI experimentation will keep running. “That said, spending on these experiments will likely be the one of the first things to go if and when corporate profitability starts to decline.”
So, who is right, the bullish of the skeptic AI camp? In the end, it’s too early to tell.
Acemoglu and Covello are clearly in the skeptic camp, at least in the near term, “with Acemoglu seeing only limited US economic upside from AI over the next decade and Covello arguing that the technology isn’t designed to solve the complex problems that would justify the costs, which may not decline as many expect.”
But Nathan’s report also includes the opinion of GS executives who are significantly more bullish about the near term economic potential of GenAI. Let me conclude by summarizing the views of one such executive, GS senior global economist Joseph Briggs, who argues that GenAI will lead to significant economic upside over the next decade.
“He estimates that, over the next decade, GenAI will raise US productivity by 9%, GDP growth by 6.1%, and will ultimately automate 25% of all work tasks while creating new tasks and occupations,” wrote Nathan. “While Briggs acknowledges that automating many AI-exposed tasks isn’t cost-effective today, he argues that the large potential for cost savings and likelihood that costs will decline over the long run — as is often, if not always, the case with new technologies — should eventually lead to more AI automation.”
Comments