Artificial intelligence may not be as intelligent as it's made out to be, according to Goldman Sachs's Jim Covello.
The bank's global head of equity research, Covello is a technology industry veteran who has witnessed 30 years of transformative shifts such as the invention of the internet, the e-commerce revolution, the adoption of smartphones, and much more.
And to him, AI doesn't exhibit the signs of a truly groundbreaking innovation like the ones he's seen before.
In an interview with Business Insider, Covello explained the two reasons he's an AI skeptic — and how to invest when the rest of the market is buying into the AI hype.
It's no secret that AI technology isn't cheap. Goldman Sachs estimates that data centers, utilities, and other infrastructure buildout costs will accumulate to over a trillion dollars in the next few years.
To Covello, the high cost of AI should raise some concerns.
Most noticeably, it's created an anticompetitive, winner-take-all market where only the largest mega-caps are able to invest in and develop the technology. Investors who have been following the meteoric rise of Nvidia are likely aware of how the chipmaker and other members of the Magnificent Seven have come to dominate the returns of the S&P 500.
AI buildout comes with intensive capital requirements that only the biggest companies can afford. Covello doesn't see any company that could come remotely close to challenging Nvidia's GPU production ability. He also points to the Netherlands-based ASML, which holds a monopoly on the lithography tools needed to produce chips.
Without competition, future costs of AI are likely to remain high. High barriers to entry mean that the technology landscape surrounding AI is unlikely to see new players anytime soon, giving the existing winner little incentive to lower prices. This contrasts dramatically with the environment that led to flourishing internet or e-commerce businesses, which Covello saw as much more democratized.
"I think there's probably too much complacency that the costs are just going to come down dramatically over time when you have some companies that are so incredibly well entrenched with their technology position," Covello said.
Covello also points out that AI use cases are replacing low-cost jobs with very costly technology — the opposite of other developments he's seen. "Most technology transitions in history have been a less-expensive solution replacing a costly solution," Covello said. He points to the e-commerce revolution, where Amazon was able to rapidly take market share away from traditional brick-and-mortar retailers due to lower storefront costs.
Covello sees call centers, a popular AI use case, as a prime example of a high-cost AI solution replacing a low-cost one. SoftBank and CVS are just two examples of companies developing an AI-enabled call center solution in an initiative to improve customer service experiences.
"A lot of people say, 'Well, we'll be able to replace call centers with AI,'" Covello said. "Even if you were to believe that, that's a very expensive technology where the world's going to spend a lot of money to build out AI capabilities over the next couple of years alone, to potentially replace relatively lower cost workers."
If AI were truly transformative, then the trillion-dollar price tag itself wouldn't be such an issue. Generative AI is touted as something that could automate a quarter of all jobs, dramatically boost GDP, and function as some sort of superintelligence.
Covello doesn't think AI technology is as groundbreaking as it's hyped up to be. In his experience covering the technology industry, AI's influence pales compared to other actually life-changing innovations.
"I covered semiconductor stocks when the internet was invented, when the cellphone or smartphone was invented, when the laptop came out," Covello said. "Those things fundamentally changed the way that we lived our life. They allowed us to do things that we've never done before."
The potential use cases of these technologies were apparent from their inception, but Covello sees no comparable roadmap for applications of AI today. At best, Covello sees AI as a tool that can make existing tasks easier or more efficient. And at worst, it's a hallucinating virtual assistant (Remember pizza glue?).
While some technologists might believe that AI's hallucinations are a kink to be ironed out as the technology improves, Covello views hallucinations as a feature, not a bug, of AI.
AI is skilled at providing answers to familiar questions that it's been trained on. However, once it encounters unfamiliar situations outside its training set, AI doesn't have the ability to reason like humans do. In these situations, it makes up inaccurate information or predictions. Covello likens it to how you might make up words or hum along when you don't know the lyrics to a song.
One example is self-driving cars, which have garnered significant attention in the AI wave. However, the technology has repeatedly failed to perform when the cars come across an unfamiliar situation or outlier. When two Waymo vehicles encountered an improperly towed pickup truck on the road last December, their software malfunctioned and led to a collision.
At the end of the day, the technology isn't truly autonomous or trustworthy — it needs constant human supervision. Pizza glue might be a relatively innocuous mistake, but a self-driving Tesla crashing into a jet definitely isn't.
To those who argue that AI will become smarter through further training, Covello points out that these models have already consumed copious amounts of data and are quite far along in their training. In fact, Big Tech is actually rapidly running out of data to train it on.
Technology that is highly skilled at analyzing historical data but unable to apply it to new situations won't dramatically change our everyday lives or replace any high-value jobs. If AI doesn't actually have the ability to conduct higher-order reasoning, it won't leave a trillion-dollar impact on the economy.
In Covello's opinion, "there's a fundamental misconception of what AI is and what AI does."
Being bearish on AI isn't a popular stance to take, perhaps partially because it means betting against the monster rally of the S&P 500 this year.
Covello has some good news for those who've bought into the AI trade or have benefited from the Big Tech buzz: in his experience, bubbles can take a long time to burst.
For now, the best bet is still to invest in the picks and shovels that are building out AI infrastructure. The winners of the AI boom so far have been high-quality, well-capitalized companies with strong fundamentals. If the AI skeptics are proven wrong, then Big Tech and other chipmakers, data centers, and utilities providers will continue to perform exceptionally well. And if AI falls flat of expectations, then these AI infrastructure companies will have the financial capability to adapt.
Covello also recommends monitoring corporate profitability, especially as companies report their second-quarter results. Corporate profitability can reveal the extent to which AI technologies are actually monetized and predict future AI spending.
"There are going to be a lot of companies that are reporting increased AI expenses that are going to hurt the P and L, but there's no revenue yet," Covello said. If these costly capital expenditures don't lead to increased revenue in the next 12 - 18 months, Covello anticipates that investors will begin raising eyebrows at the efficacy of AI.
When corporate profitability is robust, companies have more money to experiment on riskier projects such as AI. Covello doesn't think companies will pull back on AI spending until their bottom lines start to shrink and the economy enters a downturn. Given that the Fed is on track to guide the economy to a soft landing, this situation is highly unlikely.
For now, strong earnings mean the AI bubble will likely continue growing for a while longer.
"We're going to just keep monitoring the data as it comes in," Covello said. "If the facts change, then we will change our view. If there are some killer applications that come in over the next couple of years, then obviously AI is going to be a lot more valuable than I currently believe."
But for now, Covello remains a skeptic about truly transformational AI use cases: "I haven't seen any yet, and I don't anticipate there's going to be any."