Only about a quarter of tasks that could use artificial intelligence are likely to be cost effective in the next decade, according to Daron Acemoglu, Institute Professor at MIT.
Speaking on the Goldman Sachs Exchanges podcast, Acemoglu says even with big breakthroughs in AI, the impact won't be seen for several years.
That would mean AI will impact less than 5% of all tasks and boost U.S. productivity by only 0.5% and GDP growth by 0.9% cumulatively over the next decade, according to Goldman.
"The current architecture of the large language models has proven to be more impressive than many people would have predicted, but I think it still takes a big leap of faith to say that just on this architecture of predicting next word, we're going to get something that's as smart as, you know, Hal in 2001: A Space Odyssey," he said.
There "could be very severe limits on where we can go with the current (LLM) architecture," Acemoglu said.
He is also skeptical that AI can get to where it needs to be more quickly by simply throwing more GPU capacity at it.
Higher and higher quality data, rather than capacity, will be needed and it's not clear where that data is coming from, he added.
To get a return on the $1 trillion companies like Nvidia (NASDAQ:NVDA), Microsoft (MSFT), Alphabet (GOOG) (GOOGL), Meta (META), Amazon (AMZN) and Super Microcomputer (NASDAQ:SMCI) are expected to spend on AI capex in the next few years, AI wll have to solve complex problems, Jim Covello, Goldman Sachs' Head of Global Equity Research, said on the episode.
“We're a couple of years into this, and there's not a single thing that this is being used for that's cost-effective at this point,” he said. “I think there's an unbelievable misunderstanding of what the technology can do today. The problems that it can solve aren't big problems. There is no cognitive reasoning in this.”