It's a complete failure.
Stonks
Even if the whole thing is a bubble, there's no denying that the AI hype has been a huge success for the stock market.
But how good is AI itself actually at picking stocks? Two researchers, Gary N. Smith, a professor of economics at Pomona College, and Sam Wyatt, a student at Pomona College and project leader at Pomona Consulting, conducted an analysis to find out.
“AI-driven investments are particularly interesting because they offer a way to quantitatively evaluate the capabilities of a technology,” they argue in a Scientific American article.
And their findings, which have yet to be peer-reviewed, suggest that their capabilities are crude.
Smith and Wyatt looked at all public exchange-traded funds launched since October 2017 that relied partly or fully on AI systems to make stock investment decisions and found that the majority of them underperformed the S&P 500, an index of the 500 largest companies listed on U.S. stock exchanges.
Interestingly, more than half of the funds have since been closed.
An unlucky investment
For background, the S&P 500 is considered a benchmark for the health of the stock market. If the S&P 500 is doing well and your fund is doing poorly, you're missing out.
So if only 10 of the 43 funds that partially used AI systems to make stock investment decisions outperformed the S&P 500, it seems likely that the technology is seriously flawed. Overall, the average annual return of the partially AI-powered funds was 5 percentage points lower than the index's 12.4%.
But perhaps in these cases, human involvement limited the AI style. How would a fully AI fund without human intervention work?
To make matters worse, all 11 funds lagged the S&P 500, and six lost money in a generally booming market. “Overall, the 11 all-AI funds lost an average of 1.8% per year, while the S&P 500 earned investors an average of 7.6% per year,” the researchers wrote. (Recall that anyone can get low-risk equity exposure by investing in the S&P 500, so the benefits of AI-managed funds are currently quite slim.)
Smith and Wyatt argue that the technology's mistake is that AI can correlate data but not understand it.
“The Achilles heel of AI systems is that, although they are unparalleled at spotting statistical patterns, they have no way of determining whether the patterns they find are valid or meaningless,” the researchers wrote. “Until AI algorithms understand the meaning of words and how they relate to the real world, they will likely not be relied upon to make important decisions, including investments.”
More on AI: Experts say the AI bubble is about to burst
