Eight years ago, when the first AI-powered investment funds launched, industry veterans dismissed them as expensive gimmicks. Today, artificial intelligence has quietly become the finance industry's most valuable employee, performing tasks such as extracting data from thousands of documents, analyzing markets at superhuman speed, and helping investment firms manage their finances more efficiently than ever before. What it still can't do is be Warren Buffett.
As the Oracle of Omaha turns 95 this month, he remains characteristically wary of the technology that has transformed investing. At last year's Berkshire Hathaway annual meeting, Buffett compared AI to nuclear weapons, saying it "scares the hell out of me." Yet that hasn't stopped him from investing heavily in companies like Apple and Amazon, which are dumping money into developing powerful AI systems that many worry (or hope) can replace human workers.
That includes replacing Buffett himself. The Intelligent Livermore ETF, which launched this fall, uses ChatGPT, Gemini, and Claude as its "investment committee," training the large language models on decades of legendary investors' writings and asking them to pick stocks in their style. The fund's prospectus promises to harness the strategies of the investment world's most illustrious minds, turning Warren Buffett's folksy shareholder letters into algorithmic investment decisions.
Doug Clinton, who runs Intelligent Alpha, the firm behind the ETF and other AI investment offerings, said his AI models can successfully replicate many aspects of Buffett's approach. They can screen for companies with low price-to-earnings ratios and high profit margins, just like the Oracle does. For Clinton, the goal isn't to unlock some secret sauce.
"We haven't thought about, 'Hey, where's the insight that's unique from human beings?'" Clinton said. "We're trying to see where can we, at scale, outperform the benchmarks." About 80% of his firm's 30 AI-driven strategies are beating their benchmarks, he said, by an average of 1000 basis points since inception.
Even with that record, Clinton's AI committee uses human oversight. His firm includes a final layer where a human approves the end portfolio, which he found people want as a fail-safe to ensure the AI isn't hallucinating. The challenge, Clinton acknowledged, is something he calls "taste" -- the ineffable quality that lets an investor look at 50 companies that meet all the quantitative criteria and instinctively know to pick just two of them.
"That's still the delta that has not yet been figured out," he said.
Private markets present a fundamentally different challenge for both humans and AI, according to Matt Malone, head of investment management at Opto Investments, a platform that helps wealth managers build private markets portfolios for their clients. Instead of making rapid trading decisions, investors make one crucial choice: which fund manager gets their money. Then they're along for the ride for years. Most private markets operate through drawdown funds, where investors commit money that gets deployed over time and returned over an even longer period.
"We have now multimodal AIs that can extract information from a deck fairly efficiently, much more efficiently than a human can," Malone said. "And frankly, a human doesn't want to be doing that because it's boring and repetitive." His firm uses AI to pull relevant data from those PDFs, standardize it, and create databases that let them compare fund managers across metrics like how they generate returns.
But Malone's team still makes the final calls on which managers to back. The AI excels at backward-looking analysis, crunching historical performance data, while humans focus on forward-looking decisions about inflation, interest rates, and sector outlook. Sometimes that means rejecting managers that AI finds have stellar past performance.
"We may say that [a fund manager] did great investing in hotels," Malone said, "but if we don't think hotels are a good investment going forward, we still don't want to invest with this manager right now."
Elliott's skepticism comes from working in a field that has been using sophisticated algorithms for decades. Quantitative funds have long employed complex mathematical models, and the recent advances in large language models haven't dramatically changed their core work. Instead, falling compute costs and incremental improvements in machine learning techniques have been more valuable than anything to do with ChatGPT, according to Elliott.
The result is an expensive AI arms race where everyone is chasing the same advantage. Elliott talks to managers who are now layering AI tools on top of their existing quantitative strategies, spending heavily on large language models and machine learning infrastructure. The question becomes whether all this AI spending creates a real edge or just raises the cost of doing business.
Even in systematic investing, human accountability remains crucial, Elliott said, which is why the next Warren Buffett is unlikely to be AI. Outside of high-frequency trading, investors still want to look someone in the eye and hold them responsible.
"For the super high frequency market, it's about the machine, " Elliott said. "For pretty much everyone else, they're looking at the person across the table, and saying, 'I don't even care how you make the decision, but it's you who is making the decision. If it doesn't work, you're on the hook for it.'"