Ken Griffin says everyone is misinterpreting the AI revolution — and wishes Zohran and Bernie would ‘read a damn history book for once’

Ken Griffin has told a version of this story before. At Stanford Business School in May, the Citadel founder admitted that he’d recently gone home one Friday “actually fairly depressed” after realizing what AI could now do — a stunning reversal for an executive who’d dismissed the technology as “garbage” at Davos just months earlier.

What he didn’t say then was exactly what triggered it. Speaking with Goldman Sachs’s Raj Mahajan at the firm’s Apex Symposium on June 2 — audio Fortune has exclusive, embargoed access to — Griffin filled in the blanks.

The dinner story, inverted

The hedge fund billionaire has previously talked about having dinner at a table with a full roster of CEOs. At the Milken Institute conference, he described asking the table to share how AI was transforming their businesses and getting “six or seven extraordinary stories” back.

He offered a new twist on the tale to Mahajan, saying he was with a number of major CEOs at a dinner “about two years ago,” and his guests were effusive as to how AI was transforming their business.

Griffin added that he basically didn’t believe it. “I couldn’t help myself. I’m like, ‘Let’s go around the table and share stories about how AI is transforming your business.'” When he got four or five “incredible stories” back about productivity gains, he told Mahajan, he deduced that “not one involved AI.” The distinction, he added, is that these gains involved “machine learning,” “optimization” and “digitization,” which are not quite the same thing.

The distinction matters because the terms get used interchangeably even though they describe different things. Machine learning, in Griffin’s framing, refers to systems trained to recognize patterns in data and get incrementally better at a narrow task — the kind of technology Citadel has used for about a decade, citing examples like reading radiological reports or powering self-driving cars. Those systems can produce real productivity gains without exhibiting anything like general reasoning: they’re pattern-matching at scale, not making judgment calls or synthesizing new conclusions the way a human analyst would.

What Griffin means by “AI” — at least the version that alarmed him — is something closer to agentic systems that can read a task, plan a multistep approach, execute it, and check their own work, the way a finance-paper project did when it read an academic study, reproduced its methodology, and verified the results independently. A company........

© Fortune