A Goldman Sachs partner responsible for a flagship artificial intelligence project warned that the rapid adoption of AI on Wall Street could weaken the thinking skills of the next generation of financial professionals.
“There’s a huge risk here: In the age of AI, we outsource our reasoning capabilities to these models, ultimately leading to cognitive atrophy to the point where we can no longer think from first principles ourselves,” said Chris Churchman, who leads Marquee, Goldman Sachs’ digital platform for institutional clients.
According to an interview transcript obtained by CNBC, Churchman made the remarks on the latest episode of Goldman Sachs’ Exchanges podcast.
Churchman said that just as modern technology has gradually caused people to lose their navigation and memory skills, bankers could also gradually lose their analytical abilities if algorithms take on all the heavy analytical work.
“Reasoning still matters,” he said. “You still need to think through a problem and organize it into a coherent argument, and right now we’re handing that reasoning over to AI.”
Wall Street is pushing to deeply integrate AI into every aspect of trading and banking, which could be a “devil’s bargain”: In the short term, AI can boost profitability across the industry, but in the long term, it could erode the professional talent base on which the industry depends.
As AI takes over more and more of the routine work previously done by junior bankers and traders, young employees would have used that work to learn how to think and make decisions. Now, if those training opportunities are replaced by AI, financial institutions may sacrifice a professional culture that develops junior employees into experienced Wall Street professionals.
This could even reduce financial institutions’ need for junior bankers from the outset. CNBC reported last year that Wall Street firms were studying how to use AI to reduce the ratio of junior bankers to senior employees.
Churchman said banks need to strike a balance between using AI and protecting Wall Street’s traditional “apprenticeship” training system. Churchman previously ran UBS’s foreign exchange trading business and joined Goldman Sachs in 2021.
“You need to learn through practice, and a lot of knowledge is actually tacit and has never been written down,” he said.
Churchman said Goldman Sachs needs to “make sure we don’t lose some of the tacit and intuitive knowledge held by some of our best people today, while ensuring that the next generation can develop those capabilities as well.”
For example, junior traders can handle client requests for quotes under the supervision of experienced traders and learn how to make trading judgments through the process.
“We could certainly automate this,” he said, “but if we do, can we still develop senior traders who truly understand the business?”
Churchman said that when decisions involve high risks and substantial uncertainty, systems must ensure that employees retain final decision-making authority rather than becoming passive operators of AI systems.
Churchman also serves as co-chair of Goldman Sachs’ global banking and markets AI working group. He said that even Goldman Sachs, one of the world’s leading investment banks, has not yet “figured out” how to manage the transformation that has already begun.
During the podcast interview, Churchman also shared Goldman Sachs’ experience bringing AI to the Marquee platform. Marquee serves hedge funds and other institutional clients, providing access to Goldman Sachs’ market data, research reports, risk analysis, and trade execution services.
Churchman said Marquee’s AI platform is currently available only to Goldman Sachs employees internally.
From a technical standpoint, the biggest challenge is ensuring that AI-generated answers are 100% accurate and auditable. He noted that consumer-facing AI chatbots typically warn users that their answers may contain errors, but in high-end finance, tolerance for error is extremely low.
Churchman revealed that while developing the client-facing AI platform, the software even offered a surprising self-assessment.
“When we kept applying pressure and testing it rigorously, at least it was honest,” Churchman said. “It was almost like it was saying: ‘Listen, at the end of the day, I’m better at making myself sound comprehensive than actually being comprehensive.’”