A Goldman Sachs partner leading one of the bank’s major artificial intelligence initiatives has warned that the rapid adoption of AI across Wall Street could weaken the analytical and reasoning skills of the next generation of financiers if employees become too dependent on machines.
Chris Churchman, who leads Marquee, Goldman Sachs’ digital platform for institutional clients, said AI could create a form of “cognitive atrophy” by taking over the reasoning and problem-solving tasks through which young bankers and traders traditionally develop their expertise.
“There’s a huge danger here that in the era of AI, we outsource our reasoning to these models, and we have cognitive atrophy that stops us being able to reason from first principles ourselves,” Churchman said.
Register for the next Tekedia Mini-MBA.
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
His comments came during an episode of Goldman Sachs’ “Exchanges” podcast, according to a transcript provided to CNBC.
Churchman’s concern goes beyond the possibility of AI making mistakes. He noted that Wall Street could inadvertently weaken its own talent pipeline by automating the routine work that has historically served as an apprenticeship for junior employees.
“Reasoning is still important,” he said. “You still need to reason about [problems] and structure it into an argument, and now we’re delegating reasoning.”
The issue is relevant to investment banks because much of the expertise required to make complex decisions is developed through repeated exposure to real-world situations rather than formal training.
Churchman compared the potential effect of AI with previous technologies that reduced the need for people to exercise certain skills, such as navigation and memorization. If algorithms increasingly perform analysis, generate recommendations, and execute routine decisions, junior employees could have fewer opportunities to develop the judgment needed to operate independently.
That could create a long-term problem for banks. AI may increase productivity and reduce costs in the short term, but the same technology could weaken the human expertise needed to supervise those systems and make decisions when circumstances fall outside established patterns.
Wall Street firms have already been examining whether AI can reduce the number of junior bankers needed relative to senior employees. That trend could accelerate if banks discover that AI can perform significant portions of entry-level research, modelling and administrative work.
Churchman said banks therefore need to preserve an apprenticeship model in which younger employees learn alongside experienced professionals.
“You learn by doing, and a lot of knowledge is tacit, it was never written down,” he said.
Goldman needs “to make sure we don’t lose that tacit and intuitive knowledge that some of our best people have today [and] to ensure the next generation have it too,” he added.
Currency trading provides one example. Churchman, who previously ran currency trading at UBS before joining Goldman in 2021, said junior traders traditionally develop their skills by handling client pricing requests while experienced traders supervise their decisions.
“We can absolutely automate that,” he said, “but then do we get the senior traders that fully understand?”
The question marks a potential feedback loop in financial-sector automation. If AI replaces enough entry-level work, future senior employees may never acquire the practical experience that enabled previous generations to become effective risk managers. The result could be an industry with sophisticated technology but a thinner pool of people capable of understanding when that technology should not be trusted.
Churchman said AI systems should therefore be designed so that humans remain responsible for high-stakes and highly uncertain decisions rather than becoming passive operators who simply approve machine-generated recommendations.
Even Goldman Sachs has not fully resolved how to manage that transition.
The bank has invested heavily in integrating AI into its operations, and Churchman is also co-chair of its Global Banking and Markets AI working group. But he acknowledged that the firm has not yet “figured out” the appropriate balance between automation and human judgment.
The technical challenge is equally significant.
Churchman said one of the hardest problems in developing Goldman’s AI capabilities is ensuring that responses are factually accurate and can be audited. That requirement is stringent in financial markets, where a seemingly minor error in a valuation, risk calculation or market-data interpretation can have substantial financial consequences.
Marquee, Goldman’s digital platform for institutional clients, provides hedge funds and other professional investors with access to the bank’s market data, research, risk analytics and trade-execution services.
Churchman said the AI component of Marquee is currently available only to Goldman employees as the bank continues to develop the technology.
Unlike consumer chatbots, where users may tolerate occasional incorrect answers, financial AI systems have to operate under much stricter standards. Institutions need to know where an answer came from, whether the underlying information is reliable and whether the result can be independently verified.
Churchman said the AI system itself effectively highlighted this challenge during testing.
“When we challenged it hard, at least it was honest,” he said. “It was like, ‘Look, in the end, I’m better at sounding thorough than being thorough.’”
That distinction could become one of the defining issues for AI adoption in finance. Generative AI systems are often capable of producing highly convincing explanations even when their underlying conclusions are wrong. In financial markets, confidence and fluency cannot substitute for accuracy.
The challenge for Goldman and its peers is therefore not simply determining how much work AI can automate. Banks must also decide which tasks should remain part of the human learning process and how much responsibility employees should retain when machines are involved.
The economic incentive to automate is strong. AI can process information rapidly, reduce repetitive work, and potentially allow senior employees to handle larger volumes of business. But Churchman warns that if junior employees lose the opportunity to develop judgment through that same work, banks could eventually face a shortage of experienced professionals capable of challenging AI systems when they fail.