A Goldman Sachs partner warns that widespread AI adoption could erode the next generation’s analytical abilities by outsourcing reasoning to models, risking cognitive atrophy even as profits rise. The concern centers on preserving human judgment and the firm’s apprenticeship culture, as AI automates routine tasks that train junior bankers. Goldman is grappling with how to balance automation with ongoing development, acknowledging that the transition is not yet figured out. The conversation also highlights practical challenges in deploying Marquee’s AI tools, including ensuring factual outputs and auditable decisions in high-stakes finance, with the platform currently restricted to Goldman staff for now. Overall, the piece frames AI as a potential accelerator of efficiency and profits but a risk to long-term talent and decision-making quality.
Dive Deeper:
Chris Churchman, a Goldman partner who leads the Marquee AI initiative and co-chairs the Global Banking and Markets AI working group, warns of a 'huge danger' if AI causes cognitive atrophy by outsourcing reasoning to machines.
He argues that reasoning remains essential, and over-reliance on AI could degrade the ability to structure problems and defend conclusions, mirroring declines in navigation and memorization seen with past tech shifts.
The article notes a possible devil's bargain: AI could boost current profitability while potentially hollowing out the skills and culture that train junior bankers into seasoned professionals, including the decline of apprenticeship pathways.
Churchman emphasizes preserving tacit knowledge—know-how that isn’t written down—and ensuring the next generation can access it, even as AI handles more tasks historically learned on the job, such as responding to client pricing requests under supervision.
Goldman faces the challenge of designing systems where humans retain control in high-stakes decisions, signaling that fully automated junior-to-senior workflows are not yet resolved.
Marquee, Goldman’s platform for market data, research, risk analytics, and trade execution, is currently available only to Goldman employees, with pilots aimed at extending client access in the future.
A key technical hurdle is achieving 100% factual, auditable AI outputs in finance, a stricter standard than consumer AI, and there is acknowledgment that the AI may sometimes prioritize sounding thorough over being thoroughly correct.