BOK International Conference, Day 2

Presentation by researcher Sofia Kazinnik

The Federal Reserve Board building in Washington, D.C. [Reuters]
The Federal Reserve Board building in Washington, D.C. [Reuters]

By Kim Byeo-ri, The Herald Business

AI can serve as a powerful tool for transforming core central bank functions — but fully harnessing it will require central banks to overhaul not just their infrastructure, but also their organizational culture and institutional systems, a researcher argued Monday.

Sofia Kazinnik, a senior research scholar at Stanford University's Digital Economy Lab, made the case at the 2026 BOK International Conference in a presentation titled "Artificial Intelligence and the Fed."

"Central banks lag behind the private sector in AI adoption due to structural constraints — the absence of a single public mandate, strict public accountability, bureaucratic procedures and data barriers," Kazinnik said. She added that outdated technology infrastructure and siloed systems prevent central banks from effectively using or sharing their vast data for AI training.

She said AI could nonetheless become a powerful support tool for modernizing core central bank functions. On monetary policy, Kazinnik said AI can extract high-frequency data to compensate for lags in official statistics and improve the accuracy of real-time forecasts.

For financial stability, she said AI can analyze large volumes of unstructured text to proactively detect signals of systemic risk. However, she cautioned that AI agents' participation in markets carries a dual-edged quality — depending on system design and objective functions, they could either dampen or amplify financial herding behavior.

Drawing on job and budget data from the Federal Reserve System, Kazinnik estimated "AI-augmentable labor hours" and found that productivity gains could be achieved broadly across knowledge work throughout the Fed system.

"Based on 360 job roles within the Fed, I derived each regional Fed's AI exposure and annual total labor input hours," she said. "Generative AI broadly increases knowledge-work productivity across the Fed system."

In open market operations alone — a function handled exclusively by the New York Fed — AI could streamline roughly 1.17 million work hours per year, she calculated. For Treasury Services, the estimated gain was 3.18 million hours annually, while Cash Operations could see 3.51 million hours of work streamlined.

Kazinnik said central banks must build differentiated training pathways tailored to specific job functions, along with internal governance structures, to support AI adoption. "To fully realize AI's potential, central banks must fundamentally transform not just computing infrastructure, but also their organizational workflows, institutional systems and norms — alongside the technology itself," she said.


kimstar@heraldcorp.com