WASHINGTON — Acting Comptroller of the Currency Michael Hsu warned against regulators giving banks or other financial firms too much leeway in exploring applications for artificial intelligence, saying instead that regulators and firms should "co-learn" about this emerging technology together.
Speaking Thursday at an AI conference put on by FinRegLab, a nonprofit organization focused on technology in financial regulation, Hsu was asked about the potential for regulators to give banks an AI "sandbox" where the firms could explore potential applications of the technology without fear of enforcement or liability risk. Hsu said AI represents a unique challenge in that there are risks associated with banks and regulators moving too quickly in adoption of the technology and a roughly equal risk of moving too slowly. He said the sandbox approach may not be the most effective way of achieving that balance.
"The caricature of the worrisome sandbox is really a request from a company, a bank, the industry to say 'We want a space with no liability, and if you happen to learn from that, great, but that's what we really want.' I think that's dangerous," Hsu said. "I get it, I can understand why that could be a vehicle for … improvements, but the starting premise is 'Leave us alone.'
"I think a better way to approach it is one where, again, the focus is on learning — how do we maximize learning, both for the organization or institution that wants to do it, and for the supervisors and the regulators. How do we co-learn?" he said.
A better regulatory approach would be for both banks and regulators to explore the technology and identify risks and benefits together rather than having banks accelerate AI innovation while regulators hit the brakes, Hsu said.
"AI has this [Reinforced Learning from Human Feedback] thing, it's almost like we need RLHF for supervision," Hsu said. "[We need] supervised learning as regulators, where we're in a space where we're co-learning as these things are being developed, rather than giving permission for a [sandbox] and then if something bad happens, someone says, 'No, you told me it would be OK.' I think this requires a bit more of an interactive engagement for this to work."
When asked whether the OCC and other regulators have the tools to establish a co-learning AI regulatory model, Hsu said, "I think so. I think we have to try."
Hsu's comments came after a panel earlier in the morning with Senate Banking Committee member Mike Rounds, R-S.D., in which Rounds
"The oversight coming from the agencies is one that says, 'OK, here's what we want to pursue,' so it's a joint venture, so to speak, and everybody wins," Rounds said.
"That's really what this is all about — you're a business, and you need tools, and you expect your financial service organization … to present you with more data in a more timely fashion," he said. "You want them to offer these products and services to you, but if they're restricted because of regulatory oversight — this is not allowed, we don't know how to do that — then we'll never get ahead, and other people in other countries will do this to us."
