Innovative AI models should be trained to adhere to a ‘constitution’ or a set of regulatory rules aimed at mitigating the risk of harmful behavior, according to a senior policy maker from the Bank of England.
In a speech at CityWeek in London, Randall Kroszner, an external member of the Bank of England’s financial policy committee, discussed the differences between fundamentally disruptive and more incremental innovations, along with the distinct regulatory challenges each presents.
Kroszner noted, “When innovation is incremental, it is easier for regulators to understand the consequences of their actions and to undertake regulatory measures that align with their financial stability goals.” In contrast, he emphasized that in the rapidly evolving field of AI, the pace of innovation can be disruptive, making it significantly harder for regulators to determine the appropriate actions to ensure financial stability and to anticipate unintended consequences for growth and innovation.
He raised concerns that the upcoming Digital Securities Sandbox—designed for utilizing emerging technologies like distributed ledger technology in the issuance, trading, and settlement of securities—may not adequately address the challenges posed by artificial intelligence technology. He remarked, “Fundamentally disruptive innovations, such as ChatGPT and similar AI tools, often involve extraordinarily rapid scaling that tests the limits of regulatory tools. In such situations, a sandbox approach may not suffice, and policymakers may need to innovate further in response to disruptive changes.”
Kroszner also referenced recent comments from his colleague Jon Hall, who discussed the potential risks associated with neural networks evolving into ‘deep trading agents.’ Hall warned that these agents could have incentives misaligned with those of regulators and the public good, potentially exacerbating market shocks and undermining stability.
To mitigate these risks, Kroszner proposed that training neural networks to adhere to a ‘constitution’ or a framework of regulatory rules could be beneficial. He suggested that this ‘constitutional’ approach could be piloted in a sandbox environment as a means to guide new innovations while promoting financial stability. “In cases where disruptive changes occur at such a rapid pace that a sandbox approach may not be applicable, a ‘constitutional’ framework may be the most suitable method to pursue,” he stated.
Discover new challenges and opportunities presented by artificial intelligence in the banking sector at Finextra’s inaugural NextGenAI conference on November 26, 2024. Register your interest here.