Bank of England Policy Maker Calls for AI Models to Respect Regulatory Rules to Reduce Harmful Behavior

Bank of England policy maker, Randall Kroszner, is advocating for innovative AI models to be trained to respect a ‘constitution’ or set of regulatory rules in order to reduce the risk of harmful behavior. In a speech at CityWeek in London, Kroszner highlighted the challenges posed by disruptive innovation, particularly in the case of artificial intelligence.

Kroszner emphasized the need for regulators to understand the consequences of their actions when dealing with rapidly evolving technologies like AI. He pointed out that incremental innovation is easier to regulate, but disruptive innovations, such as AI tools like ChatGPT, present a greater challenge for regulators.

The senior Bank of England official suggested that the central bank’s Digital Securities Sandbox may not be sufficient to address the regulatory challenges posed by AI technology. He proposed the idea of training neural networks to respect a ‘constitution’ or set of rules to mitigate the risks associated with their rapid scaling and potential misalignment with regulatory goals.

Kroszner also referenced a speech by his colleague, Jon Hall, who warned about the risks of neural networks becoming ‘deep trading agents’ and potentially destabilizing the market. He argued that a ‘constitutional’ approach combined with a sandbox could be an effective way to guide new innovation and support financial stability.

Overall, Kroszner’s call for AI models to be trained to respect regulatory rules reflects the growing recognition of the need for responsible innovation in the rapidly evolving technology landscape. By implementing a ‘constitutional’ approach, regulators may be better equipped to address the challenges posed by disruptive technologies like AI.