Emma RadmoreSheldon Mills, speaking at the FCA’s Supercharged Sandbox Showcase event, has headlined his review. He stressed the review does not change the FCA’s regulatory approach but that it wants to understand what the opportunities and risks are as AI becomes more capable, and see how it can reshape competition and the customer relationship and therefore how regulation may need to adapt.
The key is to look ahead, and to work out how to design for the unknown. He spoke of three particular types of AI :
- assistive AI, here today to explain and compare products, fill in forms and generally support consumers while leaving decisions in the consumers’ hands;
- advisory AI, which is emerging, with systems that will “nudge”, recommend and encourage action, for example to change suppliers or lenders. These can give good outcomes, but equally raise concerns about transparency, neutrality and the basis of advice; and
- autonomous AI, which is approaching to create agents acting within set boundaries, to take action to move money, negotiate renewals and so on, which has the power to be transformative in helping consumers get better returns and better value products – but is might make mistakes, consumers might not understand enough to stay in control and commercial incentives might shape the recommendations people see.
So the FCA wants to understand how firms can unlock opportunities safely. It wants to see how AI might embed or amplify bias, and not be able to explain why a particular decision was made; and how to address the risks of AI making a decision which appears logical but isn’t actually right for the particular consumer.
It wants to work out what “reasonable steps” look like in terms of SMCR accountability, when models and the way they behave is constantly changing.
But of course, the period for comment is short – ending on 24 February, with the report due in the summer.