"Isaac Asimov, AI, and the Goals of Financial Regulation"
One of the main issues currently under discussion in European financial regulation is whether competitiveness objectives should be part of the purview of European regulatory and supervisory agencies, alongside their primary mandate of safeguarding financial stability. The debate has gained momentum following the UK reform introduced in 2023, which included objectives of this kind, and in tandem with the regulatory simplification efforts being championed by the European Commission. Notably, Brussels recently launched a consultation on the competitiveness of the financial industry and is expected to publish a report in the coming weeks. It is important to clarify that both the UK reform and the proposals currently on the table in the EU refer to competitiveness in a dual sense: that of the financial sector itself and that of the broader economy it helps to finance.
Most authorities and several academics happen to be opposed to the inclusion of competitiveness objectives, based on seemingly compelling arguments:
- (i) the most effective way for authorities to support the competitiveness of the financial sector is to ensure that they fulfill their financial stability objectives;
- (ii) assigning multiple objectives ultimately weakens the accountability of an independent agency;
- (iii) the European Supervisory Authorities (ESAs), including the EBA, ESMA, and EIOPA, are accountable to the Commission, which is responsible for ensuring that the objectives set by the political authorities are properly balanced; and
- (iv) the Single Supervisory Mechanism (SSM, which is part of the ECB) is a supervisory rather than a regulatory body, so assigning it regulatory objectives would make little sense.
Interestingly, some of these arguments are contradictory, such as (ii) and (iii), since agencies are either independent or they are not. But addressing them one by one:
- (i) while it is true that financial stability and competitiveness are mutually reinforcing over the long term, it is reasonable to assume that some trade-off exists between them in the short term;
- (ii) while it is also true that multiple objectives erode accountability, bringing an implicit trade-off into the open ultimately improves accountability;
- (iii) in practice, the ESAs operate with a considerable degree of independence from the Commission; and
- (iv) the SSM regulates through the back door, using mechanisms such as so-called supervisory expectations.
To shed light on this debate, I propose the following exercise: imagine that financial regulation and supervision were entrusted to an artificial intelligence system and that its mandate had to be defined through an instruction or “prompt,” to use the technical term. One option would be to set a single objective: preserving financial stability. A second alternative would be to add a secondary mandate: preserving financial stability and, without prejudice to that objective, championing the competitiveness of the financial sector and its contribution to economic growth.
This second formulation closely resembles the objectives of monetary policy: under Article 2 of the ECB Statute, the ECB’s primary objective is to maintain price stability, but, without prejudice to that objective, it shall support the general economic policies of the Union.
The main argument for introducing a secondary objective is that European regulatory and supervisory agencies cannot disregard the competitiveness of the financial sector and of the wider European economy. In economics, situations involving multiple equilibria are common: when plotting two variables on a graph, it may be possible to achieve the same value for one variable with two or more different values for the other. In this case, let’s imagine that the same degree of financial stability could coexist either with a highly competitive economy or with a weakly competitive one. Under the “prompt” described in the first option, the AI system would draw no distinction between these two realities, which would be absurd.
To use another example, if we asked an AI system to set highway speed limits and told it that its objective was to prevent accidents, but said nothing about allowing people to get where they’re going within a reasonable amount of time, the speed limit would likely end up being set at an extremely low level — say, 10 kilometers per hour: maximum safety, zero efficiency.
Eighty years ago, Isaac Asimov set out the laws of robotics in his “I, Robot” series. In them, he illustrated the importance of properly specifying and prioritizing objectives in order to avoid absurd outcomes. We should revisit those lessons and avoid leaving important objectives out of the equation.