3 questions for Anne-Cécile Krieg, Deputy Head of the department in charge of model risk management. The publication, last July, of the white paper “Operationalizing the risk management of artificial intelligence systems” shows, if there was still a need, to what extent the subject of Artificial Intelligence (AI) constitutes an accelerator of competitiveness and , at the same time, a risk accelerator for financial institutions.
So how do we approach this challenge? The point with Anne-Cécile WarDeputy Head of the department in charge of model risk management, who spoke this November 26 at the ACPR conference on these same subjects.
Anne-Cécile, what are the major benefits in terms of competitiveness that AI offers to players in the financial sector?
I believe that the benefits of AI in our universe are no longer in doubt! Processes like document review and data classification are greatly accelerated by artificial intelligence. It is also a technology that we use and which has proven itself, particularly in compliance filtering. Large flagship projects have also brought benefits to our clients while strengthening our risk management. I am thinking, in particular, of the overhaul of the framework for authorized overdrafts for individual customers or the MOSAIC (More Security With Artificial Intelligence) project which makes it possible to automate processes for detecting fraud on means of payment, with an analysis of flows, and the triggering of alerts upon the identification of abnormal or suspicious events.
And as a corollary, what constitute the major risks?
Artificial intelligence induces two main types of risks, which are two sides of the same coin: the risk of depriving oneself of AI out of fear and the risk of not sufficiently regulating its use.
This framework must in particular take into account technical issues amplified by more sophisticated approaches (for example, encoded biases) or new ones in the context of AI (such as cybersecurity). It also involves a multidisciplinary risk analysis (IT, model, operational risks).
How does a group like ours organize itself to integrate AI with controlled risk?
Compared to other industries, we are fortunate to already have a framework that we can capitalize on; this is based on regulatory expectations, for example in terms of modeling, data or control. We are enriching it to address the specific challenges brought about by AI, in particular on the basis of a collective approach undertaken with our peers in order to identify best practices. This fruitful collaboration resulted in the writing of a white paper on the subject, involving BNPP, La Banque Postale and Société Générale. This document reviews all the risks identified and offers a sort of “toolbox” for each of them.
Internal training is also a key issue, requiring a multiplication of channels in order to involve all hierarchical strata, the provision of generalist and expert training. As we know, learning and knowledge function as safeguards.
Sharing of best practices, feedback, training – these are certainly the empirical dimensions that will ensure a reasoned deployment of the undeniable benefits of AI.
Find out more: the white paper dedicated to AI gen (July 2024)
