David Bounie, Prof. david.bounie@telecom-paris.fr Tlcom Paris, - PowerPoint PPT Presentation

David Bounie, Prof. david.bounie@telecom-paris.fr Télécom Paris, Institut Polytechnique de Paris ww.xai4aml.org 1

The team David Bounie, Economics Winston Maxwell, Law Stéphan Clémençon, Applied Mathematics Astrid Bertrand, PhD 2

The partners

The AML process in a nutshell Law enforcement Bank processes FIU Suspicious Transaction Referral to juridical Customer & transaction Alert Alert confirmed Report authority data, sanctions data Complementary sources*: Web, social media, newsfeed… AUTOMATED REVIEW SYSTEM: FIRST LEVEL SECOND LEVEL FINANCIAL LAW ENFORCEMENT deterministic or improved by machine REVIEW REVIEW INTELLIGENCE UNIT DATA GATHERING learning 4

99% of alerts have little or no utility to law enforcement Law enforcement processes Bank processes Out of 100 alerts, 10 are FIUs analyze only 2 FIUs forward 1 to law converted into SARs (>90% false SAR cases out of 10 enforcement positives) TMS generates 100 alerts 5

€200 billion in criminal funds circulate every year in Europe… …Yet only 1% of criminal funds are confiscated. 1 1 Europol, “From suspicion to action - Converting financial intelligence into greater operational impact.”, Sept. 2017. 6

AI can help improve AML systems, but creates new forms of risks for individuals. Current AML systems exhibit high false positive alerts while showing low success rates in catching illicit funds. Explainability is a key factor to mitigate those risks. 7

How AI can help? Suspicious Customer & transaction data, Transaction Referral to sanctions data Alert Alert confirmed Report juridical authority Complementary sources*: Web, social media, newsfeed… AUTOMATED REVIEW SYSTEM: FIRST LEVEL SECOND LEVEL FINANCIAL LAW ENFORCEMENT DATA GATHERING deterministic or improved by machine REVIEW REVIEW INTELLIGENCE UNIT learning Customer Anomaly detection Prioritization of Vizualisation tools* segmentation alerts* * Not Always deployed by banks. AML stages where Machine Learning can be valuable 8

Why is explainability important? AML-CFT: Why is it suspicious enough to shut down? SOCIETAL Fundamental rights: Why are my assets being frozen? POINT OF VIEW Is my algorithm functioning properly? TECHNOLOGIC AL POINT OF Do I detect any drift in behavior over time? VIEW Is my model AML compliant? REGULATORY Contrarily to other profit-maximizing firms that will always prefer efficiency over explainability , Financial Institutions won’t risk a fine for POINT OF VIEW AML, a field that has no business value. 9

Challenges for explainability in AML Multiplicity of business Multiplicity of audiences: Internal controller, Auditor, ACPR, cases: bank/insurance TRACFIN, Individual impacted by the model, CNIL. R&D CHALLENGES FOR Avoid bias & discrimination EXPLAINABILITY Go beyond post-hoc approaches (LIME, SHAP, etc.): Graph Conv. Networks, etc. IN AML Access to the data : Economic viability : Data used for the explanation can The explanation should also take raise data privacy issues. into account economic factors. 10

Thank you for your attention! Point of contact: david.bounie@telecom-paris.fr Research Chair website: www.xai4aml.org 11