An Executive Guide on How to Use Machine Learning and AI for AML Compliance
1 like277 views
The document is an executive guide on utilizing machine learning and artificial intelligence for anti-money laundering (AML) compliance. It outlines the evolution of AML solutions, the necessity of machine learning in detecting complex financial crimes, and the importance of creating a 'ground truth' model for anomaly detection. Key takeaways emphasize the proactive use of AI technology to combat financial crime effectively.
An Executive Guide on How to Use Machine Learning and AI for AML Compliance
- 1. https://www.alessa.caseware.com/https://www.alessa.caseware.com/ An Executive Guide on How to Use Machine Learning and AI for AML Compliance
- 2. https://www.alessa.caseware.com/https://www.alessa.caseware.com/ Definitions 2 1. Artificial intelligence (AI), sometimes called machine intelligence, is intelligence demonstrated by machines, in contrast to the natural intelligence displayed by humans and other animals. 2. Machine learning (ML), is a field of computer science that uses statistical techniques to give computer systems the ability to "learn" (e.g., progressively improve performance on a specific task) with data, without being explicitly programmed.
- 3. https://www.alessa.caseware.com/https://www.alessa.caseware.com/ 3 Evolution of AML Solution @IBM
- 4. https://www.alessa.caseware.com/https://www.alessa.caseware.com/ Today’s State Actors and Criminal Organizations are employing highly sophisticated tradecraft techniques that: • Mask their activities • Defeat existing detection techniques • Employment of Dark Web • Continuously adapting tradecraft to defeat defense in depth • Mapping target-victims defensive capabilities, vulnerabilities as well as partners for least path of resistance, creation of backdoors 4 Evolution of Financial Crime Deception Occurs at Multiple Levels Initial Infection And Probing Malware Operation Exfiltration Where AI systems are deployed to defend against threats
- 5. https://www.alessa.caseware.com/https://www.alessa.caseware.com/ • We do not program AI to detect threats, we train it to do so! • Designed to detect in real-time unknowns using machine learning algorithms • These systems will learn and interact to provide expert assistance in a fraction of the time it now takes 5 AI vs IA Intelligence AugmentationArtificial Intelligence AI IA
- 6. https://www.alessa.caseware.com/https://www.alessa.caseware.com/ 6 Anomaly Detection Artificial Intelligence Capabilities in AML Suspicious Behavior Monitoring Cognitive Automatic Robotic Processing
- 7. https://www.alessa.caseware.com/https://www.alessa.caseware.com/ Key aspects: • Complex problems require a toolset of (standard) analytics techniques and tuning efforts in collaboration between ML and business experts. • Approach needs to be combined with expert feedback: Unsupervised learning and supervised techniques • The entire solution consists of much more than just the model itself. 7 Our Machine Learning Approach Exploratory Analysis Data Cleansing Apply various ML modules Results Analysis Create Ground Truth Create Ground Truth
- 8. https://www.alessa.caseware.com/https://www.alessa.caseware.com/ 8 Ground Truth By studying all past transactions, a model is created that learns what it has seen and is taught that this is normal. The resulting knowledge is referred to as the “Ground Truth.” Subsequent transactions are then compared to this Ground Truth and the inability to reconcile this transaction is returned as a score. This score can have thresholds attached to them, and using these thresholds we can label it as an anomaly. Due to this procedure, it is very important to have “pure” data while training the model. Feeding it actual anomalies will have it learn them as well, much as a child can be desensitized by television. Tabular data is cleaned and fed to the model. Scores are returned and can be visualized.
- 9. https://www.alessa.caseware.com/https://www.alessa.caseware.com/ 9 Anomaly Detection original row autoencoded row reconstructed row • Unsupervised • Used to detect unknowns • Forms a basis for more targeted detection systems Scoring output Individual label breakdown
- 10. https://www.alessa.caseware.com/https://www.alessa.caseware.com/ 10 Fraud Detection Model A Model A B Transfer learning • Supervised • Transfer Learning • Ability to target very specific scenarios
- 11. https://www.alessa.caseware.com/https://www.alessa.caseware.com/ 11 Predictive Analytics Model A C Model BVS. • User Behavior Analysis • By learning from past trends, they can be extended • Can be used to create new rulesets
- 12. https://www.alessa.caseware.com/https://www.alessa.caseware.com/ 1212 Takeaways • No, Artificial Intelligence (AI) won’t take over the world. • Yes, implement AI and Machine Learning Technology to stay ahead. • Stop taking a reactive/defense approach to fight financial crime. • Take proactive/preventative approach (go on the offense) by using AI to fight financial crime.
- 13. https://www.alessa.caseware.com/https://www.alessa.caseware.com/ An Executive Guide on How to Use Machine Learning and AI for AML Compliance
Editor's Notes
- #3: Discussions of artificial intelligence (AI) have created a certain amount of unease by those who fear it will quickly evolve from being a benefit to human society to taking over. Even Stephen Hawking and Elon Musk have warned of AI’s threats. However, we’re not all operating from the same definition of the term and while the foundation is generally the same, the focus of artificial intelligence shifts depending on the entity that provides the definition.
- #4: Why Good Processes Alone Can’t Make AML Stick Anymore Government agencies and financials services organization primarily focused on rules based transaction monitoring to fight money laundering. These types of solutions focused only on defensive capabilities such as process improvements in regulatory compliance, risk reporting and mitigation. Disadvantage of rule based AML systems alone generate a large number of false positive and only detect what you are looking for (what we call the “know” factor). The importance of the variety, velocity and volume of data has forced organization to rethink their AML technology strategies and evolve from government regulation rules based solution to more sophisticated AI based solutions. Then talk about emerging markets and on…. Why is this important you? As long as we adhering to regulatory bodies…what does it matter. Machine data is digital information created by the activity of computers, mobile phones, embedded systems and other networked devices. Such data became more prevalent as technologies such as radio frequency identification (RFID) and telematics advanced.
- #5: Deployed AI systems of their own to intrude in you systems and customers by Identity Theft 2. Account Takeover Behave like normal law abiding citizens. Impact the bottom line, brand exposure and increased fines.
- #6: Know that we know that criminal organizations are leveraging sophisticated techniques. What do we do about? And How? Which will be cover later.
- #7: I’m going to drilldown on more level to introduce several AI capabilities being used today by organizations such as yours. These capabilities include: 1. Anomaly Detection – is an advanced technique used to detect behavior that don’t fit within the a normal data profile. 2. Suspicious Behavior Monitoring – focuses on what we call known labels such as fraudulent behavior. 3. Cognitive Capabilities – such as cognitive virtual assist that leverage NLP to provide compliance SME insight into changes in regulatory requirements. 4. Automation Robotic Processing – automates repetitive manual activity such as remediation and workflow process. All of these capabilities have 2 common dimensions: ML and Data. I will hand it off to Josip who will take about the several uses and approach.
- #8: Thank for segue Intro myself Intro ML, math, process Discuss process, show model place The solution is not just the model
- #9: Fraud is a unique problem Solution to this, GT Unsupervised learning Baseball coach
- #10: Maybe a little backwards Follows a child’s learning path Using GT, tries to recreate the trans Difficulty shows up as an anomaly
- #11: Transfer learning Because the grouping is already known Add fraud Can now isolate specific instances
- #12: Future (is now?) Predictive solutions Using the two previous models Extension GANs
- #13: - Stop Looking in your rear view mirror. - Look forward to get insight to prevent financial crime before they happen.