NASA’s new dark energy space telescope can also detect killer asteroids
technologyreview.com·3h ago
AI Research
arXiv CS.AI·Jun 9·5 min read
Explainable AML Triage with LLMs: Evidence Retrieval and Counterfactual Checks
TL;DR
Investigators face challenges in quickly triaging a high volume of anti-money laundering (AML) alerts due to strict regulations. A new explainable AML triage framework leverages large language models (LLMs) to summarize evidence and perform counterfactual checks, ensuring compliance and accuracy.
✦ Why It Matters
Engineers can implement LLMs in regulated environments to enhance decision-making accuracy and compliance in AML workflows.
Key Takeaways
- Utilize LLMs for evidence-based decision-making in AML.
- Implement structured output contracts for LLMs to ensure citation.
- Conduct counterfactual checks to validate triage recommendations.
- Adopt evidence bundling techniques to improve auditability.
How It Works
The framework combines evidence retrieval from various sources, such as policy guidelines and transaction data, to create a comprehensive view for triage. It employs a structured output contract that requires LLMs to provide clear citations, distinguishing between supporting and contradicting evidence.
Counterfactual checks are used to test how small changes in the input affect the triage outcome, ensuring that the rationale behind decisions is coherent and justifiable.