Learning to Replicate Expert Judgment in Financial Tasks
Thinking Machines Data Science has developed a method to replicate expert judgment in financial tasks using machine learning, enabling automated credit risk assessment and transaction monitoring while maintaining accuracy and consistency.
Background
- **Thinking Machines** is a Singapore-based AI consultancy (unrelated to the 1980s firm) that builds custom ML systems for finance and government.
- This article describes a project where they trained a model to **replicate expert judgment** — e.g., a senior analyst's loan underwriting decisions — rather than predicting an outcome like default.
- Why do this? Outcome data (did the loan default?) is often scarce, noisy, or takes years to observe. But historical expert-labeled cases are abundant, letting the model learn the expert's reasoning directly.
- The model uses supervised learning on past expert decisions plus interpretability tools, so banks can see *why* a decision was made — key for regulators in Singapore, Hong Kong, and others that require explainable AI in financial services.
- Broader context: Many Asian banks are adopting AI for credit scoring, fraud, and compliance. Replicating judgment lets them scale rare expertise while staying auditable.