"Learning to Replicate Expert Judgment in Financial Tasks" – Thinking Machines
Thinking Machines developed models to replicate expert judgment in financial tasks by training on labeled data from domain specialists. Their approach aims to automate complex financial decisions while maintaining the accuracy and nuance of human expertise.
Background
- Thinking Machines is a data science and AI consulting firm based in the Philippines, not a hardware company despite the name.
- The article describes a project where the firm built a machine learning system to replicate the decision-making of human experts in financial tasks — specifically, loan underwriting or credit assessment.
- This is a practical application of "expert systems" or "behavioral cloning" in AI: instead of writing explicit rules, the model learns from historical decisions made by experienced loan officers.
- The key challenge addressed is that expert judgment is expensive, scarce, and inconsistent, so automating it can make financial services more scalable and consistent in markets like the Philippines.
- The approach likely involves supervised learning on labeled data (past loans, expert decisions) and possibly techniques to handle subjective or borderline cases where experts disagree.