Driverless AI Use Cases in Finance and Cancer Genomics

This video was recorded in San Francisco on February 9th, 2019.

Slides from the session can be viewed here: https://www.slideshare.net/0xdata/mar...

‍Marc Stein is the founder and CEO of Underwrite.ai. Underwrite.ai applies advances in artificial intelligence derived from genomics and particle physics to provide lenders with non-linear, dynamic models of credit risk which radically outperform traditional approaches. Marc’s career has always revolved around deep interests in artificial intelligence, quantum physics, genomics, sugar cream pie, and all ice cream flavors found at Berthillon and the challenge of how to combine all these in practical applications.

‍

Why Nonlinear Problems Need Nonlinear Models

In this talk, Marc Stein, Founder and CEO of Underwrite.ai, explores a core idea that drives both his work in finance and genomics. Many of the most important problems in the world are nonlinear by nature, yet they are often approached with linear models that oversimplify reality.

Human beings tend to reduce complexity into binary rules. Approve or deny. Healthy or sick. Low risk or high risk. While this makes decision making easier, it ignores the interconnected variables that actually drive outcomes in complex systems.

Rethinking Credit Risk Modeling

Stein begins with credit modeling in the United States. Most lenders rely on a small set of variables such as FICO score, debt to income ratio, payment history, and credit inquiries. These models work reasonably well for borrowers with strong credit profiles, particularly those with FICO scores above 700.

The problem appears as you move further down the credit spectrum. Traditional models become less predictive, excluding large portions of the population who may be creditworthy but do not fit neatly into simplified scoring frameworks.

Underwrite.ai challenged this approach by asking a different question. What happens when you evaluate thousands of attributes instead of just a few? By applying nonlinear machine learning models, the team found they could significantly improve predictive accuracy, especially for underserved borrowers. This made it possible to expand credit access while still maintaining strong returns for lenders.

Insights from Cancer Genomics

The foundation for this approach came from Stein’s earlier work in cancer genomics. In prostate cancer diagnosis, modern diagnostics typically focus on a small group of biomarker genes, producing accuracy rates in the high seventies.

By analyzing all 25,000 gene expression levels rather than isolating a handful, Stein’s team discovered that thousands of genes correlate with cancer risk. When these signals were modeled together, prediction accuracy increased into the nineties. The lesson was clear. Complex biological systems cannot be accurately understood by looking at only a few variables.

Linear and Nonlinear Models in Practice

Stein reinforces this point with real world lending examples from South Korea and the United States. In South Korea, where credit systems are already highly efficient, linear models perform well. Even there, nonlinear approaches were able to push performance higher by incorporating more data and more sophisticated feature interactions.

In the United States, where consumer lending models are less efficient, the gains were even more pronounced. Nonlinear models outperformed traditional scoring methods across the entire risk spectrum and were particularly effective at concentrating defaults into higher priced tiers. This improves both risk management and profitability.

The Bigger Picture

Whether applied to credit risk, cancer diagnosis, or financial inclusion in emerging markets, the message remains the same. Highly complex problems demand nonlinear solutions.

Stein closes by emphasizing the importance of asking why, not just what. When nonlinear machine learning models are applied thoughtfully, they unlock opportunities to solve problems that were previously considered too complex to tackle and create meaningful impact at scale.