Responsible AI in Credit Risk: FICO Insights at Edinburgh Conference 2023

The 18th annual Credit Scoring and Credit Control Conference kicks off tomorrow, a premier global forum for showcasing innovation in responsible and ethical artificial intelligence (AI), machine learning (ML) and related data science. Sponsored by the University of Edinburgh’s Business School, this venerable technology event brings together credit professionals from across Europe and around the world to learn, share ideas and pursue the spark of inspiration.

AI is arguably today’s hottest business and technology topic, and I am honored to be on the mainstage at this prestigious event, delivering the AI keynote on Thursday. My presentation, “Responsible AI in Credit Risk: Palatable, Interpretable, Ethical, and Auditable,” illustrates how a standard interpretable AI model drives transparency in the application of Responsible AI to credit scoring –– an essential capability in enabling AI model explainability, the ability to expose imputed bias, address and remove that bias, and then audit model performance for ongoing accountability.

FICO’s recent machine learning and AI breakthroughs will be additionally presented by three data scientists, two of whom will give talks on patent-pending techniques for which I am co-inventor.

Explanation Dropout: Practical Counterfactual Explanations for Machine Learning Models
Matt Kennel, Senior Principal Scientist

Matt will discuss how AI model explainability can be improved by effecting the dropout of many variables that have the same explanation, thereby improving the “signal to noise” ratio in interpreting model explanation. This innovative approach can be superior to traditional methods such as the Shapley value.

First-to-Saturate Principle for Consistent Explanations of Neural Networks
Krzysztof Nalborski, Lead AI Scientist

Krysztof will share FICO’s novel approach to identifying the unique modes of activation of latent features in an analytic model that contribute deterministic explanation of neural networks, by limiting latent feature activation to the first mode to “max out” (saturate) first. This technique helps data scientists to better understand the properties and typically dense connections in the hidden layer(s) of neural networks, a key predictive component that is inherently challenging.

Developing Broad-Based Credit Scores to Balance Many Objectives
Gerald Fahner, Senior Principal Scientist

Gerald’s talk will illustrate how FICO applies cutting-edge data science responsibly to address real-world business and technology objectives. He will discuss FICO’s latest AI approaches to balance multiple desired scoring objectives, both high-level (predictive power, robustness, interpretability, fairness) and finer-grained goals such as predictive differentiation across various different loan types. 

If you’re in Edinburgh, please come to my Artificial Intelligence keynote and talks to learn about the latest techniques on the forefront of achieving Responsible AI in credit scoring transparency. You’ll see why the Edinburgh conference provides a chance to test, discuss, and shape our innovation responsibly — a data scientist’s dream and a welcome break from the challenges of our daily work.

How FICO Can Help You Build Responsible AI

• Watch my LinkedIn Live conversations on Responsible AI
• Discover FICO’s AI, machine learning and advanced analytics solutions
• Download our AI Playbook: A Step-by-Step Guide for Achieving Responsible AI
• Review The State of Responsible AI in Financial Services (fico.com)