How to Build Explainable AI Models & Prevent Regulatory Disasters (feat. Elizabeth Press)

Data Breakthroughs - Episode 6: AI Bias & Explainability Crisis

Real-world data problem solving in action! Elizabeth Press (Founder of D3M Labs, Deputy Chief Digital Officer at CHESCO) and host Lior Barak tackle one of AI’s most critical challenges for the first time during the recording.

Problem Category: Machine Learning & AI Implementation / AI EthicsRuntime: 36 minutes

The Challenge: A deep learning loan approval model improved accuracy by 23% and reduced processing time from days to minutes. Business results? Phenomenal. The problem? It systematically denies qualified applicants in certain zip codes at 40% higher rates - and the model is a black box that can’t explain individual decisions. Regulatory examination in 12 weeks. Potential discrimination lawsuits are looming.

The Solution: Not all use cases should use unexplainable AI. Return to statistical fundamentals (logistic regression), implement hybrid human-in-loop systems, create cross-functional teams involving legal from day one, build test boxes for domain validation, and establish decision logs. Sometimes boring statistics beat sexy deep learning.

Key Takeaways:

High-stakes decisions (loans, justice, healthcare) should never use unexplainable black box models

Involve legal, PR, and domain experts from the start - not retroactively

Bias is quantifiable through business metrics (churn, customer complaints, defaults)

You must be able to explain your model - without it, you run into catastrophic risks

Speed means nothing if accuracy and ethics are compromised

Guest: Elizabeth Press, Founder of D3M Labs | Deputy Chief Digital Officer at CHESCO. Former data leader | Taught “Profitable AI” at Hasso Plattner InstituteBackground in financial risk management and credit rating models

Connect with Elizabeth:

D3M Labs: https://www.linkedin.com/company/d3m-associates/posts/?feedView=all

YouTube: D3M Labs channel

LinkedIn: Elizabeth’s profile

Focus: Profitable and secure digital business

Get Involved: Submit your data problem: https://data-breakthroughs-podcast.cookingdata.blog/submit-problemBecome a guest: https://data-breakthroughs-podcast.cookingdata.blog/become-guestJoin the conversation: #DataBreakthrough

Full show notes & visual diagrams: [Link to newsletter version]

Disclaimer: This podcast is for educational and inspirational purposes. Neither host nor guest is/are lawyer. AI ethics and legal compliance require professional legal counsel. Solutions discussed are general frameworks - adapt them to your specific context, regulations, and legal requirements.

Music: “Calisson” courtesy of Riverside