Why Is Explainability So Important Right Now?

As AI systems and machine learning-powered tools proliferate in our everyday life, both practitioners and critics are becoming more vocal about the need to know how they produce the outcomes they do. The costs of not knowing—financial, social, technical—have become too high. This week, we invite you to read three excellent posts that place explainability front and center. (If your interests lie elsewhere, have no fear: we’ll also branch out into self-parking cars, board games, and other topics—so do read on…)

• Get the basics of explainability right. In the context of ML, explainability can mean different things depending on the specific moment of the product lifecycle you’re looking at. Fortunately, Aparna Dhinakaran is here with a primer that’s both accessible and comprehensive, laying the groundwork for further learning and deeper understanding.
• Reframe your thinking around XAI. After spending years researching explainable AI, Bryce Murray invites us to think less about algorithms and more about end users. Why? As Bryce states in his post, “Explainability by design gives AI engineers the most freedom to develop relevant explanations to empower the users of the algorithm(s).”
• Learn how to explain your ML model in Python. For a more hands-on approach to the question of explainability, look no further than Khuyen Tran’s visual step-by-step tutorial on SHAP and Shapely values. It’s a patient walkthrough that covers a lot of ground, from basic definitions to a detailed implementation.

Photo by Andrew Neel on Unsplash

If you’re looking for other hearty topics to dive into, you’re in for a treat this week—TDS contributors covered a lot of ground recently.

• Explore the power of clustering to produce public-health insights. Many of you no doubt know how COVID-19 has made the life of vulnerable patients even more difficult than usual. Christabelle Pabalan and coauthors Victor Nazlukhanyan and Daniel Carrera decided to take a closer look into one such group, and showed the compounding effects of the pandemic on Parkinson’s patients.
• Consider a streamlined approach to building and scaling data teams. After launching data teams at both Instagram and Spaceship, Chris Dowsett shares key lessons from a process that can feel confusing, if not downright chaotic. Taking a cue from microservice architecture, Chris recommends a roadmap that breaks tasks into small parts and keeps workflows nimble and easy to improve.
• Read about a creative way to tackle gender bias. Tasked with increasing exposure and reach for an organization that promotes gender equality in the tech sector, Nina Sweeney relied on her data-analytics chops to find the NYC subway stations where their target audience was most likely to come in contact with their message.
• Use data science to beat your friends and family at Monopoly. If you’re frustrated by a long game-night losing streak, or even if you’re just curious about the unexpected intersections of data science and board games, don’t miss our recent conversation with Jake Mitchell, whose series on the topic explores the stats behind Chutes & Ladders, Monopoly, Connect 4, and others.
• Follow along as a genetic algorithm learns how to park a car. In under 500 lines of code, Oleksii Trekhleb trains a car to parallel-park—and, along the way, shares his entire journey of creating a complex, fascinating project, with plenty of code and visualizations included (not to mention parking animations).

Thank you for joining us this week! If you enjoyed the thematic rollercoaster of these reading recommendations, we hope you’d consider supporting us and our authors by becoming Medium members.

Until the next Variable,
TDS Editors

Recent additions to our curated topics:

Getting Started

• The What, Where, and How about Continuously Learning on the Job as a Data Scientist by Quoc Tien Au
• How to Find Your Way through the Different Types of SQL by Marie Lefevre
• Binning Records on a Continuous Variable with Pandas Cut and QCut by Allison Stafford

Hands-On Tutorials

• Render Interactive Plots with Matplotlib by Parul Pandey
• The FP Growth Algorithm by Joos Korstanje
• Signs You Are Using Data Visualization Tools Wrong by Tessa Xie

Deep Dives

• The Mystery of Feature Scaling Is Finally Solved by Dave Guggenheim
• How Percentile Approximation Works (and Why It’s More Useful than Averages) by David Kohn
• How to Create Products that Rely on Machine Learning by Søren Vedel

Thoughts and Theory

• Derive Insights from Health Data Using Knowledge Graph Technologies by Honghan Wu
• t-SNE Machine Learning Algorithm — A Great Tool for Dimensionality Reduction in Python by Saul Dobilas
• A Fair World in ML Modeling Where Men and Women Should Be Treated the Same Way by Sun Weiran