Explainable AI

Etica dell'intelligenza artificiale

Joe Franklin

Associate Data Literacy and Essentials Manager, DataCamp

What's explainable AI?

• AI systems whose internal workings are understood by humans
• Goal: Making AI's decision-making clear, understandable, and explainable
• Helps understand why and how AI makes decisions
• Major step towards ethical AI usage

¹ Icon made by vectorsmarket15 from www.flaticon.com

The central pillars

• Transparency, fairness, accountability are central
• AI conclusions should be accessible and logical to humans

• Models built with explainability at their core
• Uses interpretable models like decision trees or linear regression
• Power in seeing the process, despite possibly lower performance

¹ Icons made by juicy_fish & Becris from www.flaticon.com

How does it work?

How does it work?

Local Interpretable Model-agnostic Explanations (LIME)

• LIME as a translator that helps the model communicate
• Creates a simpler version of the model's decision process for a specific prediction
• Example:
  • Explains a movie's hit prediction based on factors like director popularity and high budget

¹ Icon made by Freepik from www.flaticon.com

SHapley Additive exPlanations (SHAP)

• SHAP: A detective of AI, revealing feature importance
• SHAP in Action
  • Director: 50%
  • Cast: 30%
  • Genre: 15%
  • Budget: 5%

Future of XAI

• Many more techniques and approaches exist in XAI
• The gap between XAI and traditional AI is shrinking
• Ongoing research is improving AI interpretability

Let's practice!

Etica dell'intelligenza artificiale