Open and Understandable: Designing Explainable AI for Everyone
summary
Proposes replacing one big model with modular feature models (each detects “has wings”, “has beak”, etc.) and wiring outputs into a knowledge graph to produce stepwise, multi‑hop explanations. Compares LIME (local explanations) and SHAP (global importance) to a demo using a random forest on synthetic 8‑sensor IoT data with anomaly threshold 1.5; LIME/SHAP show which sensors matter but the graph maps sensors → conditions → failure and yields causal paths. Claims graphs improve root‑cause reasoning but require manual ontology work (Neo4j/NetworkX used).
AI-written summary · about
chapters (9)
- 0:00Explainable AI introduction
- 1:42Why explainable AI matters
- 3:53LIME explanations
- 7:29SHAP feature analysis
- 9:15Knowledge graphs in AI
- 11:04LIME vs SHAP vs graphs demo
- 16:43Limitations of explainability
- 18:08Scalable explainable AI design
- 21:17Q&A
related talks
35mWhy Do You Need an Explainable Ai?
41mThe Explainability Problem: Towards Understanding AI
33mExplainable AI in the LIME-light
24mWhy and How to Build Explainability into your ML Workflow
01h 26mThinking in Graphs: The Cognitive Foundations of Agentic AI
59mBetween the Layers– Interpreting Large Language Models
45mAI-404: Explanation Not Found
32mExplainable AI Explained for Developers