Forget about AI and do mathematical modeling instead!
summary
Argues that much AI is a hype bubble and gives concrete counterexamples: over 2,000 COVID projects deliver little value, recommender teams spend millions yet often do worse than matrix factorization, and a production model for nuclear plant efficiency simply rediscovers the second law of thermodynamics. Shows an IoT prototype "inner Cube" where height is solved reliably with a free-fall equation instead of a neural net, and explains packaging optimization as a classic bin packing problem best solved by heuristics, not deep learning. Recommends math when data are sparse, interpretable, or feedback-driven.
AI-written summary · about
chapters (11)
- 0:00Intro
- 1:13The wonderful World of Al
- 6:07Predicting the Efficiency of Nuclear Power Plants
- 7:39Size Prediction for Packaging
- 9:53Covid-19: Inference vs Prediction
- 11:19Need for Interpretable Models
- 15:32Interpolation vs Extrapolation
- 16:12Problems with Al/Deep Learning
- 21:12Free Fall!
- 27:36There is no Black & White, it's a Spectrum
- 29:57Simplified Mathematical Modelling Cycle
related talks
46mAI and ML, separating hype from what actually matters
46mRub a Little AI on It
01h 51mAI and Social Entrepreneurship
26mWhy Machine Learning Algorithms Fall Short...
16mAI: Common Misperceptions and Uncommon Perceptions
44mAI Engineering Anti-patterns
26mHow to develop useful and unbiased AI
41mMachine Learning can't do the thinking