Pedro Domingos, Professor, University of Washington
summary
Argues that five concrete algorithms — inverse deduction (symbolic induction), backpropagation (deep nets), genetic programming, Bayesian inference, and support vector machines — each solve a different, necessary subproblem and must be combined. Describes Markov logic networks as a working hybrid representation, mentions two lab robots Adam and Eve where Eve discovers a malaria drug, and cites the billion‑parameter Google cat deep net; proposes mixing genetic search for structure with gradient methods for weights toward a single Master Algorithm.
AI-written summary · about
chapters (24)
- 0:00Intro
- 0:21Where does knowledge come from
- 1:46Machine learning
- 2:56How computers discover new knowledge
- 7:08Symbol lists
- 7:21Induction
- 9:37Illustration
- 11:09Connectionism
- 11:52Brain
- 14:00Error driven learning
- 16:05Backprop
- 16:59Evolution
- 20:09Evolutionary Robotics
- 21:09Visions
- 24:45Spam Filters
- 25:23Analogies
- 26:19Analogical Reasoning
- 29:23Recommender Systems
- 30:34Machine Learning Tribes
- 32:41Representation
- 36:48Key Ideas
- 37:33Benefits
- 42:27Quality of life
- 43:32The Master Algorithm
related talks
57mDarwinian Machine Learning: Principles of Machine Learning in Evo-devo, Evo-eco and Evolutionary...
32mHow Neuroscience Makes AI Intelligent
46mCode + AI: Will Robots Take Our Coding Jobs? ML Applied to Programming
41mArtificial Intelligence: Let's Think About Learning
43mAI Systems: Foundations for Artificial Minds
01h 00mBetter Learning through Programming Languages
29mDeep Learning, Neuroscience and the future of AI
42mReasonable AI