AI: A Return to Meaning
summary
Describes building Watson as a hybrid system that combines hundreds of shallow NLP algorithms whose scores are learned weights via machine learning, generates up to 10,000 answer–passage pairs per clue, and scales to 2,880 cores, 15 TB RAM to answer Jeopardy questions in ~2–4 seconds and win against top players. Shows WatsonPaths chaining multi‑step, passage‑based inferences without hand‑coded if‑then rules, and argues progress requires bootstrapping data, logic and human cognition so machines can predict and also explain.
AI-written summary · about
chapters (10)
- 0:00Introduction to AI journey
- 1:03Theory-driven AI origins
- 6:42Understanding vs doing in AI
- 17:59Theory-driven AI explained
- 25:23Data-driven AI approach
- 31:25Personal story on logic vs stats
- 35:44IBM Watson’s sweet spot
- 51:01Watson’s Jeopardy win
- 59:05Future of AI understanding
- 1:04:43Q&A
