Definition
From the 1990s onward, AI's centre shifted from hand-written rules to statistical machine learning, which learns from data. Better algorithms, growing data from the internet and cheaper computing made this possible. Milestones include IBM's Deep Blue beating Garry Kasparov at chess in 1997, and IBM Watson winning Jeopardy! in 2011.
In simple words
Instead of asking experts for rules, people asked: what if we just count what happens in millions of examples? It turned out that simple methods with lots of data often beat clever methods with little.
Why it matters
It's the shift that made modern AI possible, and a useful habit of mind: data quality and quantity often matter more than a fancier algorithm.
How it works
| Year | Milestone |
|---|---|
| 1990s | Support vector machines and statistical methods spread |
| 1997 | Deep Blue beats the world chess champion |
| 2001 | Random forests are published |
| 2006 | Hinton's team revives deep networks |
| 2011 | IBM Watson wins Jeopardy! |
Common mistakes
- Treating chess engines as proof of general intelligence.
- Assuming more data always wins. Biased or irrelevant data doesn't help.
Check yourself
Summary
- AI moved from rules to learning from data.
- Data and computing drove the change.
Finished reading?
Sources: Russell & Norvig, Artificial Intelligence: A Modern Approach, ch. 1