How machine learning works
- Collect examplesMillions of pictures, sentences or measurements
- Train a modelAdjust billions of numbers to reduce mistakes
- Test it on new dataCheck it works on things it hasn't seen
- Use itTo predict, recognize, write or recommend
Not magic
AI finds patterns in data. It can be confidently wrong, so people still need to check its work.
Milestones
| Year | Event |
|---|---|
| 1956 | The term 'artificial intelligence' is coined at Dartmouth |
| 1997 | IBM's Deep Blue beats chess champion Garry Kasparov |
| 2016 | DeepMind's AlphaGo beats a top Go champion |
| 2020s | Chat assistants and AI coding tools reach millions of people |
Progress has sped up as computers and data have grown.
Strengths and open questions
Good at
- Finding patterns in huge data
- Drafting text and code quickly
- Speeding up research
Open questions
- Effects on jobs and work
- Accuracy, bias and trust
- Who decides how it's used
Sources: Stanford AI Index; historical records
Remember
AI is a powerful pattern-finding tool; how people choose to use and check it will shape its effects.
Tap or press Enter to flip
Answer key
- What does machine learning learn from?Examples in dataModels are trained on large sets of examples.
- When was the term 'artificial intelligence' coined?1956At a 1956 workshop at Dartmouth.
- Which program beat a world chess champion in 1997?Deep BlueIBM's Deep Blue beat Garry Kasparov.
- Why should people check AI's work?AI can be confidently wrongAI can produce convincing but incorrect answers.
- Which is an open question about AI?How it will affect jobsIts effect on work is widely debated.