Machine Learning
Learning from examples instead of rules
The story: You can teach a child to recognize a dog by writing rules: "four legs, fur, a tail". Then a cat walks by and breaks every rule. What actually works is showing the child lots of dogs and lots of not-dogs. The more they see, the better they get.
In AI/AWS terms: ML models learn patterns from data instead of following hard-coded rules, and they improve as they see more data.
For the exam: ML learns from data and improves with more of it. Traditional programs follow rules someone wrote.
When ML is a good fit
The story: Try writing down exact rules for "is this email spam?" Suspicious words, odd senders, too many links, strange timing: they overlap in endless ways and change every week. And even if you could judge by eye, you couldn't read a million emails a day.
In AI/AWS terms: Use ML when:
- The rules are too hard to code. Spam filtering depends on many overlapping variables that people can't write down as rules.
- The scale is too big for people. A person can review a few hundred emails, not millions.
For the exam: ML fits when rules are too complex to write by hand, or the volume is too big for people.
When ML is not needed
The story: You don't hire a fortune teller to add up a shopping bill. The math is known, and a calculator gets it right every time.
In AI/AWS terms: If the answer comes from simple rules, computations, or predetermined steps, program it directly. ML would add cost and uncertainty with no benefit when the logic is already known.
For the exam: If simple rules or a calculation give the answer, ML is not the right choice.