Introducing AI
How AI Learns
Show slide seven, or project Is This a Cat?
Pick a student volunteer to be the “AI robot.” Tell them the secret rule: everything that is orange is a cat, and anything that is not orange is not a cat.
Tell the other students that the AI has been told the first three images are all cats.
Ask them to guess:
- Will it say the black cat is a cat?
- Will it say the fox is a cat?
Have the “AI” volunteer give the answer: the black cat is not a cat, and the fox is a cat.
Show slide eight if you are using the slides.
Ask:
- Is that the right answer? (No. A black cat is a cat, and a fox is not a cat.)
- Would a person give that answer? (Probably not! After seeing three cats, you would almost certainly be able to tell the difference between a cat and a fox.)
Show slide nine if you are using the slides.
Ask:
- Why do you think the “robot” said that?
- What rule was it following?
Let students discuss for a few minutes. Don’t try to reach any sort of consensus or definition yet.
Show slide ten if you are using the slides.
Tell students that the reason AIs seem like they “think” and “decide” is because they look for patterns in what their makers show them. They test those patterns and the ones that work become rules.
Let the “AI robot” student explain what the rule was:
Everything that is orange is a cat, and anything that is not orange is not a cat.
Show slide eleven if you are using the slides.
Ask students:
- How did it make this rule? (All of the cats it was shown were orange.)
- How would you change this rule so that it would recognize that black cats were cats, and that foxes were not cats?
Let students discuss this for a few minutes. Make sure to point out that you can correct an AI by telling it what answers are right or wrong.
Explain that you can do this in two ways: by having its makers test and correct it before people start using it, or by having the AI correct itself each time it’s wrong. (Most AIs do both!)
Show slide twelve if you are using the slides, or take out the bouncing balls.
Tell the student volunteer the secret rule: everything that is round bounces.
Ask students: If the first three items all bounce, will the AI say that the last one will bounce?
Have the “AI” volunteer give the answer: the last ball will bounce.
Show slide thirteen if you are using the slides, or take out one of the round fruit.
Ask students: Will the AI say that the fruit will bounce?
Have the “AI” volunteer give the answer: the fruit will bounce.
Show slide fourteen if you are using the slides.
Ask students:
- What pattern did the AI spot?
- What rule did it make?
- Why did it give the right answer once?
Why did it give the wrong answer once?
Let the “AI robot” student explain what the rule was:
Everything that is round bounces.
Explain that because that we don’t always know when AI makes a mistake, because (as in this case) it may give the right answer for the reason.
Show slide fifteen if you are using the slides.
Ask students: If you showed an AI a basketball, a volleyball, and an orange, would it guess that a soccer ball would bounce? Would it guess that an apple would bounce?
Let students discuss this for a few minutes. Make sure to point out that you can make an AI give better answers by showing it more examples so it doesn’t make rules that are too narrow (for instance, showing it cats that are not all orange, or that not all round things bounce.)
Explain that you can do this in two ways: by showing the AI more examples before people start using it, or by having the AI use every right or wrong guess as a new example. (Most AIs do both!)