Week 6 · Unit 1, Lesson 4 Big Data and Its Relationship to Artificial Intelligence · 3 lessons
3
How AI feeds on data
ICT Prep 2 Term 1, Unit 1, Lesson 4: pp. 35–39
GoalExplain why AI needs many good examples and what its limits are.
New words
training examples · the data a model learns frommodel · the program that has learned patterns from datasensitive data · private data, such as faces or medical facts, that must be protectedautomation · letting a machine do a routine task with less manual work
Explanation
Think of teaching a classmate to sort cards. If you show only two cards, he will guess. If you show many cards of different shapes and colours, he learns the rule better. Most modern AI that learns from examples works this way: more good and varied examples usually help, but poor examples give poor answers. Language AI learns from a lot of text, vision AI from many pictures, and prediction from past records. AI can also find hidden patterns, learn from new data and do routine tasks with less manual work. There are limits. A model that only memorises its examples fails on new ones. And some data is sensitive, such as faces or medical facts: a grown-up and rules must protect it, so we use invented examples only.
AI task
Examples it learns from
language (understanding text)
a lot of text
vision (recognising pictures)
many pictures
prediction
past records
finding hidden patterns
big tables of data
Worked examples
Youssef teaches a classmate to sort cards into “circles” and “squares”. After three cards, the classmate guesses wrongly on a new card. After thirty varied cards, he gets it right. The more varied the examples, the better the learning.
A clinic’s appointment list shows an invented pattern: more visits on Mondays. The AI can find this pattern, but no real patient data is needed to understand the idea. Real medical data is sensitive and protected by rules.
Class activity
Design your own card-sorting game for a family member: make twelve cards with two kinds, hide the rule, and let them learn by seeing the cards one at a time. Write what happened after three cards and after twelve.
Practice
1
Why does AI usually learn better from more examples?
More good, varied examples show the pattern more clearly.
2
What does poor data give?
Poor answers.
3
In the card game, why did the classmate guess wrongly after only three cards?
Three cards are too few examples to show the rule.
4
Why do we not use real faces or medical data in lessons?
They are sensitive data; grown-ups and rules protect them.
