← Back to course
AI Explorers 6-8 / Week 07 / Friday
5/6
Week 07 · Three Kinds of Bias

Friday

Bias beyond the mission room
// When a fair-looking score hides a gap
⏱ about 20 min

Friday: Bias Beyond the Mission Room

Rocket is planning the club's next star party. "I asked my three best friends which night is best. They all said Friday!"

Raven tilts her head. "Your three best friends all have soccer on Saturday, though."

"Oh," Rocket says. "So my survey only heard from people with the same schedule."

Nova circles the planning sheet. "That is the same trap as the training box," she says.

"A small, similar sample can sound like everyone. Who else should you ask?"

Rocket picks up a pencil and starts a much longer list.

The same idea, everywhere

Any time people learn from a sample, they can miss cases the sample left out.

A survey of only your friends, or a sports team picked by watching one game, can lean one way.

AI learns from data too. Big Idea 5 of the AI4K12 Five Big Ideas says biased training data can leave some people less well served than others.

So people should talk about AI's effects and set rules for designing it fairly.

SPOT THE SAMPLE PROBLEM
  • Read the question.
  • Tap your answer.
A class picks a field trip by asking only the students who sit in the front row. What is the problem?
Which sample would better represent the whole club?
Biased training data can leave some people less well served. What should people do?

Week review

Bias in a group's habits and processes is ____ bias.
A sample that includes every kind of case fairly is ____.
Sky Sorter's comet score was how many right out of 2?
Week reviewTrue or false?
The three NIST kinds are systemic, computational and statistical, and human-cognitive.?
Checking accuracy for each group can reveal hidden bias.?
Bias always comes from someone wanting to be unfair.?
Real AI learns from data that people usually supply, so people's choices matter.?
WHY THIS EXERCISEPeople supply and choose the data, so their habits and samples shape what the model learns.
A FAIRNESS CHECK FOR SKY SORTER
  • Tap a card.
  • Then tap its spot.
1First
2Next
3Then
4Last
Rocket asked 3 friends. If he asks 12 more club members, how many people will he have asked? Type a number.
WHY THIS EXERCISE3 plus 12 is 15, and a bigger, more varied sample hears from more kinds of people.
On paper, draw three boxes labeled systemic, computational and statistical, and human-cognitive. Sketch one Sky Sorter example inside each.

You can now name three kinds of bias and hunt for each one. Tomorrow is your Explorer Quest.

← Thursday