கKatralAI

Little Katral · Lesson 3 · 8 min

Fair and unfair machines

In one sentence

If the examples are unfair, the machine learns the unfairness — and repeats it faster.

Kavi

Kavi

Kids mentor

Shall we see how a machine learns — like training a peacock?

Read the idea, watch the picture, then try it with your hands.

Watch the idea

The concept

Tilted set

Team A · 20 photos · Team B · 2

Fixed set

Both teams visible · the call gets fair

Unfair examples, unfair guesses

Bias is not a feeling inside the computer. It is a tilt in the examples. If a hiring helper only saw one kind of resume, it may treat other good people unfairly — faster than a human would.

Who is missing?

The first question about any dataset is: who is not in the pictures? Missing people become missing fairness. A fair teacher looks at the pile before trusting the guess.

Fix the set, then the model

You cannot scold a model into being fair. You change what it sees. More kinds of examples, a second look from a human, and a rule that says stop when unsure — that is how fairness is built.

Worked example

The one-team referee

A match where the referee watched only Team A in practice will keep calling Team A the winner. Show practice film of both teams. The whistle gets fairer. Data is the practice film.

Picture to keep

Who is in the data?

  1. 1Look at examples
  2. 2Find who is missing
  3. 3Fix the set