Lesson 2 of 7
Five kinds of bias
Show on a big screenUnfair results can start in five different ways.
The rule
Bias has five common kinds: data, historical, measurement, algorithmic and human.
Experts group bias in different ways. These five are common.
| Kind | Ask | Example |
|---|---|---|
| Data bias | Who is missing from the data? | A face app trained on one age group |
| Historical bias | Is the data copying unfair choices from the past? | A club program trained on unfair old lists |
| Measurement bias | Is the wrong thing being measured? | Judging a teacher by how fast papers come back |
| Algorithmic bias | Does the design weigh one factor too much? | A prize program that gives huge weight to who replies first |
| Human bias | Did the makers' likings steer the choices? | Picking a program's examples only from topics its makers like |
Watch out!
Kinds can overlap. One unfair result can have more than one cause.
The last two rows use four words. An algorithm is the step-by-step rule a program follows. A factor is one thing a program looks at. Weight is how much a factor counts. To steer is to push choices one way.
Read each program. Name the kind.
Loading lab…
Try the lab to finish this lesson.