Bias Detective
Find out when an AI turns unfair, and why.
What you'll learn in this level
- How you really fix bias in an AI
- Why each cause needs a different remedy
- Why more measures aren't automatically better
Bias Detective
Spotting bias is the first step. The second is harder: how do you repair it? Every cause has different remedies — and none of them works against everything.
In this level you examine three systems, name the causes and pick suitable measures. The rule: targeted, not as many as possible.
How it works
- 1
Look at the data
First you check: who do the data show? And who is missing from them?
- 2
Name the causes
Then you say which kind of bias is here. Bias means: the AI treats one group worse.
- 3
Choose measures
Each cause has a different matching remedy.
- 4
Check the effect
At the end you check: has it really got fairer?
Guess first, then experiment
You are about to tick which causes explain an AI's bias. Does it help to just tick them all to be safe?
Scenarios 1: Image Recognition Bias
Data Analysis
An AI system identifies people in photos but works better for some groups.
[
{
"group": "Group A",
"accuracy": "98%"
},
{
"group": "Group B",
"accuracy": "65%"
},
{
"group": "Group C",
"accuracy": "95%"
}
]Tip
Look closely at the data: who or what does it represent — and who might be missing?