AI UNIT 3 • STAGE 1 OF 5
It isn't the machine's opinion. It's inherited from people and data.
When people hear "the AI is biased," they sometimes picture a machine with an attitude. That's not it. In Unit 2 you saw where it really comes from: a machine copies its examples, and if the examples lean, the machine leans. Bias is a skew the machine inherited, not an opinion it holds.
If bias were just the machine "being mean," you'd fix it by scolding the machine. Because it comes from people and data, you fix it by changing who's included and who decides. That puts the responsibility back on humans, where it belongs.
Not every wrong answer is bias. A one-time wrong date is just an error. Bias is a pattern, a consistent lean in one direction, across many answers. Ask the AI a few related questions and look for the lean.
One odd answer is an error. The same slant showing up again and again is bias.
As you read the answers, ask: is this a specific fact that's wrong (an error), or a whole shape that leans the same way every time (bias)? Learning to tell them apart is the core skill of this unit.
Errors are random. Bias is systematic. When an AI describes Native people the same narrow way over and over, that repetition is the tell.
In your Field Notes, write your own definition of bias in a machine, note one error and one pattern you saw, and say how you told them apart.
Over the next stages you'll gather evidence of two specific harms. First up in Stage 2: erasure, the harm of being left out.
Start your case file:
Saved automatically.