Lesson 2 · AI Explorer
Thinking critically about output.
AI learns patterns from human-made data. That includes the unfair patterns.
If most "doctor" images in the training data showed men, the system learns that association and repeats it. It isn't deciding anything about doctors — it's reflecting what it was shown.
That's why an image generator asked for "a nurse" and "a surgeon" often produces predictably different people. Nobody programmed that. It came from the data, which came from us.
You met this two chapters ago in Python, and you probably didn't notice.
Your eligibility checker had a cutoff — ₹300,000 income, 75% attendance. The challenge asked who that treated unfairly, and you found the cliff edge where two rupees decided everything.
That unfairness was written by a person in one visible line. You could point at it and change it.
AI bias is the same problem with a crucial difference: it's spread across millions of numbers with no line to point at. Same kind of harm, far harder to find and fix.
5 more parts in this lesson
Walkthroughs, code labs, mini-games and the checkpoint quiz unlock when you buy the course.