Joy Buolamwini is showing how algorithms can be unfair and how she’s fixing that problem, like when a robot gets confused by different faces.
Imagine you have a special robot friend who can recognize people just by looking at them. But one day, the robot starts missing some faces, especially if they're darker-skinned or women with head coverings. That's because the robot was trained mostly on lighter-skinned faces, like the ones it saw most often. It’s like learning to count only using your right hand, you might not notice that your left hand is just as good!
The robot needs more practice
Joy found out this problem and decided to help the robot learn better. She added more pictures of different people into the training set, so the robot could see all kinds of faces, like how a child learns to count with both hands. This made the robot much smarter and fairer.
Fixing unfair robots
Joy is helping make sure that algorithms don’t favor one group over another. It’s like making sure everyone gets equal time on a playground, no one gets left out just because of how they look!
Examples
- A facial recognition app fails to recognize dark-skinned faces because it was trained mostly on light-skinned faces.
- An algorithm used for hiring picks candidates who look like previous successful hires, ignoring diversity.
- A recommendation system always suggests the same kind of movies, even if you've watched different genres.
Ask a question
See also
- What are powered by machine learning models?
- Can You Tell When A Video Is Fake?
- What are smart helpers?
- Who Made That Decision: You or an Algorithm?
- Who is Expected SARSA?