How to Erase Unconscious Bias From Your AI Datasets
summary
Facial-recognition example HowOld.net mislabels a recent photo as 38 then 24, showing dataset fragility; criminal-risk tool COMPAS is found by ProPublica to be 2× more likely to falsely label Black defendants high risk and white defendants low risk. Recommendation: integrate continuous bias testing into product lifecycles, train on demographically diverse datasets, and assign a dedicated data scientist to monitor retraining so models don’t silently amplify biased inputs.
AI-written summary · about
chapters (4)
- 0:00Start
- 0:24Facial recognition software
- 1:26Machine bias can hurt users
- 3:13Add bias testing to product dev cycles
related talks
45mBias in Artificial Intelligence
31mAvoiding Digital Bias
21mFairness and bias in AI
35mZero Bias AI
47mEthics & AI: Identifying & preventing bias in predictive models
38mSocial Implications of Bias in Machine Learning
31mDoes This Algorithm Have a Soul?
38mSocial Implications of Bias in Machine Learning