Fairness and bias in AI
summary
AI models inherit problems from training data: historical bias, representation bias, measurement bias. Examples include COMPAS showing twice the false positive rate for Black defendants and Apple Card assigning women roughly 20 times lower credit limits despite identical inputs. Practical fixes target data and models: rebalance datasets, stop using poor proxy labels, add fairness constraints to learning objectives, and calibrate decision thresholds per subgroup so metrics (false positives, true positives) line up across groups.
AI-written summary · about
chapters (7)
- 0:00The impact and promise of AI
- 0:36How AI models work
- 2:31Examples of AI bias
- 5:14Sources of data bias
- 9:37Defining and measuring fairness
- 13:53Strategies for fair AI
- 17:13Conclusion and next steps
related talks
56mAn AI with an Agenda: How Our Cognitive Biases Leak Into Machine Learning
34m(Un)Ethical Machines? Why AI Bias Is a Problem and What to Do About It
35mZero Bias AI
35mBuilding Inclusive Machine Learning Models
27mCrouching Harm, Hidden Bias—Addressing issues of fairness....
26m(Un)ethical Artificial Intelligence: How to Keep the AI Fair for Everyone
38mSocial Implications of Bias in Machine Learning
45mBias in Artificial Intelligence