Computing Bias

Part of: Impact of Computing

Bias Lives in Computing Systems. Computing innovations can reflect existing human biases because people build them and people supply their data. A biased system produces systematically unfair outcomes for some groups. The exam calls a key point out clearly: bias is not always intentional : it often slips in unnoticed. Where Bias Enters. - Biased training data : a hiring model trained mostly on resumes from one group learns to favor that group. - Biased design choices : deciding which features matter encodes the designer's assumptions. - Biased use : even a fair tool can be applied unfairly by its users. Detecting Bias. A common technique is comparing outcome rates across groups . If a fair process should approve people at similar rates, a large difference is a red flag worth investigating. Reducing Bias. - Audit datasets for representation before training. - Test outputs across groups after deployment. - Include diverse perspectives on the design team. Note that detecting a gap is not proof of bias by itself, real differences can exist, but unexplained, large gaps demand scrutiny. What usually goes wrong. What you see What caused it How to fix it --- --- --- You say a system cannot

Challenge: Approval Rate Gap