Bias in Autonomous Vehicle Object Detection
Heard lighter skinned folks are 5% more likely to be detected by autonomous vehicles than darker skinned folks and it made me think about the cars I see driving around Mountain View created by Google X. They are putting millions of hours into road time to build up training data for their detection software, but if they’re doing that in an area of predominantly lighter skinned people, the training data skews that way. Figuring that part of the autonomous vehicle training data bias solution is running a significantly higher amount of road tests in places with darker skinned people. Feels weird saying that but you need the training data to remove the bias.