IBM releases Diversity in Faces, a dataset with 1M+ annotations to reduce bias

Things like this might be increasingly important going forward. In the long run AI needs to become more intelligent so that it doesn’t need big data to learn. Using large data sets tend to introduce biases more easily.