Google's new data challenge targets the geographic and cultural skew in the training datasets that underlie so much of the machine vision world today and seeks new approaches to training deep learning systems on highly biased datasets without inheriting their biases.