The second one. It's important to have a clear picture of biases and how your model could fail to generalize. In the case of social data from a deeply biased so

The second one. It's important to have a clear picture of biases and how your model could fail to generalize. In the case of social data from a deeply biased society, these issues can be seen a mile away.