I have to criticize this article for confusing the underlying software for building machine learning models vs. the models themselves. Open Source contributors to PyTorch, TensorFlow, etc. are unlikely to affect bias in data (and thus in trained models). Can anyone here think of reasons why this is not so?