Also, what does this section even mean in the context of ML models? How do you communicate that you trained an interest-estimation model on more data and will now be using the upgraded version? I’m sure Google uses many such models for producing personalized results, and probably several are updated in production every day. Does Google have to share the inner workings of their search stack with these news companies? Do they have to share the models? Does Google have to explain to these news companies how the newly trained model might treat their content differently than the previous iteration? Any of these things I expect would cause Google to throw up their hands and abandon the country. Exposing a live feed of all their trade secrets to the employees of random news companies is obviously not going to happen, and explaining why ML models do what they do is an active research area.