[sf_rob]

Publishing something does not make it public domain. I am not a lawyer, but saying they can do anything with what they scrape seems questionable.

[chatmasta]

I mean in practice they can get away with it. As of now, there are no audit trails required for training of machine learning algorithms.

[ucaetano]

I don't think that there's an issue with someone reading 100 reviews of a certain restaurant on the internet, and summarizing them by writing a new review:

"Most people who talked about it mentioned the great salad bar, but said that the fish was sub par. They commented on it being lively and child-friendly"

That's what I'd guess they are doing.