I think this will be the one thing that causes Google to revive its plan to scan all books in existence. They had started it, and built the machines to do it, and were making good progress... until Copyright hit them. BUT if they're not making the full text publicly accessible, and are "only" training AI on it, who knows if that would still be a problem. It's definitely a vast treasure trove of information, often with citations, and (presumably) hyper-linkable sources.
