Google is a company which is about scale. Hiring librarians is not a scalable solution to fixing metadata. Figure out a way to automate that fixing and I'm willing to bet Google will offer to hire you.

edit: As it happens I actually looked at working on a fairly similar problem for my undergraduate thesis (automated dating of works of unknown provenance) so if someone wants to talk about ideas for how to fix it feel free to email me.

I’m skeptical of this line of reasoning.

If we can’t hire librarians to label all those books, how did we ever get people to write, edit, bind, read, distribute, and finally scan them?

And if someone figures out how to automate the process, you'd be a fool to be "hired" by Google. That's something that should be paid for at a much higher rate than any salary.

I understand the importance of good data for academics (bad metadata on electronic book archives drove me crazy when I was researching etymology for the OED), but for most people it doesn't really matter.

The egregious metadata of Google Books is almost all date related problems. Anyone programmer who has done metadata-related work before will tell you that human compiled metadata always needs to be treated carefully, and date related metadata is usually "wrong" when the metadata comes from multiple sources.

Don't believe me? Crosswalk your data to dublin core and then tell me what date it was written, first published and the date the edition you have was published. I bet all those field go to dc.created and in standard dublin-core you have no way to identify which is which.

Now tell me why Google's data is so bad again?

The truth is that for subject-related data it is much, much better that library catalogs, for Author data it is pretty close and for date data the library catalog wins at the moment.

(Speaking as someone who spent a long time working in the field, and has had papers presented at the world library congress and at dublin core metadata conferences)