Google 411 was a project which offered 'directory services' by voice over your phone. It was pretty cool, you could call its 800 number, ask for a listing, and it would use Google search to find it and then read it back to you. Then you got to tell it how well it did. People started using it, adoption spread, and then they cut it off.

So why did Google run this service in the first place? Was it to see if they could make a business out of directory services? No, it was to collect a data set where a spoken phrase could be matched to its exact translation (the thing they were looking for) which could then be processed and re-processed to train algorithms for independent speaker recognition. On the one hand they could try to pay a million people to come in and say something and then confirm or deny they understood what they said, or they could use a bit of spare hardware and collect that information with out paying anyone a dime. They own that data set, it is extremely valuable for testing improvements in voice recognition, and it is yet another barrier to a new company trying to get into that space.

Now let's look at Google Books. The 'story' was digitize all the books and make a great library available, and maybe even offer up PDF copies of out of print books for people. It set off a legal firestorm (as mentioned in the article) it got all of the archivists on board and libraries contributed millions of volumes, and Google got a ruling out of the Supreme Court that said that digitizing a book was not an act of copyright infringement.

But the subtle part is that while folks say 'everything is on the web' (and that may be true) literally 99.99% of everything on the web is complete and utter crap, written by people who are seeking to game advertising selection engines not digitize information. Most of the stuff in books is not crap, because it cost someone significant time and labor to take that information and publish it. (romance novels excluded :-)).

Google had digitized the single largest collection of human knowledge ever, and put it in a form where 100's of thousands of machines can process and re-process it to derive ontological accurate facts into the largest knowledge base in history. If you want to test whether your algorithm can identify credible information, there is no better way to do it than to prime it with a ground truth which has much higher credibility than most of the accessible data out there.

That data set exists, they own it, and they don't have any obligation to share it with anyone. And it will allow for the creation of trained models for differentiating fact from fiction, jest for insult, and command from comment.

It is my opinion that any company that wants to be 'serious' about AI better have access to an equivalent data set or they will lose.