Actually having my name on a few patents I feel like I can speak with a bit more authority. Yes, companies file patents all the time on tech they have no actual plans to implement in production. They may not even have a proof of concept. This happens for all sorts of reasons.
Part of the reason is that companies would rather err on the side of obtaining a patent on something patentable rather than miss it & get sued by someone else. Another reason is that it can easily turn out that there are significant technical challenges in productizing an idea but you don't find out until well after you've filed the patent.
There's also a defensive aspect. For patent trolls you want to beat them to filing the patent so you have a far easier legal defence - America was first to file until very recently so that was also a significant incentive in filing defensive patents. For competitors you want to have the patent so that if it turns out a successful product by another company relies on your product heavily you can use it offensively or defensively in case that competitor comes after you (traditionally the bigger Silicon Valley companies have preferred to stick to defensive use of patents against competitors but that isn't always the case).
Finally there's a financial incentive to this from the employee side. Patents bring you bonuses and prestige. That means you have an incentive to push even questionable patents through the process. The patent filing process is completely divorced from marketing AFAIK (haven't sat on any review boards) so no one ever considers the negative PR risk from articles like this (probably rightly so).
> And if you don't think such a thing would be useful to a company like this (if it could really be made to work) you aren't using your imagination properly.
Of course it would be useful. However realistically the engineer in me thinks that when deployed at scale suddenly it can become a lot less useful than other more straightforward methods; both in terms of compute & memory required as well as accuracy. There are cheaper more accurate ways to solve this so why bother wth something so complex/expensive that carries significant PR risk? In fact, FB owning & not using this patent means they can win PR brownie points suing companies that do attempt to use this technique.