I am wondering where Google uses the result from deep learning of reCaptcha (how can a system that knows to recognize street signs is useful somewhere? how they profit from it?)
At this time, Google is aggressively pursuing research into AI of many sorts. All AI at this point is constrained by the size and accuracy of available training data. Getting human curated data is expensive and time consuming. Crowdsourcing through Captcha gets access to that without paying directly.
Why roads specifically? It can hardly be a coincidence that Google's holding company also owns a self driving car company Waymo. They need machines to learn what road signs look like in real life, so that they can respond to them.
On the first look, a captcha is equal to small OCR problem which has to be solved by a human. Apart from this frontend, a captcha tool will produce a statistics of the user. The statistics can be combined with existing ip-logfiles to get a more detailed picture of the users who have send requests to the website.
From the perspective of a search engine, the world is a social network which has to be investigated. It contains of billions of computers and humans in the real domain. Captcha tools can be combined with crawlers into a powerful user-profile generators. The purpose is to De-anonymize users in the Internet by asking them if they are humans or not.
 Bilge, Leyla, et al. "All your contacts are belong to us: automated identity theft attacks on social networks." Proceedings of the 18th international conference on World wide web. ACM, 2009.