If you're working on any kind of software that utilizes facial recognition — a secure Face ID camera system for a phone is just one example that comes to mind — you need a good amount of data in order to train the AI that powers it. Google isn't exactly new to the data collection game, but you might be surprised to learn that it's been doing so via such an old-fashioned method.
As reported by ZDNet, Google employees have been observed stopping passers-by in the street and asking if they can have their faces. More specifically, if they can take photos of their faces to use the data. A friend of the writer Chris Matyszczyk quoted the Googler saying something along the lines of "Hi, I work for Google and we're collecting data to improve the next generation of facial recognition phone unlocking." He was then tasked with holding a phone in selfie mode in front of his face and moving it around to analyze every feature of his appearance before being offered a $5 Amazon or Starbucks gift card for his troubles. Not a Google Play discount, though, curiously. Even though it's undoubtedly a little strange, this is about as up-front as Google gets when it comes to data collection.
This took place in New York City, but Google apparently has other teams in many different cities doing similar work. Sure enough, one of our own commenters (Frisi) detailed an interaction with Google data collectors in Miami recently that also involved actions such as taking a modified Pixel 3 XL out of a pocket and looking at notifications while also capturing face data. A "future Google product" was as much of an explanation as they got, but it's clear Google is building up to something big.
The most recent Pixel 4 leak comprised probable screen protectors showing expected cutouts for two front camera and ambient/proximity sensors, as well as a larger cutout of as yet unknown utility. Put this together with the evidence of native face authentication functionality in Android Q, as well as the Project Soli movement-mapping tech, and it doesn't seem much of a stretch to speculate that Google is working on a secure Face-ID feature to be debuted with the upcoming fourth-generation Pixels. That would be the preferred justification for the anticipated large top-bezel on Pixel 4, at least, but we'll have to wait and see what happens.
Google describes why it needs this data, and how it will protect volunteer privacy
Earlier today, Google formally revealed its face recognition and gesture tech to the world, and it appears that the company also followed-up with The Verge regarding "buying" face data from folks on the street. The purpose behind the test was to ensure that it didn't accidentally work better for some people than others, casting the widest possible net "to ensure that the Pixel 4 works with a diverse set of faces." Google later added, "our goal is to build the feature with robust security and performance. We’re also building it with inclusiveness in mind, so as many people as possible can benefit." That should all help to reduce any unintentional bias on the part of the system — a very real problem when tests aren't extensive enough.
When it comes to the data that the company has already gathered from the compensated volunteers, Google is allowing them to decide what ultimately happens to it:
Although face samples inherently can’t be anonymous, each participant is assigned an abstract identity number. We separately keep each participant’s email address, in order to remove data upon request.
Although participants previously agreed to allow Google to keep their face data for five years, Google is also cutting that time down to just 18 months. The information Google has gathered from volunteers in that time won't be associated with any other identifiable details about the individual, and it is stored encrypted and access restricted. When it comes out, the Pixel 4 won't need to send your data to Google for its facial recognition to work, this all merely applies to those that participated in Google's public testing.