You can imagine that the second the phones hit the market, security researchers (and government agencies) will start testing ways to fool Face ID and that some will have limited success—which, for cases disclosed to Apple, will improve deterrence from those workarounds.
Can Face ID be set up so that multiple people’s faces unlock the same iPhone?
For now, an iPhone X will recognize only a single face. That could change in the future. But you can no longer give a spouse, partner, or other person access to your phone through a biometric means, like you could by enrolling one of their fingers with Touch ID. You’ll need to share a password with them.
Will Face ID recognize people of color’s faces as well as it does white people’s faces?
We hope Apple has learned from machine-learning and body-recognition debacles at other companies that have led to people of color not having the same accuracy of automatic photo tagging, facial recognition, and other problems. (See the video of a man unable to get a hand drier seemingly to recognize his hand color.)
The kind of machine learning used widely now for voice, image, and other recognition relies on training databases. Companies or academic projects have to find often hundreds of millions to s of examples that they can mark correctly to feed into a deep-learning system to have it develop the pathways that let it recognize features more generically instead of as exact matches.
In the past, these training databases have apparently been heavily biased towards white faces and often towards men, leading to racially insensitive and upsetting results. Apple VP Phil Schiller said in the keynote that Apple used a billion images to train Face ID, but not which faces.
In its announcement and on its website, Apple features a number of people of color more in proportion with the global population than American or European ones in the Face ID and other TrueDepth sections, as well as showing heavily freckled faces and women with elaborate and enormous hair.
Apple’s senior VP of software engineer, Craig Federighi, later expanded on this to TechCrunch, explaining, “We’d done data gathering around the globe to make sure that we had broad geographic and ethnic data sets.” In Apple’s white paper, the company notes it created “a representative group of people accounting for gender, age, ethnicity, and other factors.”
Apple also keeps Face ID enrollment strictly on device, which means it can’t learn directly from real-world usage about how well its algorithm performs for given individuals’ faces. This is great from a privacy standpoint, but might lead to awkward results. (Federighi confirmed to TechCrunch that customers’ use of Face ID won’t in any way be folded into its training. “We do not gather customer data when you enroll in Face ID, it stays on your device, we do not send it to the cloud for training data,” he told the news outlet.)
Sign up for Computerworld eNewsletters.