Doing so would be wildly inefficient. You need your passport to check in for your flight. When boarding the plane, you don't need to check someone's face against the entire passport database, only against the passports which have been checked into the flight, which is particularly important in terminals with shared domestic/international departures where entry is possible without a passport and the passport can be transferred to somebody after check-in.
But the main reason why it's anyway infeasible is because flights are rarely full of people of a single nationality; scanning peoples' faces against a wide-ranging passport photo database effectively means scanning their faces against some kind of global database of the faces of everybody from around the world. Assembling a database like that, apart from the ethical issues, is politically infeasible.
Remember - this system doesn't advocate for people to not need their passports at all when traveling. They still need them - just not after check-in. The question is whether the system is more or less accurate than a human being at matching passport photos to faces, and based on what we know about racial bias, the answer to that is probably that facial recognition is more accurate by now.
>> The new veriScan system developed by the Metropolitan Washington Airports Authority—with guidance from U.S. Customs and Border Protection—scans the faces of travelers approaching the gate. The system then compares the photo to a gallery that includes images of that person—either their passport photo for U.S. citizens or the photo taken of foreign nationals when they entered the country.
So that's definitely not just the passports that have been checked into the flight- foreign nationals in particular may have entered the country at any time and place other than before boarding a flight at Dulles.
Which is why there are privacy concerns in the first place- because it can be used to track the movment of people across multiple trips.
you don't need to check someone's face against the entire
passport database, only against the passports which have
been checked into the flight
Sadly the gates are still slow and unreliable, even at that seemingly simple task.
* Admittedly for immigration rather than boarding [1] https://en.wikipedia.org/wiki/EPassport_gates
>> Presumably it's 99% effective at matching random faces to the photos
I think what they mean is that they carried out laboratory tests where a test subject stood facing the system's camera and the system tried to match their face to a database of pictures of peoples' faces, then decided whether the subject's face matched one in its database.
There is a lot that one can do to make this kind of testing give much better results than it should (e.g. use a small database, rather than the entire US passport photo database plus photos of visiting foreigners). This does not presuppose malice- a bit of sloppiness or choices dictated by budget restriction are plenty enough to "spike" the results.
On the other hand, even assuming 100% accurate evaluation one can expect a machine vision system's accuracy to fall way down when it's deployed in the real world, where conditions can no longer be as strictly controlled as in the lab.
So as a rule of thumb, we can probably assume that a 99% accuracy in the lab translates to ~75% accuracy in the wild.