Artificial Intelligence Has a Facial Expression. It Ain’t Pretty 😝

Sept. 11, 2019

First, let’s establish some definitions:  facial recognition (FR) as being tested at some airports is not the same as artificial intelligence (AI).  Indeed, FR is not an ‘airport’ process at all; it is being tested by some airlines (not all) and a few TSA locations (not all) as an alternate means of identification to help speed up screening checkpoint throughput and as a substitute for the boarding pass.  FR certainly has some operational constraints and TSA security implications by requiring essentially real-time connectivity to a wide range of both government and industry data bases to first verify each travelers’ identity, and then determine whether that individual appears on any of dozens of those data bases as an undesirable/ unallowable traveler. 

Two immediately obvious difficulties: (a) over 2.7 million US  passengers every day, each of whom has the choice to opt out of FR for privacy – or any other reason, and (b) until there is nearly 100% airline participation and about 100% reliability of connectivity  and access from all of TSA’s commercial airport locations to all the data bases, there continues to be a need for a redundant  parallel non-electronic manual system which may or may not provide equivalent levels of security.  There is a bit of wiggle room in the fact that much of the background checking begins to occur the moment you made your reservations 3 weeks ago, but the final go/no-go review connection comes at the front of the queue.

OK – that’s one small piece of the logistics side.  Possible? Maybe. Expensive? You bet.  Realistic?  Not in my book.  Many of you will recall the lengthy battles regarding the TSA Behavior Detection Officers (BDO), a program that TSA unceremoniously terminated when experts ridiculed its’ “detection” success rate  as no better than a random coin toss at best; its selection criteria were almost comically random.  That’s a discussion for another day, but the background is available from a 99-page report from the Government Accountability Office (GAO) and another from the American Civil Liberties Union  which, among other things, concluded the program fueled racial profiling.

Comes now the Association for Psychological Science 68-page report with an excellent and scientifically founded  study on a hopefully distant cousin to Behavior Detection/ Facial Recognition/ Artificial Intelligence titled: “Emotional Expressions Reconsidered: Inferring Emotion from Human Facial Movements.”  Although the notion has bubbled up in several places as “the next best idea in aviation security”, there is not yet any apparent  official nexus to TSA’s big ideas of the future of AvSec.

The study is replete with interesting graphics and examples about the gross ambiguity of such extremely soft sciences, although it clearly has  some application in the scientific/ psychological community.  I think my favorite is the discussion regarding cross-cultural societies where, for example, several manifestations of “types” of expressions such as smiles in one culture may be deemed an  insult in another, and may be sub-defined by a spoken word or gesture.  There are risks in trying to figure out how to objectively “read” emotions in people by detecting their presumed facial expressions, such as scowling faces, frowning faces, and smiling faces.  How about anger, contempt, disgust, fear, happiness, neutral, sadness, and surprise. As I claimed  during one of my previous anti-BDO rants, I’ve experienced most of those in a single queue during a Friday evening push at a Cat X airport.  As the APA study points out, there is also a cultural/educational bias:

“Emotional expressions are taught to preschool children by displaying scowling faces, frowning faces, smiling faces, and so on, in posters (e.g., use of a  “feeling chart for children” in a Google image search), games ...  and episodes of Sesame Street (among many examples).”

Maybe TSA should consider a pre-board sphygmomanometer blood pressure check along with the facial review, to inform the cabin crew of your current emotional disposition selected  from an entire subculture of emoticons – warning the crew of any angry 😠,smiley😊, frowny☹, clueless 🤡,  or certifiably dangerous 💀 persons.    One must always qualify guesstimates of TSA’s future endeavors; there have been 13 TSA Administrators (including 6 Acting) since its inception, but I have summed up this innovative technological  advancement for your and TSA’s consideration for AvSec purposes: Don’t.  Just don’t.

You’re welcome.