George Rebane

[This piece appears in the 10jul10 Union as my regular column for this month (here).  It was unfortunately retitled as 'Even babies practice profile'(sic).  This is the submitted version.]

Profiling has been an American bugaboo for over a generation now.  Most people don’t really know what it is exactly, but they do know it is somehow bad.  The dictionary defines ‘profiling’ as “the use of specific characteristics, as race or age, to make generalizations about a person, as whether he or she may be engaged in illegal activity.”

Profiling raises its head when we talk about catching terrorists or fugitive illegal entrants in our country.  We’re supposed to do both without the using the generalizations that make profiling such a powerful tool.  Most other countries have elevated profiling to an effective police and investigative tool by using the technology that underpins it.

You see, profiling has a more precise and technical notion of ‘making generalizations’ that connects it to learning.  Computer scientists discovered a few decades ago that the correct and most powerful way to combine new observations or data with what is already known is through Bayesian learning based on the rule named after the Reverend Thomas Bayes (1702 – 1761).  At the heart of Bayesian learning is what is popularly known as profiling – the use of powerful generalizations that most correctly cobble the new with the known.

It all starts with understanding that everything we know is not 100% reliable – in this universe nothing is or can be known with absolute certainty.  Bayes Rule is the technical means to combine measurably uncertain new stuff with old stuff at an established level of uncertainty.  Like, for instance, observing a furtive mid-eastern looking male standing in an airport security line combined with the known statistics on Islamic terrorists.  Bayesian estimation gives you the correct and best way to compute the probability that he is a terrorist, and then decide whether to expend more resources to ‘confirm’ (at a yet higher level of certainty) whether or not he actually is.

And every time you go through such a Bayesian decision process, you add more to your knowledge base – here about Islamic terrorists – making successive decisions more accurate and effective.  In short you learn.  Perhaps the leading exemplar of profiling is El Al, the Israeli airline.  Boasting the world’s best airline security record, they make no bones about using the best technology to keep their passengers safe.  Political correctness is the least of their concerns.

In the meanwhile, US airports allocate completely randomly their limited number of thorough searches of boarding passengers.  That’s why here you often see silver-haired Caucasian grandmothers being spread-eagled, while more plausible candidates for plastic explosives in their shorts are hustled through.  Why do we do this?

We do it because we are not willing to assume the political risk of the inevitable false alarms that a more powerful Bayesian profiling policy would yield.  We are prepared to assume the greater existential risk of losing an airplane full of people, or worse, than have the ACLU haul the profiling jurisdiction to court on a charge of excessively examining their client having a more likely terrorist profile that may include race, age, religion, gender, ethnic background, and country of origin.

In a similar vein, we would rather the country continue flooding with illegal entrants through a porous border than use the obvious, but not perfect, telltales of a fugitive alien to catch them.  This political paradigm is so imbedded in our culture that our leaders use their own mea culpas to highlight profiling and continue the strict bans against it.  For example, some years back the Reverend Jesse Jackson admitted tearfully to an audience that, yes, even he had inadvertently profiled when he crossed a city street in twilight to avoid encountering three young black men approaching him a half a block away.  Next time, he promised to risk all.

Many of us working in the field of machine learning have long suspected that nature evolved all critters to employ some form of Bayesian learning, causality, and, therefore, profiling in their brain bones.  So now comes the latest research on how babies and infants learn (‘How Babies Think’ in the July issue of Scientific American).  This research demonstrates that the developing young‘uns optimally use their time and brains to build and strengthen their internal Bayesian networks to interpret the world around them and decide what to do next.  It’s amazing that we can actually see how fast the little buggers learn by profiling to their hearts’ content and parents’ joy.

George Rebane is a retired systems scientist and entrepreneur in Nevada County who regularly expands these and other themes on KVMR, NCTV, and Rebane’s Ruminations (www.georgerebane.com).