Company says facial features reveal terrorists and pedophiles 80% of the time


[Read the post]


I guess there are some suckers out there that will buy this…


The extreme legal protections for proprietary source code may be hiding fraudulent claims.


Faception uses 15 secret classifiers


I didn’t think that there was still money in phrenology.

Also, you can get better than 80% accuracy just by automatically saying NO to both questions.


The same suckers in the UK MOD that bought nonfunctioning magic mine detectors?



…or the polygraph with its astronomical success rates…


According to fifteen different secret criteria I have, I’d say he looks like an unethical asshole.


An Israeli start-up

Explains a lot. The extensive and pervasive security apparatus there has bred all kinds of nonsense voodoo bullshit.


Sounds legit!


Yeah, I wasn’t surprised that it was an Israeli firm either, which ought to in itself be surprising.


Astronomers are rather good at deducing amazing things about distant tiny objects from a few photons. They shouldn’t be referenced, even ironically, when mentioning bullshit artists.

Can’t reveal the source code? How many trials did you do that had results no better than chance, and that you discarded? Because that’s what the pharma industry has been doing a lot of lately.


I worked at a C"R"O for some years. Most of the “studies” were financed by marketing departments, not R&D.

I disagree generally, but anyway. We could compromise on “with its astrological success rates” though. Not what I had in mind, but still funny.


Amazing new tech!


Indeed. I can offer their customers a magic stick that will, when used on a sufficiently broad swath of the population, detect non-terrorists with at least a 99% accuracy rate.


Oh for fuck’s sake! We are rolling back the advances in society and right now we are bringing modernism back to the point where Cesare Lombroso had it in the late 19th century and the “science” of classical criminology.

Hang on, what happened after he died again? Oh yes, workers rights. Women got the vote…


So, if we take the single statistic we have at its word, in the true/false miscreant/innocent matrix this system will give ratio of 80:20 true positive to false negative.

I know that human beings generally don’t have a very good comprehension of probability, but you’d think the message would have gotten out by now that when you’re trying to sort a very small number of goats from the sheep, even a nominally good ratio of false positives will make your system almost useless. And when you’re sorting not goats and sheep, but people — people who can vote, write letters to their political representatives, sue for damages, or beat up your operatives — those false positives matter.

The fact that we’re not told the rate of false positives or true negatives should be a red flag to anybody, but especially those that they’re trying to sell the system to.


Yeah, and you can tell which ones just by looking at them.


Why did you think that?


Playing a contrarian: (While nobody should take the claims of revealing pedophiles and terrorist seriously on the mere basis of facial features…) there is some non-deterministic link between physiology and behavior. E.g. massive jaws are a fairly reliable sign of high testosterone levels in men and those in turn tend to predispose owners of said jaws to statistically elevated levels of aggression and recklessness. Of course, no individual with such features should be automatically pigeonholed solely on that basis - but in aggregate, I would argue there is a real effect.