Stanford researchers claim new facial tracking software can determine your political affiliation

Jan 14th

Because artificial intelligence wasn't already frightening enough, researchers decided to teach computers how to identify a person's political ideology based upon their facial appearance and expressions.

The study was led by Stanford researcher Michal Kosinski, who already caused a stir in 2017 by programming machines that could determine whether you are gay or straight based on your appearance.

"We were really disturbed by these results and spent much time considering whether they should be made public at all," said Kosinski at the time.

Doubling down on his "concern," Kosinski decided to spend more time teaching facial-recognition software to eke out our deepest secrets and beliefs.

To teach the software how to predict political affiliation, Kosinski ran a million faces through the system, along with available data on their political beliefs. Most of these came from Facebook and dating sites (isn't it great that we all provide so much free data to our new overlords?).

Based on these million faces and the corresponding political data, the machine was taught to look for 2,048 facial features and expressions to determine if someone is conservative or liberal.

Chart showing how faces are cropped and reduced to neural network representations.
Not creepy at all!

"Political orientation was correctly classified in 72% of liberal–conservative face pairs, remarkably better than chance (50%), human accuracy (55%), or one afforded by a 100-item personality questionnaire (66%)," wrote Kosinki. "Accuracy was similar across countries (the U.S., Canada, and the UK), environments (Facebook and dating websites), and when comparing faces across samples. Accuracy remained high (69%) even when controlling for age, gender, and ethnicity."

That's some pretty scary stuff right there.

Being able to determine someone's politics from a camera shot is infinitely scarier than those internet polls that offer to determine which type of potted plant you are (while trying to collect valuable market data), and even more accurate than an extensive personality test.

As TechCrunch points out, this already has real-world implications, with the Communist Party of China's mass surveillance rollouts that track citizens wherever they go, both physically and digitally.

Imagine if government officials could use CCTV, computer and phone cameras, drones, and even satellite imagery to predict someone's political ideology based on a snapshot of your face. It's not hard to imagine the CCP scanning people's faces in the near future to convict them of being political or religious dissidents.

Of course, Kosinski doesn't want to be blamed for teaching SkyNet/The Matrix to accurately target entire groups of people at whim.

"Don't shoot the messenger," he said. "In my work, I am warning against widely used facial recognition algorithms. Worryingly, those AI physiognomists are now being used to judge people's intimate traits – scholars, policymakers, and citizens should take notice."

As for the actual data on what physical features separate leftists vs. conservatives, Kosinski was very sparing in sharing. The only things he indicated were that left-leaning individuals are more likely "to face the camera more directly... [and] to express surprise" and that they were "less likely to express disgust."

I wonder what computers will be able to do next!


There are 103 comments on this article.

You must become a subscriber or login to view or post comments on this article.

© 2020 Not the Bee. All rights reserved.