Black Lives Matter: U.S. Protesters Tracked By Secretive Phone Location Technology
Mobilewalla says its dataset, which Datta told me “is never real time—we can store hundreds of petabytes of historic data,” covers 1.6 billion devices and 25 billion daily data points globally. As for this report, the company says it is based on “devices observed for geographic areas around specific protest locations in New York, Los Angeles, Minneapolis and Atlanta,” both during the day and at night, across three days at the end of May.
The data itself focuses on gender, ethnicity and those it infers come from inside versus outside the city, where phones with a “common evening location within a 150km radius of city center” are considered to be “inside.” The vast majority of tracked phones, unsurprisingly, were from inside the relevant cities. Although Atlanta, the firm says, saw more than one in five phones coming from outside.
Mobilewalla
Most protesters were males, in all cities, and the clear majority were under 34 years of age, again in all cities. The report then steps into ethnicity distribution—the vast majority of tracked phones belonged to “caucasian/other.” Despite the percentage of African Americans being much lower, the report then also presents its “gender distribution” for African Americans, but not for other ethnicities.
I asked Datta about the accuracy of the gender and ethnicity classification. “We do a lot of AI to assign attributes to individuals,” he explained. “For instance... if you use a female centric app once or twice you may or may not be a female, but if you use it 600 times over 10 months the likelihood goes up.”
Mobilewalla
As for ethnicity, Datta told me “we look at signals—there are a small number of signals that make the likelihood the person is of a certain class, ethnicity or gender or anything. We gather data where we know the class and use AI to extract patterns and project those patterns on others, it’s all probabilistic.” It is impossible to know how accurate the data is, although Datta says the firm runs controlled tests for commercial customers to prove out its models.
Mobilewalla
None of those being tracked had any idea at the time, nor do they know now. Mobilewalla says that “the information is based on mobile device data observed and derived from [its] consumer data set. The data is aggregated based on mobile devices present at protest locations and then cleansed and de-identified in accordance with applicable privacy and consent regulations.”
Datta told me that all the data, which it buys and does not collect itself, is “opt in,” but when challenged on this, he accepted that this “opt in” relies on so-called permission abuse, where apps gather way more data than they need to function. This data includes location pings, access to our contacts, web activity, phone information and even hardware if they have sought the right permissions.
