One of the more popular recommendations that has emerged for combating “deep fakes” is to the use of digital signatures on smartphones and cameras that would certify that a given image or video was recorded in person rather than digitally doctored from afar. Central to this proposal is the idea that the public would be comfortable with having the imagery and video they capture be digitally traceable back to their specific device and that they would be willing to encode their GPS coordinates into the content they share. Twitter’s experience over the last seven years suggests the smartphone-owning public may be less amenable to this idea than previously thought.
The idea of digital signatures as a way to authenticate images as being unaltered is one with a long history in professional photography. In the last few years it has experienced a resurgence of interest as a way to combat the growing danger of AI-generated falsified imagery in the form of “deep fakes.” One popular idea is to have cellphones digitally sign images and video captured by their onboard camera, including recording the GPS coordinates of where the user was when they recorded the imagery.
Given the ubiquity of GPS-enabled smartphones and the growing popularity of location-based services, the idea of digitally signing imagery and encoding a user’s GPS coordinates into the image might seem like an obvious solution and one that would generate little pushback from users already accustomed to broadcasting their locations in realtime.
However, Twitter’s experience with geotagged tweets suggests users actually care more about their locative privacy than technologists care to admit.
Since 2009, Twitter has allowed users to geotag their tweets, affixing their device’s GPS coordinates to the post so that it can be placed on a map precisely where they were standing when they sent it.
For those concerned about revealing their precise GPS coordinates, the company also allows users to specify a coarse bounding box around their actual location, such as revealing only the city or country they sent the tweet from.
Combined, both precise and bounding box geotagged tweets accounted for more than 3.5% of tweets in January 2012 but have fallen steadily to less than 1.5% of tweets as of the end of 2018.
Fewer and fewer Twitter users are comfortable with sharing their GPS-verified location.
Even worse, it turns out that a large percentage of those precise GPS-geotagged tweets were inadvertent. Prior to April 2015, when a Twitter user chose to report only their coarse location, like the city they were tweeting from, a spokesperson confirmed that Twitter would in reality quietly record their precise GPS coordinate in the tweet’s JSON record without telling them. This coordinate was not visible to the user – they saw only the city name like “San Francisco” in Twitter’s interface, but researchers and data miners who consumed Twitter’s various data feeds could see the user’s precise GPS coordinates, often accurate to what side of an intersection they were standing on when they sent it.
In April 2015 Twitter changed its policy regarding precise geographic information and began recording a user’s GPS coordinates only if a user explicitly selected to share their precise location. Overnight this led to a 70% reduction in GPS-tagged tweets.
It turns out that the widespread availability of GPS-tagged tweets on Twitter was not a result of citizens across the world being comfortable with sharing their precise geographic locations. The majority of the mere 1-2% of users willing to share their locations were actually willing only to share their city or country-level location and had no idea that Twitter was secretly sharing their precise GPS coordinates with the world.
Once GPS coordinates were recorded only for users that explicitly requested GPS-level geotagging, they all but vanished.
In their place, 98.7% of Place-tagged tweets are to either the city, state or country the user tweeted from. Just 0.43% of Place-tagged tweets are to a specific business address.
In short, even among users so comfortable with sharing their thoughts with the entire planet that they happily broadcast them to the Twitterverse, only the most microscopic percentage of those users are willing to share even the business-level location they are tweeting from, let alone their precise GPS coordinates.
Apparently, people value their locative privacy far more than technologists have believed.
If Twitter users are not comfortable sharing even their city-level locations, why would they accept using their even more revealing precise GPS coordinates to digitally sign the imagery they capture with their phones?
Putting this all together, digital signatures might seem like an easy solution to deep fakes for someone sitting in the technological utopia of Silicon Valley. Yet, the experience of Twitter over the last seven years suggests that even its oversharing users are loathe to share their authenticated location, suggesting that digital signatures may not fly with the public. Unless device manufacturers make digital signatures mandatory and force the public to accept this latest privacy loss in the way they have accepted each of Facebook’s changes over the years, it is unlikely that the public will willingly adopt any form of digital signature that invades their locative privacy.
