After years of rolling our eyes as CSI investigator wait impatiently for blurry images to magically be enhanced to a nice clear image, somehow restoring data which had never saved to begin with, Google Brain is now bringing a similar technology to fruition.
“The first part, the conditioning network, tries to map the the 8×8 source image against other high resolution images. It downsizes other high-res images to 8×8 and tries to make a match.”
“The second part, the prior network, uses an implementation of PixelCNN to try and add realistic high-resolution details to the 8×8 source image. Basically, the prior network ingests a large number of high-res real images—of celebrities and bedrooms in this case. Then, when the source image is upscaled, it tries to add new pixels that match what it "knows" about that class of image. For example, if there's a brown pixel towards the top of the image, the prior network might identify that as an eyebrow: so, when the image is scaled up, it might fill in the gaps with an eyebrow-shaped collection of brown pixels.”
The thing is, all they’re doing is making an image of a celebrity. If they started using pictures of everyone, then they’re just compiling photos of the most average person imaginable. Nothing about this is actually creating a real image. Despite this, even the most intelligent of journalists are suggesting this be used to aid police. The idea is that if it’s not used for a trial, it’s ok if it helps them find the suspect.
What if you’re just a very averaging looking person? It reminds me a lot of a recent story related to IP mapping:
“For the last decade, Taylor and her renters have been visited by all kinds of mysterious trouble. They’ve been accused of being identity thieves, spammers, scammers and fraudsters. They’ve gotten visited by FBI agents, federal marshals, IRS collectors, ambulances searching for suicidal veterans, and police officers searching for runaway children. They’ve found people scrounging around in their barn. The renters have been doxxed, their names and addresses posted on the internet by vigilantes. Once, someone left a broken toilet in the driveway as a strange, indefinite threat.”
It turned out that when the company MaxMind started mapping IP addresses, they took a couple shortcuts.
“As any geography nerd knows, the precise center of the United States is in northern Kansas, near the Nebraska border. Technically, the latitudinal and longitudinal coordinates of the center spot are 39°50′N 98°35′W. In digital maps, that number is an ugly one: 39.8333333,-98.585522. So back in 2002, when MaxMind was first choosing the default point on its digital map for the center of the U.S., it decided to clean up the measurements and go with a simpler, nearby latitude and longitude: 38°N 97°W or 38.0000,-97.0000.”
“As a result, for the last 14 years, every time MaxMind’s database has been queried about the location of an IP address in the United States it can’t identify, it has spit out the default location of a spot two hours away from the geographic center of the country.”
Seems to me like some average looking people could end up with a whole lot of knocks on their doors if this kind of software is starts getting used to aid law enforcement. Bigger question, why is Google even making this software? What use case is there where it wouldn't be taking liberties it probably shouldn't?