Opinion: Big Tech Can’t Stop Facial Recognition by Itself

Even the best efforts of three big providers can’t stop the technology’s spread or misuse. Licensing agreements might allow police departments to use parts of it even if they can’t use specific algorithms.

by Cathy O’Neil, Bloomberg Opinion / June 17, 2020
Surveillance security camera.(Dreamstime/TNS) TNS

(TNS) — Amazon, IBM, and Microsoft have generated a lot of excitement by announcing that they will, at least for a while, stop selling facial recognition software to police departments. Municipalities have imposed moratoriums. Some activists are calling for an outright ban, on the grounds that the technology can disproportionately – and wrongly – identify people of color as criminals.

If only it could make a bigger difference.

Concerns about facial recognition are well founded. There are no rules governing tolerance for errors that can ruin people’s lives. Vendors often make unrealistic promises, causing police to do silly things like overstate the software’s reliability. And even good technology can perpetuate racial bias: In New York City, for example, the identification database consists largely of the disproportionately black and brown people whom the police have already processed

Unfortunately, even the best efforts of three big companies can’t stop the technology’s spread or misuse. Licensing agreements might allow police departments to use parts of it even if they can’t use specific algorithms. Plenty of other purveyors, such as Clearview AI and Palantir, are available to fill the breach. And federal law enforcement agencies, such as Immigration and Customs Enforcement, will be hard to cut off.

The biggest issue, though, is that any effort to curb facial recognition runs counter to the country’s efforts to combat terrorism. Back in 2005, Congress adopted the Real ID Act to address a problem that the 9/11 attacks exposed: 18 out of the 19 terrorists involved had acquired fake IDs from local departments of vehicles. The legislation requires officials to verify that any individual has only one license — a task that entails, among other things, collecting biometric data and sharing them among different state and federal agencies.

Despite various delays and some resistance from states (Maine initially opposed it), the requirements look likely to go into effect from October 2021. All U.S. states and territories must comply, or the identification documents they issue will become useless for domestic travel. It’s hard to see how they will do so without subjecting people to facial recognition technology. Many already use it and share their databases. (And yes, things get confusing for identical twins.)

Full disclosure: I have a company that audits algorithms for bias, and we worked with the Washington State Department of Licensing on facial recognition. One thing I found encouraging was that as a sanctuary state, Washington doesn’t share records with ICE, so people getting driver’s licenses don’t have to worry about the information being used to deport them. On the other hand, this means that when the whole system is up and running, undocumented immigrants in non-sanctuary states probably will have to worry.

In short, Congress would have to change the Real ID Act before bans on facial recognition could be meaningfully enacted. This, in turn, would require Americans to think hard about what really matters: protecting ourselves from devious terrorists trying to game our systems, or defending the vulnerable among us and upholding civil liberties? Unfortunately, those interests are all too often in opposition.

©2020 Bloomberg News, Distributed by Tribune Content Agency, LLC.

Looking for the latest gov tech news as it happens? Subscribe to GT newsletters.

Platforms & Programs