Keeping Big Brother in Check: Facial Recognition Technology

Biometric technology identifies you based on your biological characteristics: your fingerprint, your iris, your face. 

It can even measure how you walk, your gait, said Steve Crown, vice president and general counsel at Microsoft, during Tuesday’s Center for Strategic & International Studies’ event on the human rights issues associated with the most daunting of biometrics subsets: facial recognition technologies. 

The technology is mostly deployed in two ways: verification of people and identification of unknown individuals. 

The former is based on what is known as a “one-on-one match” to match a face against one record, according to a 2019 report by the National Institute of Standards and Technology, whereas the latter tries to find a match in a “one-to-many” pool of options. 

Currently, a common use of facial verification most of us are familiar with is unlocking your phone, or perhaps your laptop. Many see this type of technology as a way to simplify and speed up a process while being almost impossible to hack or crack. 

Nevertheless, there is a growing school of those questioning whether these advantages can eclipse the social risks – especially in law enforcement or governmental controls. 

It’s one thing to be locked out of your phone – it’s an entirely different thing to be locked up in jail because of a false positive in an algorithm. 

Using only photographs in operational databases used by federal law enforcement agencies in its 2019 study, NIST found that, in one-to-one matches, more Asian and African American faces received false positives than the photos of Caucasians – this rate was even higher for African American females. 

According to the three panelists at the CSIS event, safeguards are definitely needed and they all start with transparency. Transparency of what goes into the algorithm embedded in the facial recognition technologies, how and when the technology is deployed, and assessing whether it has or does violate any human rights in any instance. 

With reports floating around on the growing number of states trying to require millions of Americans to use facial recognition technologies to access their unemployment benefits, this question seems increasingly important. 

In Minnesota, Grand Rapids Police Chief Scott Johnson wrote an op-ed last month in which he mentioned that many police departments, including major metropolitan ones like the Los Angeles Police Department, stopped using the technology because it has too many flaws. Despite seeming to be of great advantage in law enforcement, both in catching and deterring criminals, Johnson asked to what extent does it need oversight and safeguards to ensure we don’t end up with the Big Brother government foretold by George Orwell’s “1984”? 

Without any existing laws or standards that guide the biometrics industry in the U.S., CSIS’s Human Rights Initiative’s two-part report – launched and presented by its author, CSIS Senior Analyst Amy Lehr – recommended that the private and public sector alike follow the United Nations’ Guiding Principles on Business and Human Rights. 

This has to happen from the early stages of developing this technology: from collecting the database of facial images to training the facial recognition technologies systems and creating the algorithm, to its integration into software on platforms and onto applications for deployment. This assessment also has to take place when the technology is sold and deployed by operators. 

According to Lehr’s report, the rights to privacy, security, non-discrimination, effective remedy, arbitrary arrest and detention, among others, will be subject to risk when developing and deploying this technology. 

Lehr applauded the European Union’s draft of legislation for artificial intelligence, which includes facial recognition technologies and is similar to her recommendations, and aims to establish guidelines for law enforcement use of AI as well as commercial systems – both with a pre- and post-deployment evaluation. For law enforcement, facial recognition technologies would be deployed on a case-by-case basis allowed by a “judicial authority or by an independent administrative authority of the member state.” 

When it comes to artificial intelligence, machine learning, and any technology to create networks that replicate the brain’s neurological pathways, it is vital that governments and companies alike implement testing, reporting and transparency, Crown said. Companies like Microsoft and Amazon have already said they will not sell facial recognition technologies in the absence of government standards. 

Once a machine can learn to determine an individual’s zip code or the racial composition of their community, Crown explained, unexpected and unintended discriminative data points can prop up within the technology’s processing. Crown pointed to Microsoft’s Fairlearn and other tools that were created to look at how “a black box machine actually is prioritizing [and] where [weights are] actually being afforded.”

When implementing safeguards, “we need more visibility to that and more people engaged in the process,” he charged, noting the private sector also needs to partner with non-governmental organizations and civil society organizations that can help keep these technologies in check – especially before lawmakers actually set any policies in stone. 

And in Latin America, which also lacks any guidelines in this area, the human rights risk may be even more worrisome and the need for transparency is just as important. Litigating two cases against state governments in Mexico, Grecia Macias, a lawyer at Red en Defensa de los Derechos Digitales, said she is oftentimes faced with government officials that say they don’t understand the technology but just know that it works. They use the war on the drug cartels as an excuse to deploy it wherever they want, she explained, but don’t have set standards. In a time where there are increasing protests around the feminist movement or mothers protesting mass disappearances, she explained, the technology might be used to detain activists after-the-fact just for protesting and exercising their freedoms of assembly and expression. 

There are many complicated issues for U.S. lawmakers to grapple with when it comes to the potential facial recognition technologies abuses. On one hand, if a child is taken and there is a photo of the suspected kidnapper, Crown explained, it can help find them quickly. On the other hand, Macias countered, there’s the privacy protection issue for children and surveillance in real-time or en masse. This kind of issue, like in the EU draft, would warrant judicial oversight, Crown said. 

“So there might be an emergency process by which you can get approval and then you absolutely have to have transparency,” he said. “It doesn’t do any good to put in place rules if you can’t actually enforce them or see that they’re being followed.” 

To read the full text of CSIS’s reports on the human rights risks of FRT development and deployment click here.