Facial Recognition

Some form of facial recognition technology goes back to the 19th century, but after the September 11, 2001 terrorist attacks it became ubiquitous.

The first thing that might be called facial recognition technology emerged in France in the late 19th century. It consisted mostly of photographic portraits and measurements meant to help law enforcement officials identify criminals. By the 1960s, facial recognition systems evolved to include using human features to teach computers to recognize faces. Today, computers can teach themselves to recognize faces.

“Why has the use of facial recognition become such a hot-button issue now? The most obvious answer is that the technology has been improved, streamlined and commercialized to the point that it has become widely accessible.”

The September 11, 2001 terrorist attacks made facial recognition technologies suddenly more attractive, especially for law enforcement, but its use of facial recognition technology has significant potential downsides. Whether they use facial recognition systems for national border protection or policing city streets, law enforcement officials have no way of insuring that the technology doesn’t encode familiar human prejudices. The data that forms the basis of facial recognition algorithms may suffer from lack of diversity. For this and other reasons, several cities in the United States have banned this technology and others are considering doing so.

Computers learn to recognize faces in six steps.

The first time a computer recognized a face was in 1994 in an office at Rockefeller University in New York. The way computers learn to recognize faces unfolds in six phases:

• Facial recognition software starts with the image of an ordinary face, but it doesn’t see that picture as the face of a particular individual.
• The software registers the face as pixels, to which it assigns “numerical values” that create patterns that indicate facial features.
• The software considers webs of “landmarks” or skin textures that help it “see” the face as an idiosyncratic, individual face.
• The software has to view the data at different angles to recognize faces in various postures, except when someone is facing the camera directly.
• In the 1990s, programmers had to write facial recognition algorithms. Today, the process takes place through “machine learning”: Computers iterate knowledge to teach themselves. Programmers make the information more precise by providing additional images.
• Facial recognition technology is only as accurate as its data, which means systems provided with data only from white men are limited.

Facial recognition technology is commonly used by law enforcement at all levels across the United States.

One sheriff’s office in Colorado embraced the technology for wide use. The office participated in a state-wide data-sharing program that allowed it easy access to information such as arrest records and mug shots. In this instance, facial recognition technology proved valuable in finding leads in otherwise cold cases. The agency is well aware of the potential problems with facial recognition. Its officers don’t make arrests on a single “hit” or without further evidence. In criminal investigations, they use facial recognition technology as only one tool among others.

Facial recognition technologies are mostly unregulated, but that’s likely to change.

In 2018, the American Civil Liberties Union ran pictures of all the members of the United States Congress through Amazon’s facial recognition software to compare these portraits with 25,000 mug shots. More than 20 were false matches. This demonstrates the fallibility of facial recognition technology.

“Law-abiding citizens should be pushed from databases, the only searchable people should be those with outstanding warrants for violent crime.” – Matthew Feeney, director of the Project on Emerging Technologies, Cato Institute”

The United States currently has no federal laws regulating facial recognition technology, but numerous bills are under consideration. Congressman Jimmy Gomez of California [a Democrat and member of both the Congressional Hispanic Caucus and the Congressional Asian Pacific American Caucus] was not surprised when he was among the false matches; he knew he was already more likely to be pulled over by a police officer. San Francisco, Oakland and Boston all ban municipal law enforcement from using facial recognition technologies.

This technology may affect everything from the way people do their jobs to how students and teachers act.

Researchers in China have submitted close to 1,000 patents for facial recognition technology, far more than researchers in the United States. In China, technology experts suggest a variety of uses for facial recognition technology. Some researchers believe it has the potential to bring about significant change in how people live.

“As China’s technology sector continues to spread its ideas and investment dollars around the globe, these tools come with risks.”

One technology uses face recognition software to analyze behaviors associated with learning to determine the effectiveness of classroom teaching. The problem with this, according to education professor Neil Selwyn, is that the system might give teachers an incentive to conduct their classrooms in ways that would show up well on the software or that students could figure out how to manipulate it.

Another technology deploys facial recognition technology to find consumers and feed them appropriate advertisements. Such a system could be installed on billboards or in a mall. Ben Sobel, a researcher on law, technology and digital media, points out that this system could end up targeting sensitive information about a person, like a disability. A new technology uses face and voice recognition software as people sign in and out of their jobs. Aiha Nguyen, who leads a project at the Labour Futures Initiative at Data & Society, suggests that workers, especially in the United States, could wonder if their employers would share information about them with immigration officials or law enforcement.