The Fight Over Police Use Of Facial Recognition Technology | News Cover

The News Cover: The majority of Americans appear in a facial recognition database, potentially accessible by their local police department and federal government agencies like ICE or the FBI. It's not something you likely opted into. But as of now, there's no way to be sure exactly who has access to your likeness. 

Over the past 10 years or so face recognition, face surveillance has moved from the realm of science fiction to the realm of science reality. But in light of recent protests for racial justice, facial recognition technologies have come under scrutiny for the way in which they're deployed by police departments around the country. Protesters worry they're being tracked, and communities of color say this tech will exacerbate bias. 

Nobody can get clear answers about who is using facial recognition and how. These technologies do not work the same across different groups of people. And oftentimes the people that they fail most on are those who are already marginalized in society. In response, IBM, Amazon, and Microsoft have all stated that they'll either stop developing this tech or stop selling it to law enforcement until regulations are in place. 

And in late June, Democratic members of Congress sought to make these pledges permanent, introducing a bill banning government use of facial recognition technologies. But lesser known companies like Clearview AI, NEC, and Rank One plan to pursue business as usual, and say this tech has an important role to play in the justice system. If somebody violently harms somebody else, that is an ongoing threat to public safety and tools that can be used safely should be available. 

Whether or not they support a ban, researchers and activists across the political spectrum are increasingly speaking out about privacy concerns, algorithmic bias and the lack of transparency and regulation around this tech. I don't think we should take an approach that a technology is inherently good or bad. But I'm not comfortable at the moment with the lack of regulation, having this technology being used among law enforcement. 

This is a type of technology that has profound implications for the future of human society. And we need to have a real conversation about whether we can have it in our society at all. Facial recognition technologies use biometric information, that is body measurements specific to each individual, to match a face from a photo or a video to a database of known faces. This database could be composed of mug shots, driver's license photos, or even photos uploaded to social media. 

It's likely that you already use this tech in your daily life, as advances in artificial intelligence over the past decade have greatly improved its capabilities. Every time you use your face to unlock your smartphone, accept Facebook's photo tagging suggestions, or sort through a Google Photos album by person, you're seeing facial recognition at work. This isn't really the type of thing that lawmakers are seeking to ban, and some are definitely eager to see the everyday users expand. 

I think there are a lot of applications that are potentially quite exciting. You know, going to a grocery store and being able just to walk out of the store without having to pay, you know the store just identifies you and automatically done it to you. But whether you're tagging photos or searching through a vast government database, it's the same technology at work. And that has others concerned. We're worried that people are going to start to normalize this technology and that could bleed into acceptance on the government side, which is extremely concerning. 

Real-time surveillance of video footage is often considered the most worrisome use case. Right now, the tech is far from 100 percent accurate. But in theory, using facial recognition on video feeds would make it possible to alert police when a suspect shows their face in public or track where anybody is going and who they're associating with. 

China and Russia already do this, and some U.S. cities, like Detroit and Chicago, have acquired tech that would make it possible. Detroit's video surveillance program started in 2016, when security cameras replaced at eight gas stations. In 2017, the department bought facial recognition software, giving them the capability to scan these cameras video feeds. Over the last, under three years, i t has rapidly expanded. 

They have surveillance helicopters, access to drones, traffic lights with surveillance capabilities. After heated debate, Detroit banned the use of facial recognition on live video, so the city cannot track people in real-time. Chicago promises that it doesn't do this either. But throughout the U.S., using facial recognition on photographs is still common, though San Francisco, Boston and a number of other cities have outlawed all government use of this tech. So we should not forget, right, San Francisco was the first city to ban face recognition. 

The place where the sausage is being made did not want the sausage, right? Private companies like Walmart, Lowe's and Target have also trialed facial recognition systems to catch shoplifters, though they say they're not currently using it. And U.S. airports are starting to roll out face scanners at the gate, so passengers need not show their passport. There's also potential to use similar tech in targeted advertising, something that Walmart is experimenting with in partnership with a startup called Cooler Screens, which infers a shopper's age and gender in order to show more relevant ads. While the screens don't identify individuals, it's not hard to imagine a system that could, a thought that puts many on edge. 

I think many people will be concerned and creeped out, but will eventually suck it up and get used to it. If someone can come up with a way, in the private sector, to ensure that this is not easy for criminals or the government just to take advantage of, then I can see people becoming quite comfortable with it. The global facial recognition market was valued at 3.4 billion dollars in 2019, and it's projected to grow steadily over the coming years. 

However, in the wake of George Floyd's death and protests against racism and police brutality, Amazon, Microsoft, and IBM made headlines for pulling back on police access to facial recognition technology. But while these tech giants certainly have clout, as well as other significant ties to law enforcement, they're not actually the most important companies in this specific market. First off, IBM did not have a real product in this space. 

Microsoft and Amazon, neither of them were big players in the law enforcement space. They did not have a large line of business there. So one could call it a bit of virtue signalling. After announcing a yearlong moratorium on police use of its facial recognition software, called Rekognition, Amazon says it doesn't know how many police departments actually use it. Rekognition is widely used in the private sector, but previously only one law enforcement customer was listed on its website. 

For its part, Microsoft says it does not currently sell its facial recognition software to police and that it promises not to until there are federal regulations in place. IBM took the boldest stance of the three, promising to stop research and development on facial recognition altogether. But this tech wasn't really generating much revenue for the company anyway. 

But many lesser-known companies are providing this technology to the police on a large scale, and they've made no such promises to stop, upsetting privacy advocates. My view is that it's fundamentally incompatible with democracy and with basic human rights to have technology like facial recognition in the hands of law enforcement. Clearview AI is a huge player in this space. Founded in 2017, Clearview has amassed a database of over three billion images, scraped from millions of websites and social media platforms from Venmo to Facebook. 

Its catalog is far more comprehensive than anything that came before it, and the company says it's used by over 2,400 law enforcement agencies in the U.S. at the local, state, and federal levels. Because we're like the largest provider in the space and we've had so much experience, we feel that it would be a shame and a really big mistake to take it away, because all these crimes would go unsolved. 

The way Clearview works is simple. You just upload a picture and the system searches its database for matches. So Katie, do you mind if I show you how it works on your photo that you supplied earlier? Go for it. So it just takes a matter of seconds. You pick the photo that you want to search, which is that one. And as you can see, it's uploading it. It's finding photos. And here there's eight photos out of over three billion that match. 

And you can see they all come with a link to the original. I see a picture from my personal website, an old news article, CNBC's website. All things I knew were out there, but not things I knew were a part of a facial recognition database, accessible to thousands of police departments. The Clearview system itself does not reveal my name, but the links point to websites that do. So we don't actually identify someone, we help you identify someone if you're an investigator. 

While Google, YouTube, Twitter, and Facebook have all sent cease-and-desist letters to Clearview, the company says that because these images are public, it has a right to compile them. At the end of the day, it's a search engine just like Google. So we think it's a little bit hypocritical of them to then send a cease-and-desist. But fundamentally, it is publicly available information. 

Other players include NEC, the 1 21-year-old information technology and electronics giant that sells its software to about 20 U.S. law enforcement agencies, and Rank One, which says it supports about 25 different law enforcement agencies. We think face recognition is a tool that empowers people when used correctly. If it wasn't our technology, it absolutely would be somebody else's technology. 

Advocates have raised the alarm on facial recognition for years, and now their concerns are gaining momentum. One of the most oft-cited issues is the general lack of transparency when it comes to exactly who is using facial recognition and for what ends. I wish I could tell you how many police departments are using this technology. Researchers at Georgetown discovered that Detroit was one of the cities using facial recognition. 

They had been using that technology for over a year before the community got wind of it. So because we don't have this transparency, I'm not able to answer this question of house widespread this technology is. That lack of transparency ma y be by design. The concern here I think from the police is, we don't want to show our hand to criminals. 

The idea is, well if we have to be more transparent about what technology we use, then people will adapt their behavior. And I think in a functioning democracy that takes civil liberties seriously, that's a price sometimes we have to pay. What's more though, the tech just isn't always accurate. And when it's wrong, it reveals bias. A 2018 study by the ACLU tested Amazon's Rekognition software on members of Congress, running their images through a mugshot database. It incorrectly identified 28 of them as criminals. And while only 20 percent of Congress members are people of color, they comprised 39 percent of the false matches. 

This report came on the heels of a 2018 study by Buolamwini and her co-author, Timnit Gebru, which demonstrated that software by Microsoft, IBM, and Chinese company Face++ frequently misgendered women and people of color. I was working on an art installation that used face-tracking technology. It didn't work that well on my face until I put on a white mask, and so that led to some questions. Others that did de tect my face labeled me male. I am not a man. B uolamwini and Gebru found that the systems they tested were nearly 100 percent accurate when identifying the gender of white men. 

But when identifying gender in darker skinned woman, Microsoft's system failed about 20 percent of the time and IBM's system failed about a third of the time. After the study came out, Microsoft and IBM trained their algorithms on a more diverse set of faces and improved their error rates. But when B uolamwini conducted a follow-up study using Amazon Rekognition, it still m isgendered dark-skinned woman nearly a third of the time. Amazon, unlike Microsoft or IBM, actually actively attempted to discredit our research. 

And we were really fortunate that more than 70 A.I. researchers and practitioners came to say no, this is rigorous, peer-reviewed, in fact even award winning work. The issues that are being pointed out are core to our field. Like Amazon noted in their critique of B uolamwini's study, Brendan Klare, CEO of Rank One, says that gender identification is a different technology than matching a face to a database of faces. 

Obviously gender estimation is sort of a different type of application, but there never should be errors like that. And the errors they showed were egregious and those are not representative of the technology. Our technology is about 99 percent accurate on gender estimation across all races. He says that claims of algorithmic racism and sexism are misleading. 

It's an important topic. It's one that has been susceptible to a lot of misinformation. Both Rank One and NEC recently made the news after their algorithms misidentified a black man in Detroit as a suspect in a shoplifting case. The man, Robert Williams, spent 30 hours in jail. Klare says that Rank One's technology was misused in this case, because a match is not probable cause for arrest. The investigating officers did not collect independent evidence. 

If the case of Mr. Williams is not an isolated incident, if there is a pattern that emerges, we will get out of this market. Clear view says a recent test showed its system is 99.6 percent accurate and exhibits no racial bias whatsoever. Previously, the ACLU has called Clearview's accuracy assertions absurd and challenged the company's testing methodology. 

The Detroit police chief, facing scrutiny over Williams' arrest, said that if the department relied solely on its facial recognition software, it would misidentify suspects about 96 percent of the time. The huge difference in stated accuracy rates vs. actual accuracy rates could be because these systems are tested using relatively high quality images. 

But when they're deployed in the real world, security camera footage can be too low quality to yield accurate results. But just for context, the most important thing, this is much more accurate than the human eye. And I think there's only beneficial things that can happen once you get to this level of accuracy. But even if a system could achieve perfection, others can think of a whole host of not so beneficial consequences. 

It's dangerous when it works and when it doesn't. But even if the technology worked 100 percent of the time, it would be extremely dangerous because at its logical conclusion, what it really does is eliminate privacy and anonymity in public space, which is something that no free society should ever tolerate. Some are more optimistic about the role that facial recognition could play in society. Facial recognition technology could be exculpatory evidence. Like, look Your Honor, I know I'm being accused by three witnesses, but there is an image of me at Walmart at this moment. It cannot be me.

What if there was a facial recognition system and it only included images of children that parents had volunteered in the case of a missing child? That's the kind of situation where I imagine that people see the value of it. But still, others ask, are the serious drawbacks worth it? There is the scenario that we would hope to be true, which is this flawless system is used in a police force that doesn't have a history of systemic racism. But this is not the world we live in. 

And so I can understand wanting to use whatever tools are available, but we have to ask, are we bringing the correct tool to the job? And so it's one thing if oh, it worked out how you thought. But at what cost? How much data did you have to collect? We believe it's extremely dangerous, in a predominantly black city especially. It doesn't make any sense to double down on something like this at a time when the nation is calling for some systemic changes within policing and to undo systems of brutality and racial violence. 

What most experts agree on is that at the very least, more regulation and transparency is needed. Many also say this tech should not be used to help solve low-level crimes like shoplifting or drug use, though some concede that it may eventually be appropriate to use facial recognition on still photos to help solve violent crimes or kidnappings. But using it on video footage is often considered a red line. 

We believe strongly that the use of facial recognition algorithms to analyze video data either in real-time or to look back at historic video footage, that that ought to be banned permanently in the United States, that we should just never be engaged in that type of surveillance. In general, a federal moratorium on this tech could garner significant bipartisan support. 

Last year, an ACLU poll in Massachusetts revealed that nearly eight in 10 voters supported a statewide moratorium, which included 84 percent of Democrats, 82 percent of Independents and 50 percent of Republicans. We're kind of in this sweet spot right now where the technology is not quite good enough to really be able to effectively achieve the goal of cataloging every person's every public movement, habit, and association. But it's getting there. So this is the perfect time, actually, for lawmakers to step in and say, you know what we're just going to draw a big bright red line in the sand here and say, we're not going to go any further until we have a deliberate conversation about this. 

But some worry that Big Tech will use this time to lobby for overly permissive regulations. We're going to be on the lookout for legislation that is clearly backed or sponsored by companies like Amazon and Microsoft, where their lawyers and lobbyists have gone over the text to make sure that it's friendly to their business model. In some form or another, facial recognition is likely here to stay. We'll probably continue using it to unlock our phones and tag our pictures. 

It may become commonplace to use it to confirm our identity at the airport or grocery store checkout line. Maybe we'll even come to accept a world full of hyper-targeted advertising screens. Name one technology we've developed and stopped. This genie is not going back in the bottle. It really is just going to be coming down to how well do we manage it? But to what extent governments and police departments can access this technology remains an open question. And that's where the real debate lies. 

Some think the political environment right now presents a real opportunity to ban a ccess to this tech at all levels of government. I think in Congress we have a real shot at getting strong legislation to put at least an immediate moratorium on local, state, and federal use of facial recognition. And I'm optimistic that we will get a ban on facial recognition in the United States. 

Others predict legislation will stop short of a ban. I think there will be many members who take the view that the technology has many worrying applications and civil libertarian issues, but that it also is ultimately useful with the right guidelines in place. And crafting these guidelines soon will be essential, because technology like facial recognition is advancing at a rapid clip. 

It may be too late to turn back the clock altogether, but some privacy advocates say that this debate is emblematic of the idea that just because we can build something doesn't mean that we should. Sometimes algorithmic justice means you don't make it, you don't deploy it. I think these technology companies need to have a sit down and ask themselves before they innovate, is this something that necessarily needs to exist in the world?