Hello. I'm Ryan Cook, and this is Civic Tech Chat, a show that looks at the way technology, politics, and policy impacts the world around us. The tools we use, the way services are delivered, and how we talk about and set policy all shape our society. We'll gather around and have a chat about these things together and more. Either that, or maybe I'll rant about a topic for a while. Anyway, let's go ahead and start the show. Welcome folks to another episode of Civic Tech Chat. This time, we're gonna be talking about facial recognition and its role in government. But before we get started, I do wanna let you all know that we've started a discord for the podcast. There will be a link with an invite down in the episode description. Do feel free to go check that out. It's a small community right now, but hoping to grow it. It's a great way to reach out to me and let me know things that you might want us to cover or to just hang out and talk about civic tech. Well, with that said, let's go ahead and go to our topic. So you might have heard that the federal government is seeking to expand the use of facial recognition technology. And you might then be wondering, well, what is that tech? What's it all about? And it has to do with developing the ability to identify someone using an image of their face. Often, this uses machine learning techniques that seek to teach some software how to match up an image with some other sort of reference. It's generally somewhat imperfect as it's as it is right now, and the ways that it's imperfect can start to become troublesome, which is something we'll talk about a bit later. These technologies are often used in a number of different ways. For most folks that can be things like face ID or Windows Hello. So when you go to unlock your phone, you look into the camera and that device tries to figure out if you are the authorized user of it. Same can be said if you're trying to sign into certain laptops, you look into the webcam and it does the same thing. Though it can be used for a plethora of other uses. It could be used for transportation security, trying to identify you before you get on a plane, or maybe when you're trying to cross an international border, or law enforcement can also use it to do a number of different activities, like maybe trying to identify persons of interest. This has become a topic that's become salient somewhat recently, and that's in part because the US Government Accountability Office has generated a report talking about the ways that agencies within the US government are attempting to use facial recognition. Some highlights of that report include the fact that 16 agencies reported using facial recognition for digital access or cybersecurity purposes. 14 of those mentioned that they allow folks to use it to unlock their issued mobile devices. So again, that is maybe that more benign use where it's like, hey. We're using this to help figure out our identity authorization for devices and things of that nature. Six agencies then reported using the technology to generate leads for criminal investigations. So this you can think is identifying a person of interest, maybe comparing images that they have against mugshots that are in a database, also identifying criminal victims through the use of images that are openly available on social media. We'll end up talking about this specific one a bit more later when we talk about policy. This is definitely one where folks are putting a lot of attention. And then lastly, five agencies reported using the tech for monitoring surveillance for physical security. So thinking about this, that would be maybe identifying folks that are on a watch list. So maybe you have this, like, list of folks that for whatever reason you don't want in your facility and you're using data from your cameras to try to automatically inform security personnel that they have entered the facility or the area. And basically, the idea that it was limit the amount of memorization individual security personnel would have to make while doing their jobs. And as you might imagine, there is indeed policy about brewing about this. In The United States, two senators, senator Ron Wyden and senator Rand Paul, the first a Democrat from Oregon, the second a Republican from Kentucky, introduced a bill together called the Fourth Amendment is Not for Sale Act. This bill would seek to ban the US government and its law enforcement agencies from buying location data and personal information without obtaining a warrant beforehand. In addition to that, it would seek to block the purchase of data that's considered, in quotes, illegally obtained. The way I I would think to interpret that is things like hacking, someone intentionally breaking a contract to get access to some data, or other sorts of trickery. Also recently, the FTC wrote up a post that is, talks about this issue a bit. I believe it was posted on April 19. It explores the use of AI technology a bit more broadly and its use for automated decision making. Facial recognition also kinda fits under this umbrella depending on how it's used. If it's used for example to try to make decisions, then it it kinda falls under what they're talking about in this report or blog post. And it does mention some interesting things. For example, the FTC talks about three existing laws in this context that they think are useful. One of them being section five of the FTC Act, which prohibits unfair deceptive practices. And an interesting thing that they note in that is that this would also include the sale or use of racially biased algorithms. They also mention the Fair Credit Reporting Act, which they say would be of use in cases where an algorithm is used to deny people things, like employment, housing, credit, and other similar sorts of benefits. And the third thing they mention is the Equal Credit Opportunity Act. This one makes it illegal to use a biased algorithm that results in credit discrimination based on things like race, religion, national origin, sex, marital status, and similar such things. This post also attempts to give some guidance around the ethics of of all of this and how to communicate about it. And there are some interesting headlines that I kinda cherry picked out from this that I found interesting. One of them is tell the truth about how you use your data. Another, watch out for discriminatory outcomes. Another is don't exaggerate what your algorithm can do or whether it can deliver fair or unbiased results. And perhaps my favorite one is the one that's at the very bottom of this page, and it says, hold yourself accountable or be ready for the FTC to do it for you. It sounds like they're they're quite serious about wanting to enforce these laws and provisions. All of this brings us to the question, well, why? What's what's the big deal with this? Why are policy makers paying enough attention to this do things like write bills, to have agencies releasing materials about it. As you might have gathered from what we've talked about in that policy space, folks are becoming aware that these technologies are quite imperfect. And the ways that they're imperfect, in particular with facial recognition, can cause a great deal of risk for individual people out there in the world. Let me, lay out a scenario for you. Let's say there is a law enforcement agency somewhere that has facial recognition software set up. We're gonna use that identify persons of interest. Use cases, our example. While they're going about their day, and let's say the software is analyzing images they found, and it turns out that it mistakenly matches that image up to an image of you. Suddenly you're now thrust into a situation where you're being investigated, and that puts you in a position where your livelihood put be put at risk. You could be put in a situation where you have to pay for expensive lawyers. What if you can't afford all of that? So I I I make this point and I tell I give this scenario because there's this idea that there's these machines out there as a narrative that are making automated choices and because it's a computer, that's that machine, that it's naturally impartial just because it's not a person. But that narrative is fraught with problems. The machine or the computer isn't gonna be any more impartial than the design of the model itself. Computers only execute the instructions that they're given. Even if you, you know, incorporate a technology machine learning, it's then taking data and instructions and attempting to learn how to do something from all those things. And all those things are both provided and designed by humans. Because of that, human bias is gonna seep into whatever methodology is used to make those choices. But you shouldn't just take my word for it, I'd like us to talk for a moment about a study where some folks did a deep dive about this. And this example was written by, and I apologize in advance, I'm gonna mess up these names right now, but it's written by Joy Bulamuini, or Bulamwini and Timnit Gebru. Those two authors put this paper together, and they sought to evaluate bias in automated facial analysis along race and gender, effectively comparing different offerings that were available and trying to figure out how accurate they were along those classifications. And there were some really interesting key findings that they came to. Really two in particular that I want to bring up right now. The first one is that all evaluated classifiers perform better on male faces than female faces, with differences in error rates ranging from 8.1% to 20.6%. The second one is that these also perform better on lighter faces than darker faces with an 11.8 to 19.2 difference in their comparative error rates. I imagine you realize hearing those numbers that those are quite significant. And if you recall again back to that example of the potential consequences that someone might face if they're mistakenly identified, say as a suspect by one of these, you can start to see how, especially given the current inequities we already have, especially in our criminal justice system, that if these technologies are applied in a risky or inappropriate way, it's just simply gonna exacerbate those and make society even more inequitable than it currently is. Well, I appreciate all of you taking the time to listen to this and chat with me about this important topic. If this has stoked your interest and you'd like to do a deeper dive into the effects that automated decision making can have on vulnerable populations, I would highly suggest checking out a couple of books. One of them is called Automating Inequality by Virginia Eubanks, and another is Weapons of Math Destruction by Cathy O'Neil. I will link both of those in the episode description. And as always, don't forget to subscribe to this podcast app in the listening app of your choice to get more content like this, as well as other long form interviews. Again, thank you for listening, and I will catch you next time. You can follow us on Twitter using the handle at civic tech chat. Visit us on the web at civictech.chat, or subscribe to us