From the U.S. Government Accountability Office, www.gao.gov Transcript for: FBI Face Recognition Technology Description: Audio interview by GAO staff with Diana Maurer, Director, Homeland Security and Justice Related GAO Work: GAO-16-267: Face Recognition Technology: FBI Should Better Ensure Privacy and Accuracy Released: June 2016 [ Background Music ] [ Narrator: ] Welcome to GAO's Watchdog Report, your source for news and information from the U.S. Government Accountability Office. It's June 2016. Automated face recognition technology has become more accurate over the last few decades. According to the FBI, this technology can help law enforcement agencies identify criminals in their investigations. Diana Maurer, a director in GAO's Homeland Security and Justice team, led a recent review of the FBI's use of face recognition technology. Jacques Arsenault sat down with Diana to talk about what they found. [ Jacques Arsenault: ] Face recognition technology has come a long way since it was mostly the stuff of science fiction books or movies. Can you tell me how is face recognition technology used today? [ Diana Mauer: ] I think the first thing people need to understand is it's not used the way you see it on TV or in the movies. People are used to seeing a TV show, the bad guys' pictures taken at a crime scene. They put it into a database and within five seconds, they have a positive match. That is the long-term goal; we are far from achieving that goal. The way it's currently used is that images are being fed into these systems. The FBI is checking against a number of databases, but they're using it for investigative leads. The technology is not advanced to the point where we can have what's known as positive identification. We cannot say for certainty that that person was exactly the individual in the photo. [ Jacques Arsenault: ] And so then, where do these images come from that they're putting into the database? [ Diana Mauer: ] The images come from a variety of sources. Many of them come from criminal mugshots that have been collected over the course of many years at various state, federal, and local law enforcement agencies. Some of the images also come from when people apply for specific kinds of jobs or they have to get a criminal background check or for employees who need to get a security clearance. There are also, in some cases, images that are being collected from the department of motor vehicles from a number of different states. So, it's from a variety of sources. [ Jacques Arsenault: ] And about how many photos altogether? [ Diana Mauer: ] It's about 30 million different photos are available for different kinds of searches. [ Jacques Arsenault: ] So then, what do we know about how accurate this technology is? [ Diana Mauer: ] Accuracy is a really good question. We know that the FBI tests its system for accuracy in a very controlled environment and it meets very specific performance parameters. What we don't know is how accurate it is in real-world use, and that's because the FBI is not doing what's known as operational testing, which means test how well the system works when you're using it for what it's actually designed to be used for, which in this case is trying to identify bad guys. [ Jacques Arsenault: ] So, when thinking about accuracy, it reminds me, in a very less technological way, of speed cameras that might be out on the roads, that it's important that they're tested in the factory and that they kind of work when they come out of the box, but it's also really important to test them in real-world conditions and make sure that they're measuring accurately. And it sounds like you're seeing some of those same challenges in a much more complicated technology. [ Diana Mauer: ] That's absolutely the case. You can imagine that when criminals' images are captured during the course of committing a crime or in other ways they're not necessarily the best quality. They might be trying to hide their face. They may be disguised in some way. So, that makes it very challenging to use those photos and compare them to a mug shot, where the images are being taken in a very controlled, very consistent environment. That's a significant technical challenge. They've made great strides. It's a very important and powerful tool for law enforcement, but they need to do a better job of ensuring the accuracy. [ Jacques Arsenault: ] Now, turning to privacy. You know, hearing that there are 30 million photographs, it's clearly not just mug shots but a wide variety of people whose pictures are taken and may not know that they're going to be used for any purpose like this. What are the privacy implications that you found in this? [ Diana Mauer: ] Well, under federal law, agencies are required to let the public know when they're using personal information and they're also required to let the public know how that information is being used. So there's transparency, so that the public is aware. We found that in the case of the FBI's use of facial recognition technology there were problems in both scores. We think it's important for an informed democracy for the voters to understand the potential trade-offs. On the one hand, it is important for law enforcement to have access to this kind of information. It helps keep us safe. On the other hand, innocent people have a right to know when their pictures are being taken from the department of motor vehicles and provided for FBI's use for investigating criminal leads. [ Jacques Arsenault: ] So then what recommendations is GAO making to the FBI in this report? [ Diana Mauer: ] We had a series of recommendations for the FBI on both accuracy as well as privacy. On the accuracy front, we recommended that the FBI do more stringent testing of the technology before it's used. We also recommended that they begin conducting tests of how accurate it is in real-world environments. And, third, we recommended that the FBI validate and work with their external partners to ensure that their information is of sufficient accuracy for FBI's purposes. On the privacy front, we recommended that the Department of Justice issue clear information to the public on how facial recognition technology is being used at the Federal Bureau of Investigation. We also recommended that they determine why the privacy impact assessments for both uses of FBI's facial recognition technology were not completed and were not made public prior to their initial use. [ Background Music ] [Narrator:] To learn more, visit GAO.gov and be sure to tune in to the next episode of GAO's Watchdog Report for more from the congressional watchdog, the U.S. Government Accountability Office.