That’s Not Me

On the morning of February 16th, 2023, Portia Woodruff was getting her two daughters ready for school. I was brushing my daughter's hair up, and I heard a loud bang at the door. I was upstairs when I heard it, so I went downstairs, myself and my youngest daughter went downstairs, and when I went to the door and kind of peaked out, it was the officer standing at the door. A police officer from the Detroit Police Department. She said, "Are you Portia Woodruff?" I said, "Yes." She said, "Well, I have a warrant for your arrest." And I was like, "Okay, no. You have to be mistaken. What do you mean? You have a warrant for my arrest." She said, "No, we have a warrant for your arrest. I will need you to step outside. You can go with us, and I'm like, "Well, no, what's going on?" The officer told Portia that she was being arrested for robbery and carjacking. I'm like, "Can you tell me when this happened?" She's like, "Oh, well, it happened last year." In addition to the officer at her door, Portia saw that there were five other police officers outside her house in her driveway and on the street. Portia told her daughter to go and wake up her fiance. She also called her mother. I said, "Mom, they're here. They say they have a warrant for my arrest. I don't know what's going on." Portia's mother told her she should look at the warrant. When Portia read it, she noticed that the robbery and carjacking, she was being accused of hadn't happened the year before. It had happened just a few weeks earlier, and Portia was about to have a baby. And clearly, your eight months pregnant, you're visibly pregnant. Yes. Yes. We were kind of going back and forth, and they weren't listening to me. My mom's like, "Hey, you might want to look at the paperwork. There's no way." And then once my child's father came down, like, "Hey, what are you guys talking about? What's going on?" They were disregarding everything everyone said. So, of course, my children were there. My youngest star crying. My oldest daughter, she's looking at me, but trying to calm her sister down at the same time. And it's like, "Okay. What's going on?" My mom's pregnant and the police are trying to take her away. I was fraying too, but I'm to the point, like, I'm helpless because what can I do? Nothing I'm saying that it's not, they're not taking it in, basically. Portia agreed to go with the officers. I was handcuffed in front of my children. They walked me off the porch. They handcuffed me in front of light in the middle of the street, and then they placed me in the police car. She was driven to the Detroit Detention Center and put into a holding cell. There are guards speaking to me like, "Hey, you're pregnant. What are you doing here?" I don't know. Hey, I don't know. I have no clue. I have no idea. I didn't even speak to a detective until they arrested me and maybe around seven or eight. I didn't speak to a detective until maybe 12. One. Police later admitted they knew quickly after Portia's arrival at the police station that she couldn't have been the person they were looking for because of her pregnancy and also because of a large tattoo of a rose on Portia's arm, which didn't match the suspect's description. They questioned her anyway. She asked me about a tattoo. She asked me about a location. She asked me about the color of a vehicle. She asked me where was I at the day that the incident happened. How many months I was if I knew the victim in another suspect. If I frequently visit a certain gas station, she asked me a series of questions and then after. She asked, "Well, do you have any questions for me?" And I'm like, "Of course. Why am I here?" And what did she say? Her response was, "Oh, well, the victim chose a photo of you that was picked up from a video at the gas station." Police had run a picture of a female suspect, taken from a surveillance video through their AI-powered facial recognition system. She said the database brings up photos and sometimes it'll bring up the wrong photo and unfortunately things like this happen. And I'm like, "Wow. Are you serious?" I'm Phoebe Judge. This is criminal. A little more than two weeks earlier, on the evening of January 29, 2023, a man named Lawrence Walker called 911. He told the dispatcher that he'd been "car jacked" at gunpoint. He said two people, a man and a woman, were involved. His phone and his wallet were in his car when it was stolen. Lawrence Walker had spent some time with the woman who was involved in the car jacking, although he didn't know her well. He had picked her up earlier that day and said they had drank together, and he thought she might have drugged him. He described her as a black female, with brown and blonde long hair and a slim build. He said her name was Trinidad. Two days after reporting the car jacking, Lawrence Walker told the police that his cell phone had been dropped off at a gas station. An officer went to the gas station and watched surveillance footage, and saw that the woman dropping off the phone matched the description of the woman involved in the car jacking. The police officer took a screenshot of the woman in the gas station surveillance video. In the image, she's wearing a hat that covers her forehead and eyebrows. They ran this image through their facial recognition system. Doug McMillan is a reporter for the Washington Post. It's actually a system that's controlled by the state of Michigan, and they send a still image to an examiner in a lab who runs this through an AI-powered software system. The system generally comes back with not one result or not a few results. They generally come back with hundreds of results, and that examiner goes through and selects who they think might be the best match or the best fit for that potential suspect. In this case, the software identified 73 photos in the system as potential matches. An examiner narrowed it down further to just one potential match, Portia Woodruff. That examiner sent that potential match to the Detroit police investigator with the disclaimer, like they always do, that this is just a possible lead, and that this is not definitely the suspect, that it's just a potential match, and that that police person needs to go and take additional investigative steps to make sure it is the right person before making an arrest. Police created a photo lineup for Lawrence Walker to look at. An included Portia's mugshot, which had been taken eight years earlier. When Lawrence Walker saw the photo lineup, he picked out Portia's picture. The officer conducting the investigation asked him, "Are you sure this is the person?" Lawrence Walker said, "Yes." But the problem with that is that this facial recognition software is so good at finding somebody who looks like the perpetrator, that's what this machine is built to do, that if it finds somebody who looks like the perpetrator and isn't the perpetrator, that potential suspect is going to fool witnesses. It's going to fool police investigators, and it could even fool a judge in jury, ultimately, because what they've come back with is a look-alike person and not the real suspect. Portia Woodruff stayed in custody until that evening. She was arraigned and released on a $100,000 bond. The police also kept her cell phone. Portia says they told her they wanted to check her phone records to confirm that she wasn't at the crime scene. Portia went straight from the jail to the emergency room. She said she'd been having contractions. She got checked out, got IV fluids, and went home in the middle of the night. When you got home, what's the conversation that you had with your kids like? I talked to them a long while about just me trying to explain to them, like, you know, not things like this happen, things like this are not supposed to happen. I still try to have them have some type of trust for, you know, police officers, because my youngest is like, "Mom, they're supposed to help us." You know, we're supposed to call them when we're in trouble or if something bad is happening, you didn't do anything bad. You didn't do anything wrong. Why did they take you? Over the next few weeks, Portia had multiple court hearings before the case was eventually dropped. Here it is, I'm waiting on the arrival of my son, but the biggest thing that I have going on is trying to see about a case getting dismissed that I have no idea about and shouldn't even be going to court for it to begin with. Portia Woodruff was the first woman in the U.S. known to be wrongfully accused of a crime because of a facial recognition search. The ACLU says more than a dozen people in the U.S. have had this happen to them in recent years and stories keep surfacing. A Florida man says he was wrongfully arrested after local police used AI facial recognition technology. He spent 17 months in jail for a crime that an artificial intelligence program said he committed to leasing Fargo, North Dakota, used an AI facial recognition hit to tie her to a bank fraud case, despite her saying she'd never been to the state. The very first documented case of this kind of mistaken identity happened in 2020, also in Detroit. A man named Robert Williams was at work at an automotive supply company when he got a call from the Detroit police. They told him to come to the station to turn himself in. He thought it was a joke. When he got home later, a police car pulled into his driveway behind him. He was taken away in handcuffs. Police wouldn't tell him or his family why he was under arrest. He spent the night in jail. A cellmate told him he had overheard officers saying he was being accused of stealing watches. The next day, detective showed him a still image showing a man dressed in black wearing a red hat standing in front of a watch display. He says, "O, that's not you, I say no, that's not me." This is Robert Williams, testifying in front of Congress. He turns over another piece of paper. He says, "So I guess that's not you either. I held that piece of paper up to my face and said, 'I hope you don't think all black people look alike.'" He turned over another paper and said, "So I guess the computer got it wrong." Robert Williams was eventually released and the case against him was dropped. He later wrote about his arrest. I felt like I was in a bad movie. I couldn't leave. We'll be right back. Till some of the outs, join Criminal Plus. In the 1960s, a computer scientist named Woody Bloodsoh started working on a system of measurements that could help computers recognize faces in photos. According to a profile in Wired Magazine, Woody Bloodsoh had his teenage son and a friend go through photos of about 50 people measuring how long their ears were or the width of their mouth. That data was then entered into a computer to see if the computer could match the information to a face in a set of photographs. Their initial goal was to teach a computer to recognize ten faces. It was rumored that Woody Bloodsoh's research was funded by the CIA. He eventually taught a computer to successfully recognize hundreds of white male faces. In the 1980s, he used his research when he was hired as an expert witness in a criminal trial. His testimony, which showed that an image taken from a surveillance video, did not match the defendant on trial because their noses were different sizes led to the defendant being acquitted of several crimes. Today, Woody Bloodsoh is considered one of the founders of Artificial Intelligence. When did police start to use this technology? It's not completely new. This has been used in some form by some police for more than two decades. In Florida, there's a sheriff called the Pinellas County Sheriff's Office that has been using facial recognition in some form for more than two decades. But the technology was pretty bad and it was very kind of unreliable. In 2018, the ACLU ran a test using a facial recognition software made by Amazon called "Recognition." They uploaded a photo of every member of the U.S. House and Senate and compared them with 25,000 publicly available arrest photos. The software incorrectly matched 28 members of Congress with arrest photos. Almost 40% of those incorrect matches were people of color. A year later, the National Institute of Standards and Technology, or NIST, looked at a number of different facial recognition programs and found higher rates of false identifications for Asian, African, American, and Native American people versus white people. And also at a harder time, correctly identifying women than men. But a 2025 study by NIST found that the error rate has fallen dramatically in recent years. You really started to kind of get these magical moments where you can really see demos of this technology working incredibly well and all you need to do is put in a picture, you can even test on yourself, you can put in a picture of yourself, and it'll magically grab images from all over the internet that maybe you don't even know existed. Okay, I'm recording. All right, so, okay, I'm uploading a picture of me, the straight-on headshot I am wearing makeup. I'm going to accept that I'm over 18, and I'm going to start searching, saying it's searching for, okay, so what has come up here is a lot of pictures of me. These are pictures, a lot of, I've never seen some of these pictures before, so I'm seeing here probably 50 pictures of me from all different types of photo shoots, pictures that people have taken of me in a professional setting, headshots, but then also pictures that I don't know where of me on stage, I'd say there's probably 50 pictures of me up here. It does have this very incredible almost uncanny ability to scour the internet or scour a database of images to find somebody who looks like you or somebody who is you. And I think police, in a time when they're very strapped for resources, this is a very promising and a luring tool for helping them solve some of the most difficult crimes, you know, some of the most difficult crimes you could think of are ones where the only witness was a video camera. The problem is, you know, we don't actually know how effective it is. In the National Institute of Standards and Technology or NIST, test facial recognition programs, it uses two high quality images of front-facing people. The Federation of American Scientists says, these are, quote, "relatively ideal conditions" and that in the real world, error rates might be much higher. In very perfect conditions, when you essentially have a yearbook photo of somebody with perfect lighting who is perfectly framed with their face facing forward, facial recognition is very good at identifying the right person in that setting. But that's not how police are using these tools. In the wild, the photos that are often getting used in a law enforcement context are really low quality. Caitlin Roel works as a public defender. The software could get better and better and better. It could get exponentially better, but if what you're putting in is a grainy, black and white surveillance photo taken from overhead where someone's looking to the side and wearing a beanie, it doesn't matter how good the technology is because the software just can't do a good job comparing it. Eight years ago, Caitlin was working in the Bronx and was assigned a client who had been charged with robbery in the first degree. The robbery had happened at a TJ Maxx. Caitlin says surveillance video showed someone stealing some socks, and then when a security guard approached them, the person waved a box cutter at them and left. A few months later, Caitlin's client was picked up by the police. I just couldn't figure out why they connected him to this robbery. She didn't make any statement saying she had done this. There was nobody found the socks. He didn't leave something that belonged to him. It was just in a rest months later, so I called the prosecutor and said, "Why do you think my guy did this?" And he said, "You're never going to believe this," Caitlin, they used facial recognition. It was the first time that Caitlin had heard about someone being identified this way. She started digging into the case to figure out what had happened. So the police went to that TJ Maxx and they pulled surveillance video. And when they pulled surveillance video, they talked to the lost prevention officer and what they told him was, "We have this fancy new facial recognition software we're going to run this surveillance through." The NYPD then ran a still image from the video through the software. Caitlin says it was a black and white image taken from above. the person was wearing a hat. Katelyn's client came up as one of the possible matches. The police took a screen grab of just my client's mug shot and they text message it to the lost prevention officer. And what they said was, "Is this the person who robbed you?" The security guard texted back, "Yes, this was the person who robbed the store." What was particularly concerning and interesting in this case is that we had a very good but not perfect alibi for where he was that day. What was it? So his baby boy was born that day. We had photos of him in the hospital. We had the birth certificate that he was on. We showed all of that to the prosecution. The problem is that the birth time was a couple hours different than the time that the socks were stolen. So is it theoretically possible he was sitting there through the birth of his child and thought, "You know what? I'm going to run to TJ Maxx and steal some white socks." It's possible, but it's deeply, deeply unlikely. And we were in this tricky spot where both the police and the prosecution who do not understand this technology were sort of saying, "Well, yeah, sure you've got that. And when their mind's like facial recognition, couldn't have got it wrong." There's a phenomenon that researchers talk about that's called algorithmic bias. When you have a very fancy expensive machine and you put it in front of these police officers and it gives them a result, they are likely to believe that that is the right person. The computer must be telling me the right person because it's a very fancy machine and it's a very expensive machine and that is going to be probably the suspect that we're looking for. And so when police get that result, they do seem to go out of their way to try to find that person guilty and they do sometimes forget to take other investigative steps that you normally would, like interviewing witnesses, going to check financial records and bank accounts, going to do basic police work that they've all been trained to do once they get this AI result. Some of this seems to go out the door and they sometimes get blinders on where they're just kind of focusing on how do we get this person in custody, how do we get this person prosecuted? Caitlin says she thinks her client would have had a strong case, a trial, but he had a new baby and he could have sat in jail for a very long time waiting for that trial, so ultimately he took the plea. How long was he in jail for? About six months. Caitlin Rule says since this case, she's worked on other cases where facial recognition technology has been used by police, typically using surveillance photos. But she suspects it's being used more often than she and other defense attorneys know about. Police generally do not disclose when they're using this technology. In most places in the country, police are not required to say that they've used facial recognition this way to identify a suspect. I went through hundreds of police reports from departments all over the country where they're using these tools. We knew from internal logs that they were using this technology, but when we read the police reports, nine times out of ten, they were not anywhere explicitly saying facial recognition technology. At most, they were usually saying we used investigative techniques to identify this person. Therefore, most of the time, the defendants have no idea if they were identified using this machine. Ultimately, there could be people who were taken to trial, there could be people who were convicted, there could be even people who accepted plea deals, people who are behind bars right now who are misidentified by facial recognition, and we don't even know about it. So maybe there are dozens or maybe there are even hundreds more of people who are wrongly identified this way. We'll be right back. Before we get back to the show, just a note to say, this fall, I'm going on the road with a big experiment. It's a brand new live show all about me and my family and secrets. It's totally different from anything we've done before. We're going to announce the full run of tour date soon, but first, we've got a couple of special shows coming up in just a few weeks. I'll be in Durham on September 16th at the first place we ever did a live show, Noterco, and in Portland, Maine on September 18th, we've never done a show in Maine. You can find out more at this is criminal.com/live, I hope to see you very soon. My understanding is that we were the first defense side to use AI, an official recognition to help exonerate someone. Christopher O'Brien is a criminal defense attorney based in Fort Myers, Florida. In 2019, he took on a client named Andrew Conlon. Andrew had been in a high speed car accident with his friend Colton Hassett. Their car was going about 100 miles an hour when it went out of control, hidden lamp posts and trees. The car spun violently and flung Colton out the window and he died in some bushes. The car then caught on fire, Andrew was trapped inside. A bystander helped get him out through the driver's side of the car and later talked with police. He told them that Andrew had been the passenger in the car. But the police didn't ask for the bystander's name or contact information. Never asked him his name, a phone number, never asked him, "Hey buddy, can you stick around a little?" I'd like to talk to you some more seems like, you know, you're an important witness. Police kept talking with other people at the scene who said they thought Andrew had been seated in the driver's seat. At one point, police also thought that Andrew said he had been the driver. He said they must heard him. Andrew maintained that he had been the passenger, not the driver. But more than two years later, he was charged with a vehicular homicide. It's a second degree felony, which is punishable by up to 15 years of prison. When Christopher O'Brien saw that there was body camera footage of someone saying his client had not been driving, he wanted to find that person. We didn't know who it was, but he was in an orange, cut-off shirt that said, "Event security." It was a bright orange shirt, so I mean, he's pretty, you know, he stood out. They made flyers and put them around the area and ran a social media campaign. No one turned up. They worked on the case for more than two years. Christopher O'Brien was working with another defense lawyer named Patrick Bailey, who wondered if they could use facial recognition technology to find the bystander. They tried using free software, but it wasn't helpful. Instead, they wanted to use something called ClearView AI. ClearView AI is a facial recognition company that works primarily with government agencies, like law enforcement and the military. It works by scraping billions of images off every corner of the Internet, including social media platforms. It's controversial, and has faced privacy lawsuits across the globe. Patrick Bailey wrote to the CEO, "Anastro a license to use a software." He said, "I am aware that your company only provides facial recognition software to law enforcement agencies." What happens if those same agencies refuse to use ClearView AI's important and powerful software to identify an eyewitness that could clear an innocent person? This stuff is available to cops. The state attorney's office and these cops could have done that. They could have run his face and found him if they wanted, but they had plenty of witnesses who said Mr. Conlon climbed out the driver's door and they were happy with that. The CEO of ClearView AI said he would let them use a software. We fed it in with the still picture that we had been disseminating around town and on Facebook and so on. We put that picture in and ClearView AI popped up like 30 pictures of the guy within seconds. I mean, it was insane how quickly it popped up these pictures of him, one of which was him in the Orange Event Security shirt, the exact same one he was wearing on the night. I was like, "That's him. I mean, that's the guy." One of the photos showed the man dancing at a club in Tampa. they later learned that his name is Vincent, the photo is clear. You can see which club and who he was dancing with. So we called up the club and they said that they didn't know Vincent but they knew the other guy. And so we got in touch with that guy, the friend. And then apparently that friend got in touch with Vincent because one morning we were here at the office in the phone rang and it was hey I hear you guys are looking for me. We were like no no way it was like hitting the lottery finding this guy. Vincent had moved away to St. Augustine so he hadn't seen any of the social media posts or flyers. He agreed to give a deposition about the accident. One of the questions the state asked Vincent he's really an old whiteie of now they're asking well why do you think he's the passenger and he's like well I went up to the car looked in the driver's door and he was in the passenger seat his feet were down in the well like where the passengers feet go and I tried to pull him out and he wouldn't come and I had to finagle those were his words I had to finagle him out so once they heard that and they were the prosecutor looked at me he's like how are we gonna argue with that. The state dropped the charges against Andrew Conlon. The National Association of Criminal Defense Lawyers calls facial recognition technology unregulated untested and flawed and wrote that this case quote does not override the larger issues that facial recognition introduces into the criminal legal system. Clear view AI has Andrew Conlon's story up on their website one supervising attorney who works at legal aid called it a PR stunt. Christopher O'Brien says he would use facial recognition technology again if it became available to defense lawyers but he also thinks it should be used with caution by everyone. We shouldn't just abdicate our human sense of morality and and just basically good common sense that we have I'm just abdicated off to some tech and go well the tech says X so X is the answer I'm very against that. My pregnancy was the only thing that really saved me I would have been still fighting the case as we speak now probably or already in jail doing sign for something I didn't do. Portia Woodruff I felt like if the detective had of just came hey let me go and visit her and knock on her door and talk to her and ask her these questions they could have been avoided. It was just complete carelessness all the way through what do you hope happens going forward. I hope that with this type of case you know sometimes I feel like they they feel like okay where they're out they're not locked up so that's good enough for them and honestly that's how I felt like they okay where at least you're not locked up and it's like okay I shouldn't have been thought about being locked up I'm not doing anything wrong for me to be in prison or in jail for anything so for that to happen and then you throw it up like oh what things like this happen to who why six months after Portia Woodruff's arrest the Detroit chief of police said that it was quote human error that led to Portia's arrest not the facial recognition system he said he apologized to Portia Woodruff and her family he did say it was lazy police work he was absolutely right criminal is created by Lauren Spore and me Nadia Wilson is our senior producer Katie Bishop is our supervising producer our producers are Susanne Robertson Jackie Zajiko Lily Clark and Lena Celicin this episode was fact checked by Katie Cedarborg our show is mixed and engineered by Veronica Semonetti Julian Alexander makes original illustrations for each episode of criminal you can see them at this is criminal calm and you can sign up for our newsletter at this is criminal calm slash newsletter we hope you'll consider supporting our work by joining our membership program criminal plus you can listen to criminal this is love and fever eats a mystery without any ads plus you'll get bonus episodes these are special episodes with me and criminal co-creator Lauren Spore talking about everything from how we make our episodes to the crime stories that caught our attention that week to things we've been enjoying lately to learn more go to patreon.com/criminal we're on Facebook at this is criminal an Instagram and TikTok at criminal underscore podcast we're also on YouTube at youtube.com/criminalpodcast criminal is part of the Vox media podcast network discover more great shows at podcast.voxmedia.com I'm Phoebe Judge this is criminal