Cops Ran Phone Pic Through AI Slot Machine… WRONG MAN ARRESTED!

[00:00:00]So, you're a commercial crabber in Fort Myers, Florida. One day, you wake up, you do your job, you come home. Same boring Tuesday as always. But then, law enforcement shows up at your door and informs you that you you tried to lure a 12-year-old out of McDonald's in Jacksonville Beach. A city that you haven't been to in years, if at all. 300 miles away on the literal opposite end of the state. So, you tell the police exactly that to their faces. I've not been out to Fort Myers in probably 2 years. And per the reporting, the officer's answers amounts to, "I know I'm looking at your mugshot." How does he know? How is he so sure? Because a computer said so. Our computer looked at a grainy, shadowed, off-angle cell phone photo of a monitor. Yes, a photo of a screen. And announced with all the confidence of a man who's never been wrong in his life, 93% match.

[00:01:01]Apparently, that was good enough for the cops. Now, that 93% sounds scientific, right? Sounds like a lab coat said it.

[00:01:08]But we're going to go into what that number actually means, and I promise you, by the end of this video, you are going to be furious about that number.

[00:01:15]Because here's the thing the badge and bootlicker crowd doesn't want to grapple with. This story isn't just about bad technology. The tech is upstream. This is the story about what happens when the government finds a shortcut around the Fourth Amendment and decides that the Constitution is a you problem.

[00:01:33]So, I've read the complaint, 66 pages, filed June 10th, and it's worse than the headlines. Let's get uncivil about it.

[00:01:41]>> [music] >> The plaintiff in this case is Robert Dillon. He's 52. He's from Fort Myers.

[00:01:53]He's married and has one daughter.

[00:01:55]Self-employed crabber, which, if you've never met a commercial crabber, is about the most leave me alone, I have traps to check profession in America. This is not a guy with a Twitter account and a grievance. This is a guy with a boat and a schedule, a guy who actually does a real job. As for the defendants, well, it's the city of Jacksonville Beach, the Jacksonville Beach Police Department, the Jacksonville Sheriff's Office, a different agency, and also the Pinellas County Sheriff's Office, a third agency.

[00:02:26]All of them all the way across the state, and we'll get into why they're here. Plus, there's two individuals named by name. And the cause of action matters, write it down. This is not defamation, this is not a they were mean to me.

[00:02:40]This is a civil rights claim under the Fourth Amendment for wrongful arrest plus malicious prosecution. For those of you unfamiliar with it, the Fourth Amendment says, "The right of the people to be secure in their persons against unreasonable searches and seizures, and no warrant shall issue but upon probable cause." Now, probable cause is the key phrase because when it comes to arrest, it's the whole ballgame, and we'll come back to it like a recurring villain. So, what is this all about? Well, it all happened in November 2023. Some man walks into a McDonald's in Jacksonville Beach and, according to the report, tries to get a girl under 12 to leave with him. She doesn't go, smart kid. She calls her parents who are nearby who call the police. By the time the cops arrive, the guy is gone. Now, that is a real and awful thing that apparently happened. Some child was targeted. And I want to be crystal clear, somebody might have done this. There might be a real predator out there.

[00:03:39]Because of that, it's going to become the most damning part of the whole story. The next day, the responding office, in this case Jacksonville Beach Police Department, goes back and pulls the surveillance footage. And here is apparently investigative step number one, the foundation of everything that doesn't follow. The police doesn't actually get a copy of the video.

[00:04:00]Instead, he decides to take a picture of the screen with his phone.

[00:04:06]Now, I need you to understand the energy here. There is a digital video file. It exists. It's sitting right there, and his move apparently was, "Nah, I'll just photograph the monitor like my aunt screenshotting a Facebook post by aiming her iPad at the TV." The lawsuit describes the resulting images, and I'm quoting the gravamen here. There's a low-resolution image of the suspect's face, partially shadowed and off-axis.

[00:04:32]So, garbage in.

[00:04:34]The single most important input of the entire investigation, and it's a phone pic of a screen in the dark at an angle.

[00:04:41]Garbage in, garbage out.

[00:04:44]So, what does our intrepid investigator do with the world's worst photo? He starts running it against booking photos. Nothing. He starts running it against the sex offender registry, which, if you think you're looking for a child learning suspect, that database would be good, right? But apparently, there's no matches.

[00:05:02]So, the trail is cold, stone cold. Weeks go by and nothing.

[00:05:07]And at this point, a competent investigator has options, real ones. The complaint actually lists them, and of course, they're chef's kiss obvious. The suspect was picking up a mobile order, a phone order placed online, which means that there's an account with a name and a phone number and a payment method attached to the guy who probably actually did it.

[00:05:30]Did anyone ask McDonald's for the order records? According to the lawsuit, no, never requested.

[00:05:36]The manager said the suspect was a regular at this specific restaurant in Jacksonville Beach.

[00:05:42]Did they pull other days of footage to ID the regular or ask the manager if they knew the name? Apparently not.

[00:05:48]Every door that leads the actual predator, the guy who again is still out there, stayed closed because a much shinier, less effective door appeared.

[00:05:58]Something that didn't require them to actually do work.

[00:06:01]So weeks in with nothing, the Jacksonville Beach officer sends the glamour shots he took to other agencies.

[00:06:08]Anyone got anything?

[00:06:09]Enter the Jacksonville Sheriff's Office.

[00:06:12]An investigator there takes the grainy photo pick of a screen and feeds it into facial recognition software. Now, for those of you who don't know facial recognition software, let's describe what it actually does. It is not a fingerprint. It does not say that this is the guy. What it does is it takes an input image and combs through its databases, millions of photos, and then returned a ranked list of faces that look most similar according to the algorithm.

[00:06:38]Most similar, that's it. That's the problem. A lineup of look-alikes. So by design, almost everyone it returns will be innocent because probably only one person did it. In the best case, there might be one match in a pile of doppelgangers. In a normal case, it might be all doppelgangers. The real guy, of course, might not even be in the database.

[00:06:59]The system that the Sheriff's Office used here was run by a third agency, the Pinellas County Sheriff's Office. It's a statewide platform imaginatively called Faces. Pinellas owns it and leases it out to hundreds of departments across Florida. That's why a Sheriff's Office on the Gulf Coast is a defendant in a Jacksonville case. They run the machine the whole state apparently plugs into.

[00:07:21]The machine decided to return Robert Dillon.

[00:07:24]And the Sheriff's Office relays it back to the Jacksonville Beach the magic words 93% match, and apparently that was good enough for everybody. So let's talk about what that 93% actually means because this is the number that did the damage, and it's basically a lie dressed up as a statistic. 93% does not mean 93% likely it's your guy. Even if it did, 7% is a hell of a margin of error. It does not mean a 93% chance of guilt. As Dylan's own attorney laid out right there, it's an internal similarity score. An artifact of the software's own math.

[00:08:00]It is not an assertion that there's a 93% chance this is the suspect. It's functionally the computer saying, "These two pictures kind of rhyme." And law enforcement heard 93% decide not to ask any further questions, and their brains filled in basically a confession. And that doesn't even factor in the model's real-world error rate. No one flagged that the input was a shadowed phone of a monitor. No one noticed, apparently, the lower-quality images, and I'm paraphrasing the complaint's own language here, "degrade the system's ability to produce anything reliable because, of course, garbage in, 93% garbage out." And here's the part that should make every law-and-order conservative grind their teeth. Because we are the ones who actually respect cops enough to demand that they be actually good at this. The guy who runs the agency that ran the match, the Jacksonville Sheriff TK Waters, went on record and said, paraphrasing, "If you brought him a facial recognition hit and called that your probable cause, he'd kick you out of his office because that's not how it's supposed to work."

[00:09:03]And remind you, that is the sheriff. He is the representative and the final authority for the agency.

[00:09:11]A defendant in this lawsuit. Saying the quiet part out loud before the lawsuit.

[00:09:16]The boss knew the rule, the system still broke it, and now he's a defendant in a case that exists because someone did the exact thing he said would get them thrown out of his office. You can't make this up. The defense's best witness is basically suing himself. So, now that the cops have a name off the machine, what did they do next? Well, they build a photo lineup, Dylan's photo plus filler photos shown to the restaurant manager and she picks Dylan. See, a witness ID'd him. That's real evidence.

[00:09:47]No. And this is the most important legal tech concept in the entire case, so help lock in. When facial recognition gets it wrong, it gets it wrong by handing you someone who will look a bit like the suspect. That's the entire function, to find similar faces.

[00:10:02]So, when you take that look-alike and drop him into a lineup next to random filler photos, you rigged the game. The filler photos, by definition, look less like the suspect than your algorithmic doppelganger does. You've built a multiple-choice task where the wrong answer is the most attractive option on page. So, of course the witness points at him because he looks the most similar. You've engineered that outcome the second you ran the search. That is why police procedure experts rightfully say you should never go straight from facial recognition to a witness lineup.

[00:10:34]It's not a safeguard, that's Gatzke.

[00:10:36]It's a known landmine they walked directly onto.

[00:10:39]The lineup didn't confirm the machine.

[00:10:42]The lineup laundered the machine. Took an unreliable algorithm, Gatzke, and dressed it up as independent human eyewitness testimony. If the entire thing is circular, the corroboration is just the original error wearing a disguise. Now, here's where measured legal analysis starts to fail me a little because this next part is generally enraging. The police did not lack exculpatory evidence, and that is key. They had exculpatory evidence. They went and got it and then pretended they didn't, and that's a massive no-no. What exculpatory evidence did they have?

[00:11:15]Well, item one. They ran Dylan's license plates through automatic license plate reader network, the ones that log your tags all over the place. Results? No hits anywhere near Jacksonville Beach in the window around the crime. His vehicle was nowhere near the scene. The cameras said it wasn't the guy.

[00:11:33]Item number two, the suspect was a regular at that McDonald's. Dylan lives and works 5 hours away and per the complaint doesn't travel much because he's a crab fisherman who stays in the area.

[00:11:45]A guy who's a regular 300 miles from where you live and work is not a regular. That's not a clue, that's clearance.

[00:11:53]Item number three, they called him.

[00:11:56]Investigators actually called Dylan and he told them never been to Jacksonville Beach, never in my life. Item number four and this one is almost insulting.

[00:12:04]On that same phone call, Dylan told the officer that he has a distinctive scar on his face. A scar. Something you check against the surveillance footage in about 4 seconds. The footage they were too lazy to get a clean copy of.

[00:12:16]Did the guy at McDonald's have a scar?

[00:12:18]Well, we may never know because again they used a grainy poor photo. But instead of pulling that thread, per the complaint, they buried it. Didn't look into it at all. So they had several independent pieces of evidence. Each one in its own way saying wrong guy.

[00:12:32]Together they're not a hint, they're a billboard. And what did the lead investigator do with all these exculpatory pieces of evidence? He left it out. You can't do that. Now, here is the legal core of this case and here's the doctrine that applies and it's not Brady. People keep saying Brady. Brady's about hiding evidence from defense at trial. This isn't trial. This is earlier and arguably worse. The doctrine in this case is Franks versus Delaware, US Supreme Court. The rule is simple and sacred. When an officer applies for a warrant, he can't lie to the judge and critically, he can't stay silent about facts that he knows about that would destroy probable cause. He has to give all exculpatory information that he's aware of to the judge. In this case, a material omission of something you know is treated like a lie. If you leave out the exculpatory stuff specifically so the judge will sign it, and the warrant wouldn't survive with the truth in it, the warrant is poisoned, and it's no warrant at all. You can't do that.

[00:13:37]So, let's picture the judge. A judge in Duval County is handed a warrant application. It says, "Facial recognition match. Witness picked him out of a lineup." Sounds pretty good.

[00:13:46]Sign here. But, critically, what the judge was not told per the complaint, that the camera recognition network that recognizes license plates showed his car was never there. That the witness ID came directly off a tainted machine hit.

[00:14:00]That the suspect was a local regular, and Dylan lives 5 hours away. That Dylan flat-out told them that he never set foot in the city. The judge signed a warrant for half the story. You can't blame the judge, you have to blame the officers. And the warrant built on half the story, where the missing evidence screams innocent, is not a valid warrant under the Constitution. Full stop.

[00:14:21]That's not Uncivil Law's hot take, that's Frank's, 1970 still good law.

[00:14:26]That is the Fourth Amendment violation, not the AI. The AI is a dumb tool. The violation is a sworn officer curating reality for a judge to manufacture probable cause that the actual facts did not support. So, 8 months after the incident, August 2024, the warrant finally gets executed. Deputies show up at his home in Fort Myers, and Dylan does what innocent men do. He's confused, he argues. He says the thing that should have ended it. He hasn't been to Fort Myers in 2 years, and he's rarely, if ever, been to Jacksonville Beach. And per the reporting, the officer's response amounts to, "The computer says it's you." That's it.

[00:15:04]That's apparently the whole thing.

[00:15:05]That's the whole wrong thing in one sentence. A human being with a badge, apparently too lazy to do his job, with the power to put another human in cage, looked at a living man telling the truth, and deferred to a 93% similarity score, which again, even if it were true, is a 7% margin of error, and that ain't nothing. Generated, incidentally, from a phone picture of a monitor.

[00:15:27]So, he's arrested, charged with enticing or luring a child, a third-degree felony in Florida for up to 5 years in prison, and he has to post bond. So, how does a self-employed crabber make this kind of bond?

[00:15:40]Per the complaint, he had to borrow money and pledge the title to his truck.

[00:15:43]His work truck, the thing he needs to earn a living, hawked to get out of the cage for a crime that happened while he was, as it turns out, at work 300 miles away.

[00:15:54]So, he's put into a group holding cell.

[00:15:56]The complaint says he sat in silence, too terrified by the nature of the charge to even speak with anyone around him, which, incidentally, is a pretty good idea.

[00:16:04]It's one of the most stigmatizing accusations in American life, hung around the neck of an innocent man by a computer guess that wasn't even that good and based on garbage evidence.

[00:16:15]So, here's how flimsy the whole thing was. Once Dylan had a defense attorney, which again he had to pay for, they cleared it almost immediately. How?

[00:16:23]Well, he proved evidence that he was at work that day. That's it. The entire investigation the police never did, done in a couple months by a private defense lawyer. Prosecutors, of course, dropped the charges because it was stupid. About 2 months after the arrest. But, and of course this is the cruelty of it, dropped charges don't really undo everything. It took nearly a year for him to get his mug shot taken down and the arrest expunged from the record. And of course, that does nothing for news reports. A year of a child-luring mug shot floating around the internet attached to an innocent man's name with how many news reports and community posts about it.

[00:17:01]The complaint notes that community members still walk up to him in public to accuse him about it. And he says he no longer feels comfortable being friendly around children because being friendly to a kid is now a risk for a man who was publicly branded.

[00:17:14]No agency, of course, has apologized, not one. And here's the detail I can't let go of. The lead investigator, after this investigation produced the wrong arrest of an innocent man based on garbage, apparently he was promoted to corporal.

[00:17:29]That is the best evidence of failing upwards I've seen in recent memory. The system didn't just fail to punish the error and the stupid, it instead rewarded it. Now, someone in the comment section is already typing, "This is just an ACLU race thing."

[00:17:44]Uh no, no, it's not. Robert Dylan is white. I bring that up not to score a point, but because it destroys the this only happens to other people comfort blanket. Yes, the testing data is real.

[00:17:56]These algorithms do misfire worse on darker skin and on women, young and old.

[00:18:00]That's a whole 'nother problem unto itself. The facial recognition software is not good with darker skin people in the first place.

[00:18:07]But Dylan's case proves a deeper point.

[00:18:09]When the government adopts an unreliable tool and treats it as gospel, nobody is safe.

[00:18:14]There are This error came for a middle-aged white crabber who never left his country. And it might come for you, too, because apparently 93% is good enough. And unfortunately, this is not the first case like this. This is like the 15th known case like this in the country. This has happened in Maryland, Michigan, Missouri, Louisiana, Nevada, New Jersey, New York, North Dakota, Arizona, and Florida, to name just a few. This is not a glitch. This is apparently the method. Now, we are not an anti-cop channel. We are a competent cop channel. The reason to be furious here isn't that police have tools and use them. It's that a powerful government tool was used to skip actual work the Constitution requires to validate the 93% match. They could have used it as a lead, they simply used it as the ultimate truth without performing the additional steps. Conservatives have been warning about this for years. You do not hand the the state a surveillance machine and then simply trust it implicitly.

[00:19:15]Power plus a shortcut aimed at ordinary citizens while the real predator walks free because everyone stopped looking the second the machine lit up. That is the part that should haunt everyone.

[00:19:26]There is a real man out there who might have really tried to lure a real child in McDonald's, and so far he's gone away clean because an investigation closed around an instant grabber 300 miles away. Bad police work doesn't just frame the innocent, it also frees the guilty.

[00:19:41]And the Fourth Amendment isn't some red tape. We believe in the Constitution over here. It's the thing that stands between you and the computer says it's you.

[00:19:49]The lawsuit is asking the court to say so out loud, and force these departments to adopt safeguards they should have had all along. Robert Dylan is suing for damages, for bond, the lawyer, the lost income, the truck, the year of his life, and the reputational harm that you can't fully scrub. And as for injunctive relief, he's asking for court-ordered policy changes. So the next honest working man doesn't get a child predator mugshot because a sheriff's office somewhere ran phone pic of a minor through a slot machine and called it all probable cause without any further steps.

[00:20:20]Now, this case has been filed in the Middle District of Florida, and I'd be lying if I said it isn't going to be a long and very complicated road. These kinds of cases are never easy. Now, here's a rule for everyone watching, and it's one piece of legal advice that never expires. If the police tell you, "We have an AI match. You're the one we're looking for." You say four words.

[00:20:40]"I want a lawyer." You don't explain, you don't argue you're innocent for 20 minutes like Dylan understandably did.

[00:20:46]You lawyer up because the system will absolutely let the machine talk you into a cage if you let it.

[00:20:52]If this made you angry in the productive constitutionally literal way, that's our channel. So make sure to smash that like button. It generally tells the algorithm to show this to more people, which is a delicious irony given the subject matter. And the comment section is where you should land with these questions. is official recognition a tool that needs the guardrails, or do you think it should be banned all right? Until later, my friends, I've been on Civil Law, and until next time, keep it uncivil out there. Cheers, and goodbye.