Police relied on Clearview AI and put the wrong person in jail

My guess is that some other law or tort would be a better fit, but I'll note that the "actual malice" and "negligence" are different standards with the latter being a lower bar. The former only applies to public figures.

I always wondered the same about credit reporting bureaus.

You also want to sue the police department separately for wrongful arrest to keep them accountable and discourage a use of tech like this.

It's going to be really easy for them to weasel out by claiming the match was like "99% confidence" or something, so not actually false per se. This is supported by the facts: "one of the alleged fraudsters looked like Mr. Reid"

They can also claim their technology does not make an accusation, it provides a similarity score that LEO can use in their investigations. This is also supported by the facts: sheriff’s officer insisted it was a "positive match".

The sheriff's officer actually gives up the game here, revealing they improperly relied on the similarity score to deduce a suspects guilt, but an actually competent prosecutor would know better. The fault lies with the LEO in this case. Good luck suing them.

I don't think it's nearly that simple. Do you want the same for other techniques? If someone says in good faith that a fingerprint matches and it turns out it was a false positive it should be slander? Shoeprints? DNA?

Clearview AI is providing a service in good faith like all these other things. It's up to the police and courts to use that information correctly.

Furthermore notice that this guy was only arrested, and the evidence bar for arrest is lower than sentencing. I don't think we can even say the system malfunctioned here actually. He was arrested because it seemed probable he did it. That's how it's supposed to work.

Then he was released and should be given routine compensation for being arrested and turning out to be innocent.

> Oh, this is simple.

If you're commenting on something you don't understand, maybe don't start out with this?

> I'm pretty sure 'reckless disregard for the truth' would be pretty easy to prove in this case -- considering Clearview probably can't say specifically why this person selected for arrest.

Sigh

In US defamation law, the only answers you should ever give to "is this defamation?" are "hell no" and "...maybe..." And offhand, this is a case where "hell no" seems to be the correct answer.

You're missing several things here, so let's break it down.

The first, and most important thing, is that a defamatory statement must be a false statement of fact in a nonprivileged context. While "false" is easy to understand (especially given that any nitpicking gets filed under the intent bucket), "statement of fact" is a confusing thing that's basically half the lecture on defamation by itself. But one of the things that expressly isn't such a "statement of fact" is a conclusion based on disclosed facts (even if the logical reasoning used to support the conclusion is fallacious). It's also worth noting that statements in some contexts can never be defamatory--for example, legal pleadings can never be defamatory (but press conferences about legal pleadings can be!).

As we apply that to this situation, I strongly doubt that any statement is actually defamatory. We don't know what the company messaged to the police department, but based on the quotation from the affidavit for the arrest warrant, the message probably was along the lines of "facial ID on this surveillance footage matched this person" (which is a true statement of fact), and any statement that moves to a conclusion that said person committed the crime would fall into the "conclusion based on disclosed fact" non-actionable statement. Furthermore, it's possible (I'd have to look up jurisprudence here) that the statement to the police officer is a context in which nothing can be defamatory.

The second thing to point out here is the requisite intent. As the person in question is not a public figure, the standard here is actually merely negligence--the person making the defamatory statement essentially has to be in a position where they could have discovered that the statement was false with reasonable efforts that they failed to undertake.

While it's not relevant to this case, your notion of reckless disregard is incorrect. In defamation contexts, "reckless disregard" means "entertained serious doubts of the truth of the statement." I can't really envision a plausible scenario where you could show that Clearview AI meets "reckless disregard"; even an email thread where developers talk about the false positive rate being tuned too damn high likely wouldn't qualify since it's not specifically about any individual statement they send out.

The final point to make is that the burden of proof of all of this in US defamation cases is on the plaintiff. Even in the initial complaint, the plaintiff has to allege facts with some degree of specificity (and plausibility) to show that the defendant had the requisite level of intent (be it negligence or actual malice). But before all that, the actual defamatory statement needs to be alleged... and notice that we don't even know what Clearview AI told the police department. This case is very far from a slam dunk, so far that I would be very worried about facing an Anti-SLAPP motion (which would mean having to pay the defendant's attorneys as well as my own should I lose) on filing it were I the plaintiff.

I'm not sure this strategy will work. See Popehat on defamation or slander

https://popehat.substack.com/p/can-a-tarot-card-reading-be-d...

If I write a small program using OpenCV (and maybe Unity), that tracks a pinball across a table for my personal enjoyment at home, and the police decided to utilize this technology to identify suspects, the question arises: should the aggrieved suspects sue me or the police?

I mean, it's not the tool, but the user of that tool that should take responsibility. As I remember, they must be instructed that this is not a silver bullet and may make mistakes.

Ahh the the Base Rate Fallacy - always something to keep in mind. A lot of people fall for that one.

https://en.wikipedia.org/wiki/Base_rate_fallacy

I do wonder if the AI did stop after a 99,9% match or tried the whole database and there were actually a dozen 99,9n% and 99,8% matches. That meta information should be also shown to the officers requesting a search.

A police made up evidence story wouldn't get as many clicks. A lot of people get arrested for the crime of "driving while black".

How is a story about a misuse of technology not a "technology story"?

The "evidence" would not exist if Clearview AI (or any similar vendor) did not purport to offer a reliable matching algorithm and expansive dataset. It's not as if the police randomly picked out someone to frame, and then used software to fabricate evidence for the warrant.

There is more to it than that if indeed he was flagged by a facial recognition pass triggered on a toll booth integrated into some warrant database and one policeman clicking "good enough" too easily. How deeply have these systems integrated themselves into the day to day of police investigations? Are any digital warrant requests ever denied? What sorts of protections are there to make certain that the person on the other side of the warrant request is actually even a police officer?

A hundred years ago there was a real cost in time and effort and interpersonal relationships to get a judge to sign off on a warrant at a weird hour. Are the reductions of those costs brought about by using technology in this way a net win for society?

The policeman in this actual case did something wrong. Did he, out of a motivation to increase his arrest record, find a random black man across state lines to try and arrest? Did he, out of malice, choose to target this person? Did he, out of laziness, not look too closely at two pictures side-by-side and click a "Request Warrant" button? How easy, exactly, is this mistake to make? That question only makes sense because of the technology in the story.

Technology launders abuse from police. You can see for instance breathalyzer technology which is closed-source, rife with bugs which misclassify, and provide an obfuscating cover for piggy action.

I know this outside of the scope of the article, but I've seen this a bunch of times when HN commenters claim that a given topic is not relevant for HN. The gist of the claim is basically "$badaction was already illegal, this new tech did not enable $badaction, therefore we don't need to discuss this further". The claim always seems disingenuous, because it ignores a crucial fact: prior to the tech, the friction opposing a bad action was sufficient all by itself to keep the rate of the bad action within acceptable limits. Its a "nothing wrong with nuclear weapons, if sharp rocks are still allowed too" sort of argument.

Fine, yes, this is a police-made-up-evidence story. But its also a technology-enabled-a-psuedo-scientific-confidence-interval-to-make-the-made-up-evidence-more-convincing story. Dismissing it out of hand really downplays why facial recognition (or algorithmically generated feeds or the banning of human content moderation or...) is so fraught. There may well be a solution to the problems appearing, but we're not going to arrive at those solutions without discussing those problems as, well, problems.

Excellent point. The judge should've known better than to approve this warrant on the basis that they photographed the whole country and found a resemblance 500 miles away.

> This tool will not remove URLs from Clearview which are currently active and public. If there is a public image or web page that you want excluded, then take it down yourself (or ask the webmaster or publisher to take it down). After it is down, submit the link here.

That's not an opt-out. That's only a request that they remove the index to an image that has already been removed from the web. You can't ask them to remove all photos of you, nor will they remove a photo that is still up on the web.

In other words, it's the sort of bullshit that we can expect from an evil company like ClearView.

The California opt out asks:

"Are you a California resident or are you authorized to submit this request by a California resident and submitting this request on behalf of that California resident?"

There's no language stating the photo needs to be of a resident of California. So could I just call up my friend in LA and say "hey is it cool if I request Clearview to delete me from their database on your behalf?"

We certainly don't want biases, like the idea that all short men have a complex:

https://news.ycombinator.com/item?id=35330244

The standards of evidence in the justice system are universally pretty weak. From suspect lineups easily manipulated by police to phrenology-level bullshit like bite mark analysis, this is why fair estimates are that up to 10% of convicted criminals are innocent. https://en.wikipedia.org/wiki/Innocence_Project

If an innocent person is arrested because the police did not thoroughly examine the evidence, then the police officers and the prosecutor should be reviewed and criminally charged if they broke the law. No exceptions.

Yikes. And there is very little people can do to prevent this until someone makes this their election platform to change the system, right? I'm assuming this is all constitutional.

How can non-official documents exist inside a government organization? What divides the two?

> Mr. Reid’s wrongful arrest appears to be the result of a cascade of technologies

This statement is clearly false.

As evidenced by

> and Mr. Reid was not told exactly why he had been arrested

Malfeasance is the cause, the level of which should result in prison sentences for the police who made the decision to deprive a person of their freedom and more for a week.