Cops Used DNA to Predict Suspects Face and Tried to Run Facial Recognition on It | Hacker News Reader

Cops Used DNA to Predict Suspects Face and Tried to Run Facial Recognition on It

wired.com

18 pointsthehoff16 comments

US police forensics is absurd fakery once again. I'm going to be optimistic and just hope this is ignorance and not knowing maliciousness. If so it's a trend that's not going to go away without making police take lifetime training, testing, and certification like nurses and other professionals have to do. They need to stay up to date on both the law and the technologies used in their careers.

The cops should think it's a bad idea too. The data will be junk and lead to worthless expenses to track down those garbage leads.

Given all the other junk they have used historically, I'd be surprised if they thought it was a bad idea.

> The cops should think it's a bad idea too. The data will be junk and lead to worthless expenses to track down those garbage leads.

You're assuming that their goal is to find the correct culprit. But that's not the case. Their goal is to hit their quotas, which serve as lead-gen for the local prosecutor to use to either prosecute or coerce plea deals out of people so they can hit their numbers and run for higher office.

None of that requires that they actually catch the "right" person. All of that can easily be done - and regularly is done - on innocent people, or people who might be guilty of something else but had nothing to do with the proximate crime.

The goal is to manufacture a valid legal basis for the outcomes that they want, and this tool serves that purpose very well. From their perspective, it's a smashing success.

It's the same reason police sometimes use "psychic detectives" (yes, that sounds like fiction, but it actually happens[0]).

[0] https://www.cia.gov/readingroom/docs/CIA-RDP96-00788R0001002...

"You're assuming that their goal is to find the correct culprit. But that's not the case."

Not necessarily. In most cases I'm my experience the faults are out of ignorance rather than malice. Or at the very least out of ego/pride of believing they are right.

This seems very unlikely to work, but the principle of using phenotypical information from DNA feels defensible. I guess that’s the main point of the service whose terms were violated by using facial recognition?

I can't imagine that this kind of evidence would be admissible in court, right? It might be useful to give you some leads but I would hope that no one is convicted on this tech alone.

> I would hope that no one is convicted on this tech alone.

I have bad news for you about what is already regularly used in court to convict people, and the scientific validity thereof.

Any examples?

> Any examples?

Pretty much everything that makes up "forensics", except for DNA analysis (and even that carries a huge asterisk). Fingerprints, bite marks, blood spatter... the list goes on. Drug tests are borderline - reliable drug tests exist, but police often use tests with much higher false positive rates.

Believe it or not, "police psychics" also exist[0], and are used to manufacture probable cause, so that also fits the definition of "used to convict people".

Forensic psuedoscience is widely known to be junk by modern scientific standards, but courts have stuck by it, in part because "that's how it's been done for decades/centuries", and in part because they have acknowledged that it would be too difficult to convict many cases if forensics were deemed inadmissible.

[0] https://www.cia.gov/readingroom/docs/CIA-RDP96-00788R0001002...

Notable missing in the other response is polygraph, only a handful of states accept them.

https://www.lawinfo.com/resources/criminal-defense/are-lie-d...

Look up shaken baby syndrome https://news.ycombinator.com/item?id=37650402

It’s not evidence.

The dna draws the face, the face is back checked for a match, the match has his dna collected and the dna is compared. The process in-between and the face match isn’t the evidence.

You're not even going to get a good lead out of this, facial recognition works on fine details, and guessing a face from DNA has none of that fidelity.

This is akin to: Police having grainy monochrome security camera snapshot of a suspect, putting the photo in a microscope, and concluding that the suspect has salt-and-pepper hair and a square pixel shaped face.

"fair skin, brown eyes and hair, no freckles, and bushy eyebrows" isn't enough entropy for facial recognition to work. Any entropy that any facial recognition would be operating on here is completely made up out of thin air by the rendering program.

> "fair skin, brown eyes and hair, no freckles, and bushy eyebrows" isn't enough entropy for facial recognition to work.

No, but it's more than enough to profile people, which is the real point.

Yes, humans often use vague information to profile others. But that isn't what is happening here, because the output is being fed into a face recognition program. It's just going to create spurious matches based on coincidentally generated attributes inherent to the procedures baked into the image generation software.

Reply on news.ycombinator.com