Hacker News Reader

Comment by h2odragon | Hacker News Reader

> he judge had to stop the trial for 90 minutes to ask her to explain the principles behind her analysis. Black explained her rationale, but conceded that she didn't have statistics showing the likelihood of the hands matching. "That research had never been done. I could say no more than everything matched, and we couldn't say it definitely wasn't him," she says.

But we're really sure that vein patterns are unique, and c'mon! he's a pedophile! so really we can stretch the point and appeal to authority and say "the science is settled" even if it is actually more complicated; because this one case justifies it right?

The implication there bugs me.

Edit: Are vein patterns even static? Seems to me "varicose veins" etc show that they change over the scale of years, at least, right?

View on HN

I think it's reasonable to be suspicious of overzealous use of inadequate statistics, but the article quite clearly stated that Black is not making unwarranted claims about the accuracy. You're making a valid counterargument to a claim that was not made.

I don't know how static vein patterns are, but in the particular case of varicose veins, my understanding is that they're existing veins that start to bulge outward. My guess would be that capillaries are pretty fungible, but the larger you get, the slower they'd change. For the best statistics, you'd probably want to incorporate the amount of time passing between two images, and model how much change is possible/reasonable. But there's plenty of signal to be had if you're careful to not overstate it.

> but the article quite clearly stated that Black is not making unwarranted claims about the accuracy

I agree the she or the article are not making any claims. And she eventually got the study done.

But by appearing in court as expert witness and presenting her analysis, she implicitly claimed that it's accurate enough to be taken as criminal evidence, and that was before doing the science.

That's the dangerous part.

Thank you, that's a much better expression of what bugs me about this than I was able to come up with. One of the things, anyway.

> Are vein patterns even static?

Nope. I had surgery one time to deal with a varicose vein of sorts. The key "engineering bug" was that the problem vein hits the collector vein at a 90° angle, instead of a "sliplane" (been playing too much mini motorways), causing congestion. The fix is to simply cut that connection. The body routes around it, increasing throughput through other networks. I don't think any entirely new routes were created, but I'm sure the size, geometry, and paths all changed.

Worked well for about 15 years before the primary issues resumed.

I'm yet to be convinced by the accuracy of this forensic method.

Like, it would be one thing if they could actually provide stats, bayesian analysis, p/r curves, etc.

> Are vein patterns even static?

What does that matter? Imagine some other characteristic that definitely changes over time lets say your clothing. The case in question had an expert saying: “Yes, the person recorded in the video is wearing the exact same clothes that we know the father was wearing the night of the incident, and that he is also wearing to court today. Now mind you, we dont know the exact statistical lilelihood that someone else would wear the exact same pants, shirt, tie, shoes and hat.” Your attempted counter argument is “yes ok, that is definitely the same clothes that the perpetrator was wearing and that the defendant wore that night and today, but he could change his clothes tomorrow, so…”

Assume we're starting with a 1% prior of some person being the the perpetrator of a crime.

Consider some completely static characteristic. If the suspect is the perpetrator then there's an 100% chance they still have this characteristic, whereas if the suspect is innocent then there's a 10% chance they have this characteristic just by coincidence. Therefore if the suspect turns out to have this characteristic, you should update your prior to (1*0.01)/0.1 = 10%.

Compare to some dynamic characteristic. If the suspect is the perpetrator then there's an 20% chance they still have this characteristic, whereas if the suspect is innocent then there's a 10% chance they have this characteristic just by coincidence. Therefore if the suspect turns out to have this characteristic, you should update your prior to (0.2*0.01)/0.1 = 2%.

At the extreme, a completely dynamic characteristic. If the suspect is the perpetrator then there's a 10% chance they still have this characteristic, whereas if the suspect is innocent then there's a 10% chance they have this characteristic just by coincidence. Clearly in this case you shouldn't update your prior at all.

Well Yes, it was kinda the first step on new terrain. It was correctly not used as convicting evidence. But as it was filmed at night in the girls room with a hidden cam, it was a likely match, and that's probably what motivated her to keep doing research in that direction. Later cases described in the article were using more features than just vein patterns. Plus evidence unrelated to hands.

While a sample of 1000 to compare against is certainly small, it's in the ballpark where it's valid supporting evidence.

And if you want to make it better, maybe just contact Black and ask if/how you could submit your own sample?

Your question made me read the article. And it does seem to be lacking enough sample data. At one point she had 550 samples and that was called “substantial” in the article. Towards the end it had grown to 1000 samples.

To your point the study is so small it does not attempt to answer the question in whether veins are static.

There appears to be active research in this area in general. It does look like many companies and researchers are looking at ways to use vein patterns for bioinformatics

https://biosmart-tech.com/tpost/megjc72d91-palm-vein-scanner...

https://res.mdpi.com/d_attachment/applsci/applsci-10-03192/a...

I don't think the problem is with the number of samples. You're not trying to identify the culprit (automatically) among many potential suspects. You're doing what is basic scientific research: you're using all the data you have to try to _disprove_ that he's that guy - which you can do with near 100% certainty. Any missing feature can clear him. After a certain amount of effort you give up.

Which means the trust you have in the team is a lot more important than the size of the dataset.

I don't know how checks and balances would look here. Maybe make available each analysis to the defendant and let him/his team check it out?

Like another commenter said in this thread, some people were convicted on DNA evidence of a hair that wasn't even human. A larger sample won't fix that.

The size of the dataset sets the maximum odds of a false positive. We are seeing false positive problems with DNA and the FBI won't cooperate into investigating. (The problem comes from relatives--while two random people are astronomically unlikely to false positive close relatives are likely to have many matches anyway, the odds of a false positive are merely low, not astronomical.)

I think I agree with you in a broader sense, but I don’t understand why this case in particular seems to be an issue.

> "That research had never been done. I could say no more than everything matched, and we couldn't say it definitely wasn't him”

This seems to be exactly correct. It explicits says that it was not possible given the information at time to confidently identify someone from a photo, but that there was enough information to positively rule out identification in some cases - and that that wasn’t the case here.

Reminds me of the cases which were "proven" with DNA evidence of the suspect's hairs, which often wasn't just a clear mismatch but sometimes not even human hair. But those revelations only came years or decades later.

Yes, or the one where the DNA "expert" testified that the odds were "millions to one" against a false match.

Except that the (accused, but innocent) "perpetrator" was a member of (ethnic group), as was the actual perpetrator, and it turned out that a high percentage of members of (ethnic group) exhibited this allegedly rare DNA feature.

It was actually worse than that. A lot of hair-based evidence wasn't even based on DNA, as there was no technology to sequence it at the time. It was from "experts" looking at hairs under microscope, comparing the "types" and "features", and seeing if they "match", according to their expert understanding. Suffice to say, their accuracy had zero nines at the front.

Reminds me of Doctor Tobin, ruling out fentynal, and saying that smoking isn't bad for you. Based on what science? My guess you can find an expert to say anything if it benefits them or their career. Lots of bullshit artists around.

I don't know, but I've dealt with biometric authentication that looks at blood vessels in the hand.

Reply on news.ycombinator.com