Hacker News Reader

Comment by nathanyukai | Hacker News Reader

But as long as they don't rely on this single evidence, that should be fine right? If someone's hair, fingerprints all match the ones in crime scene, even if the software only have 90% confidence, it should be good enough. Btw from a blurred image, it's fundamentally impossible to have around 100% accuracy.

View on HN

Microscopic hair analysis is a classic example of pseudo-scientific "evidence" used to secure convictions. The FBI has washed its hands of it:

FBI Testimony on Microscopic Hair Analysis Contained Errors in at least 90% of Cases in Ongoing Review

https://www.documentcloud.org/documents/1996772-joint-press-...

Ditto bite-mark analysis:

https://phys.org/news/2015-05-fbi-hair-analysis-problems-rev...

Yet, this kind of "expert" testimony was accepted for years and years. People went to jail on the basis of flim-flam.

Do not underestimate the ability of juries to be swayed by impressive sounding claims. Nor their ability to be swayed, and dismiss clear mitigations, by one dominant or eloquent individual.

"Our facial recognition has over 90% accuracy" from the prosecution's expert can easily become "clearly guilty" in the jury room.

> 90% accuracy

I found this article helpful to understand why 90% accurate isn't as good as it sounds: https://www.badscience.net/2009/02/datamining-would-be-lovel...

He goes I to more detail in his book, bad science with a more detailed explanation. It fools everyone because it is counterintuitive.

Please be sure to attend every jury trial for criminal cases and communicate this to the jury.

Snark aside, it's important. There probably should be someone adequately numerate on every jury.

Most juries will have around half, perhaps more, who've done nothing at all mathematical since school 10-45 years previously and think percentages are in the difficult part of maths (ie it's more than basic arithmetic). HN is going to be very unrepresentative for numeracy. :)

You want them to achieve "beyond reasonable doubt" conclusions from probabilities, percentages, false positive rates and perhaps standard deviations, well you may as well be making your case in French or Ancient Greek that they can pick a few part-understood bits from.

Or hope for an eloquent statistician on the jury!

A good book on this is Reckoning With Risk by Gerd Gigerenzer, who uses several examples from healthcare.

Often these numbers are given as percentages, and he says that most people (even the doctors and nurses administering the tests) don't really understand them, and that we should use natural numbers (as Ben Goldacre does in the example you provide).

He uses examples from cancer screening and HIV testing.

> Do not underestimate the ability of juries to be swayed by impressive sounding claims. Nor their ability to be swayed, and dismiss clear mitigations, by one dominant or eloquent individual.

This. You will have a hard time contending with pretty graphs, confusing numbers and a guy with an impressive title/badge/degree explaining why you're guilty.

Eventually, for the sake of cost savings abd efficiency, all other means of investigation will be cut, wound down and abolished. Then you have Gilliam’s Brazil, and you won’t have any recourse because you won’t know any better either

Judges and jury can easily be convinced, you add some other "evidence" into the mix like "The buy is listening to rock mustic or plays violent video games" maybe some racist stereotypes and you got convince ted. This things actually happen with bad DNA, fingerprints specialists, the one that will hunt me is with a fireman that testified that he is very sure a house was intentionally put on fire and the father was found guilty of killing his family)it was an accidental fire.

My point is that you get some bad evidence, some bad eye witnesses or profiles and you get convicted, especially if the judge or jury gets tricked by the media in general on how good the forensic science actually is.

Reply on news.ycombinator.com