A man spent a year in jail on a murder charge involving disputed AI evidence. Now the case has been dropped The case against a man accused of murder has been thrown out by a judge after prosecutors withdrew disputed evidence of an AI-identified gunshot sound. Michael Williams, 65, who denied any wrongdoing, sat in jail for 11 months awaiting trial for allegedly killing Safarian Herring, 25. It's said that in May last year, Williams …
Yes. Any half-decent AI would be able to detect when a user was trying it on and would respond accordingly. Something along the lines of "I'm sorry Dave, I can't do that".
That they are too dumb to differentiate honest intent from mischief, the only conclusion is that they are much more "A" than "I".
They definitely do, as Microsoft has found out to its detriment.
Having a grid of microphones and recording where shots are heard is a perfectly legitimate use of technology.
But what I don't understand is that if somebody is shot in a car and there is a computer system that says "there was a gunshot here at the time" then... how in the name of all that is holy did this come to be seen as being cause to lock somebody up?
After working with ELIZA and PARRY interacting with the general public while I was at SAIL in the '70s and early '80s, I can quite confidently state that the answer for adults is an unequivocal yes.
Children treat so-called "AI" pretty much at face value. As soon as they start to hit puberty, and into adulthood, most folks treat it as a toy and try their best to break it if given the chance.
From Wikipedia:
https://en.wikipedia.org/wiki/ShotSpotter#Accuracy
"As of 2021, ShotSpotter evidence has been used in 190 court cases. ShotSpotter has admitted they manually alter the computer-calculated evidence "on a semiregular basis", and it has never been independently tested, leading to doubts on its accuracy. Vice's Motherboard noted that ShotSpotter "frequently modify alerts at the request of police departments."
I wouldn't trust it as evidence.
"it has never been independently tested"
With all the guns in that certain city it seems like 'someone' could arrange a test with only a little cash. Camera shot of gun going off with a street intersection sign in camera shot. Do that 10 times. Now issue challenge to police and their AI snitch - name the intersections within next 7 days. Turn pictures over to newspapers after a week. Test done...
Though I suppose it'd get warped into a "this gang did that shot in another gang's territory to prove their courage". Oh dear, they'd do it for free! Whether they'd ever stop doing it is the problem...
That actually sounds like an interesting idea - using blanks, of course.
The only issue I see is that, in the US, someone pulling out a gun in the middle of the street is likely to cause a flurry of calls to 911 or worse, someone else pulling out a loaded gun and challenging the tester.
That could end badly.
But the idea is interesting.
This. Gunpowder makes a distinct "BOOOOM" sound as it's slow burning and tends to explode in open air. Even with black powder cannons etc, a heck of a lot of powder doesn't burn in the barrel.
Any modern firearm makes a distinct sound as the sound is contained behind the bullet until it exits the barrel at which point there is a distinct sounding BANG or CRACCCKKK as the bullet exits the barrel and lets the sound out behind it.
Why on Earth not? It's a science experiment, remember?
Consider that I can construct a box that makes it possible to fire live rounds in my living room in a perfectly safe manner. Not that I would, mind. It's in my basement, next to my reloading bench. Handy for fire-forming wildcats and verifying chamber pressures aren't too high in real world hot-loads.
The impulse noise protection earmuffs are on hooks by the door ... please wear them down here. Ta.
> I believe the shotspotter technology was originally a military application
IIRC in WWII spotting of enemy artillery for counterfire was done that way. Time the sound, triangulate, and bomb the approximate spot where that sound came from. Given it took time to set up and move a towed artillery battery, you had great chances to do some serious damage.
not sure why the downvote. I did my electrical engineering masters at UMIST and can corroborate, you could definitely hear the occadional gunshot. Plus an explosion in 1996 when the IRA blew up the arndale. I remember as we had finished uni a few weeks before but still had a lease on the house so we hung around til July.
"Imagine living somewhere there are so many guns being fired."
I did, Once. When I lived in the Johnson Park neighborhood of Palo Alto, I could often hear gunfire from East Palo Alto and East Menlo Park (Belle Haven). It was the drugs, guns and hooker dealing capitol of Northern California at the time, and full of gang activity because of it. It has been cleaned up since then. Took a couple years, not decades.
I have never heard that kind of all-night gunfire anywhere else in the USA, outside of large cities with gang problems. Yes, there have been statistical abnormalities here and there, but the vast majority of shootings have been gang related for as long as I can remember.
Get rid of street gangs, and most of the gun violence in the US will stop. But we can't do that, because it would be "racist". Nobody with a backbone has addressed WHY people of colo(u)r who live in cities seem to enjoy killing each other ... find that root cause, figure out how to fix it, and the problem will vanish practically overnight.
As it did in East Palo Alto, where apparently people have a backbone and are willing to say "We have a problem! And it's OURS, not somebody in Sacramento or Washington's! Let's fix it".
"Nobody with a backbone has addressed WHY people of colo(u)r who live in cities seem to enjoy killing each other ... "
Because the answers are likely to be (1) very ugly, (2) conditional on a population with little or no desire to correct them, and/or (3) conditions which may not be physically possible to change with a human's short attention span.
IOW, if there really is an answer, why hasn't it been implemented already?
"I wouldn't trust it as evidence."
Interestingly, as in the case from the article, it seems that when "technology" is used as evidence in a court case AND all of the other evidence is at best circumstantial, one the "technology" evidence is challenged, it invariably gets withdrawn, collapsing the case and coincidently remove the "technology" from legal scrutiny.
It's almost as if tech companies don't actually think their tech could stand up to legal scrutiny and prosecutors rely on the hype and marketing blurb to make their case while hoping the defence won't challenge it. Using the likes of ShotSpotter to dispatch cops to a scene is one thing, but using it to try for a conviction is something that needs to be legally challenged at every opportunity until some sort of legally acceptable level of proof can be determined. I doubt they have enough sensors to accurately locate the sound to anything much closer than a city block or intersection.
"the MacArthur Justice Center studied over 40,000 dispatches in an under 2-year period in Chicago and found that 89% of dispatches resulted in no gun-related crime, and 86% resulted in no crime at all." (From Wikipedia)
Further it's mentioned that they provide an 80% accuracy guarantee.
So basically independent data shows this thing is maybe 10% accurate and they guarantee 80. Of course they can't let that hit a courtroom. NY spent nearly 20% of the city budget on this garbage last year, imagine having to refund all that. Gotta do the same thing they did when the military realized it doesn't work 20 years ago: grab the money, run, and find a new market.
the software/AI claims it identified location of gunshot, something that was never in dispute. The significant issue is someone was jailed on nil evidence. This suggests a degree of incompetence that rivals Oz public health administration and possibly, remarkably stupid bigots. It also demonstrates police training and local administration are also incompetent for allowing such inadequate staff to be hired and remain employed.
"the software/AI claims it identified location of gunshot, something that was never in dispute."
If you actually read the article, you'll note that that was precisely what was in dispute. The software claims it identified the location of a firework. A human later edited the result to show a gunshot in a completely different place. There is no dispute that there was a gunshot somewhere at some point, but the claims of what the software found are so heavily disputed that the company withdrew their evidence rather than face any scrutiny.
The vendor of a breathalyzer could modify the readings at the request of police. They pull me over on suspicion of drunk driving, arrest me and make me blow into the machine at the station. It shows I'm not drunk, but damn they were really sure I was drunk and they need to make their arrest quota so they make a quick call to the vendor's support line and they remotely access the machine and alter the reading.
I get booked into jail and am forced to plead guilty because there's little chance a judge or jury will believe me over the word of the cop who says I'm drunk (and maybe presents some highly edited bodycam footage where I stumbled a bit getting out of the car or something) and definitely zero chance they'll believe me over a reading on the breathalyzer.
Sounds like they should just release everyone convicted where shot spotter was part of the evidence used against them. Prosecutors can re-try them if they think they can be convicted without it, but that evidence should be treated as tainted just like when a cop is proved to have lied during a trial and all his previous testimony becomes suspect.
Well it has (thankfully) been over three decades since I last dealt with a breathalyzer, so I can't remember how it worked then or know how they work now. Knock on wood!
My example was more for illustration than anything else. Substitute another example if you want but I think you see what I was getting at. If the police can fiddle with the evidence AT ALL then it should be 100% discounted by the courts.
Fortunately, I've never personally had to use a breathalyser either, but I've seen them used and explained in fly-on-the-wall documentaries a few times :-)
As for Police being able to alter evidence, yes, that's always going to be a risk. Bad cops will possibly try that. Most cops would normally not be so stupid as to taint the chain of evidence though. But take any large group of people with something in common, such as profession, and you will find a percentage of bad apples. What we have to hope for and work towards is a balance between the risk of a bad cop perverting the course of justice and the Police being unable to do their jobs through overly onerous checks and balances. Note that phrase "perverting the course of justice". I used that deliberately because courts take that very, very seriously, especially if it's someone who is part of the system who does it. I'm sure people get "fitted up" relatively often but on the other hand, I'm prepared to believe it's actually a fairly rare occurrence (at least in the UK) compared to numbers of people going through the justice system.
As they aren't connected to the internet there isn't a way for the manufacturer to remotely alter the results after the fact. Trump supporters keep claiming there is (even absurd stuff like Italian defense satellites doing so, which demonstrates their failure to grasp basic physics) but have been unable to come up with a single shred of proof.
Given the amount of resources they've deployed trying to pursue these claims which have amounted to zilch, we should feel pretty good about the voting machines in the US. Not that there isn't always room for further improvement (especially in the few remaining places using machines that don't leave a paper trail and thus can't be hand recounted) but they are way better than the first generation of machines deployed in the 2004 election as an overreaction to the "hanging chads" in 2000.
"As they aren't connected to the internet there isn't a way for the manufacturer to remotely alter the results after the fact."
I can think of half a dozen ways right off the top of my head. The Internet isn't the be-all and end-all of nefarious connectivity, you know. In fact, large-scale information espionage and sabotage (which this is, by any other name) happened all over the world long before TehIntraWebTubes existed.
No, despite the above, I do not think the voting machines were subverted, mostly thanks to functional physical security more than anything else.
And no, I am not a Trump supporter. Do I look like an idiot?
If they aren't connected to the internet switching votes has to be built into the software ahead of time, and happen regardless of results - they'd need to do it even if the "desired" candidate already won, creating a risk of being caught for changing results when it wasn't necessary. That automated vote flipping would have to somehow bypass all pre-election testing, and given that every state / county operates independently it isn't like they know how the process works and can find a hole in their scheme they can leverage. There will be hundreds of different testing regimes, and creating a way to automatically modify votes in a way that will work during the election but not during testing would be extremely difficult if not impossible. They only have to be caught by one county and the whole thing falls apart and everyone involved goes to prison.
The fact that they print something that can be hand counted makes it impossible for vote changes to be electronic only - they have to accept a vote for candidate A but print a vote for candidate B, in a way that the voter will not see (which is possible for those that don't make a printout that the voter can verify, which IMHO they all should be required to) and that all testing will not catch.
This would be a lot harder than you think, because they only have to be caught once for bankruptcy and prison to result for everyone involved.
We have the hardware we need to make it work, it just isn't used everywhere. We have the procedures we need, they just aren't used everywhere. Some minimal federal standards are needed to enforce those best practices, while leaving the details of how the election is conducted, and what method / hardware is used, up to local authorities.
It shows I'm not drunk, but damn they were really sure I was drunk and they need to make their arrest quota so they make a quick call to the vendor's support line and they remotely access the machine and alter the reading.
At last! A new job for the people who designed and ran the Post Office IT system.
My problem is more one of common sense. He was in the passenger seat and shot in the head. That means he was either shot across the driver of the other vehicle (from the right) in which case the hole in his head is from the wrong side for the accused to have done it or, he is shot across the accused (from the left) in which case there's more GSR on the driver than the victim. In the case where the accused shot him, keeping in mind he is in the passenger seat, this would have to be a close proximity wound with more energy involved and a suitable mess and commensurate wound.
Either way it sounds like the accused is a person of colour, likely from a poor neighbourhood (drive-by), and was thereby f*cked by the system though not completely in this case.
Yeah, it's hardly a new behaviour. Back in the 8-bit days, it wasn't unusual to walk into a shop and find a computer sales display showing "FUCK!" or similar scrolling up the screen continuously. Pre-computer days, it'd be graffiti with swear words and/or cocks and balls painted on walls etc. I'd not be surprised to find some of the pre-historic cave art is rude graffiti too.
Some of the graffiti in Pompeii and Herculaneum would be considered shocking even today. Graffiti from ancient Egypt was usually a trifle more flowery, but the meaning was obvious ... so much so that the Victorian "archeologists"[0] were shocked, and tried to eradicate it. Moving Eastwards to other ancient cultures, the pattern remains the same.
Methinks it's a normal part of human behavior.
[0]These days we would call them tomb robbers.
I have family in Chicago & I agree. But I'd say it's not only in Chicago. Police departments and prosecutors all over the world use rubbish forensics like this all of the time. In 2015 the FBI stopped using bite marks, microscopic hair analysis, etc after it turned out to be rubbish. Just because CSI shows make something look Scientific does not mean it is.
https://www.washingtonpost.com/local/crime/fbi-overstated-forensic-hair-matches-in-nearly-all-criminal-trials-for-decades/2015/04/18/39c8d8c6-e515-11e4-b510-962fcfabc310_story.html
https://slate.com/news-and-politics/2015/04/fbis-flawed-forensics-expert-testimony-hair-analysis-bite-marks-fingerprints-arson.html
So they were right, 1984 and Minority Report are manuals and not works of fiction.
So despite minding my own business, always paying my way, never collecting a single penny in support from state since the age of 17 when I started working, always paying my taxes and generally keeping on the right side of the law I'm still basically screwed! I should expect a knock on my door in the next few months to be carted off to clinky for 15 years for something i didn't do, simply 'cos an AI firm has been employed by a lazy, cost cutting Gov, the crappy, "AI-lite" coded app that a child of 4 would be able to beat, will ruin mine and countless lives.
What a sorry and sad end to humanity as it's basically bends itself over the desk and manages to somehow roger itself through software into the apocalypse!
> So despite [long list]
When elephants dance, ants die. Some of those ants might have been virtuous and just, they will get trampled all the same, simply because they are ants. If you don't want to be trampled, you need to become an elephant...
Perhaps the AI company gets paid per conviction and has a deal with the DA for a minimum number of convictions per year, this is after all, capitalism.
Your apparent innocence is nothing compared to filling quotas and keeping the wheels of commerce turning.
ShotSpotter should have to pay this guy at least $10 million for basically editing a .wav file and claiming it was a gunshot.
ShotSpotter is trying to sell their company and hoped convictions would increase the value. I wonder just how many other cases are called into question regarding ShotSpotter's attempt to basically IPO by putting innocent people in prison for decades?
This is the country that thinks lie detectors are OK to use as evidence.
A quick summary of expert evidence in court.
If you create a quack device and say it can prove guilt or not, so long as its general accepted by experts in that field, the judge can accept it.
So who are the other experts in the field? Well as you invented the device, they are most likely people you have trained, and if they believe the lie it works. Bingo.
As far as I'm aware, only DNA is the only science backed forensic evidence out there.
Look how bad finger print evidence is to get a clue, and as for blood splatter....
https://www.legalmatch.com/law-library/article/admissability-of-polygraph-tests-in-court.html
"As of today, 23 states still consider polygraph tests to be admissible in court. However, the majority of those states require the approval of both parties before they can be submitted. Also, in most cases, the polygraph test is not being used in a criminal case, but rather in civil court, such as if there was an issue that prevented a person from getting a job or gaining security clearance"
Allegedly detects gunshots from the noise. I wonder whether they checked if it had spotted a shot at the location where Williams claimed the attack had taken place? It seems to me that this is a fundamental thing to check, unless you are just after convicting the first person you think of, in which case evidence of equal validity supporting his statement would be inconvenient.
The shotspotter AI should also state what sort of gunshot had been fired (I don't know, but I'm guessing that different calibre weapons have different sound signatures, and shotguns are different to rifles, and pistols etc.) and check that the wound is consistent with the sort of gunshot recorded. Then, of course you have the forensic evidence of which side of the victim the entry wound is on, and whether it was even possible fro someone in the driver seat to have held the weapon. Plus, of course whether there were any firearm explosives ejecta on the outside of the car, indicating an external attacker.
ShotStopper may be getting a lot of stick for their late interpretation of the 'evidence', but I reckon the plklice force also has quite a lot of explaining to do.
If AI systems are to be at all accepted in criminal law as anything other than a hint, and not requiring validation with hard evidence, they must be seen to work fairly, and not biased towards convictions.
Two things: 1) your opinion of what you saw in a varmint is confirmation bias and 2) the key word is reliably, Google wound foreshoring instance. The truth is that entry and exit in real shootings are often indistinguishable and sometimes you'd get them backwards using normal techniques.
Geez Monkey: "The truth is that entry and exit in real shootings are often indistinguishable and sometimes you'd get them backwards using normal techniques."
Interesting.
But in this specific case, I would have thought that in a close range shooting there would be ejecta only around the entry wound, not the exit wound. Particularly as the victim was shot while sitting in a car, I'd have thought it should be easy to tell the direction of the bullet.
Any pathologists on the Register care to comment?
"your opinion of what you saw in a varmint is confirmation bias"
No, it's not. It's observation in the field. Try to remember, I have no dog in this race. I'm just reporting what I, personally, have seen over almost three quarters of a century of harvesting food and shooting critters that try to ruin my crops.
"the key word is reliably"
Of course. From what I've seen, it is very, very reliable. In fact, I'm willing to bet you a very large sum of money that if you shoot an animal, I will be able to tell entrance and exit wounds on sight. You pick the critter, the wound placement, and the weapon. It'll be obvious.
"Google wound foreshoring instance"
Who is it that is suffering from confirmation bias? Google's not exactly what I would call that "reliable" thing that you were asking of me ... and by the way, so-called "foreshoring" is only one aspect of bullet wounds. Observation in the field suggests there are many aspects that one needs to take into account.
There is one bit of forensic evidence that would nail the defendant; powder residue on the victim. If the victim is close enough to the shooter there will be powder residue on the skin/clothing that can be used to determine the distance the muzzle was from the body. This distance may not be exact but will be close enough to say if the shot was from a few inches or a couple feet away. Since the shooting supposedly occurred entirely inside a car there also should be powder residue in the car. If there was none in the car and there was no residue or the victim that leaves either the muzzle was in contact with the skin/clothing which leaves a distinctive pattern or from some distance away. The fact no evidence about powder residue was mentioned indicates the shot was from some distance away.
WARNING: Grammar pedant.
I believe that the word 'Jihad" actually means "struggle" or "effort":
https://www.britannica.com/topic/jihad
It is not specifically about conflict or fighting, although that has come to be the usual connotation in the 'West'. Struggle can be peaceful as well as violent.
{Grammar pedant}\end
As for being the first up against the wall when the revolution comes, there seems to be a bit of a queue for that dubious privilege.
ShotSpotter could (and probably does) have its uses, but as Hannah Fry and others point out, a lot of this AI stuff is only meant to assist, *not* solve the case for you. But then again, when someone sells you software and claims it does everything short of toasting your bread in the morning, then yeah... you end up with crap like this. People claiming "computer says you're guilty" bug me. Hannah presented at a conference a few years ago and pointed this out, asking the audience whether you'd prefer an algorithm or a human to dole out your sentence. My response was that until AI was sufficiently matured and accurate, I would prefer a human, despite conscious and unconscious bias on the part of the human. It led to a very interesting discussion about that.
How this thing detecting a gun shot put the driver in the frame for doing it is somewhat of a puzzle to me.
That is the easy part. If you have CCTV that is good enough to see a person getting in the car, or even people around the car, you can make the statements
a) There were only person A and person B in the car at the time
b) Due to the position and angle of the wounds person A could not have shot themselves
c) Our amazing microphone system placed the gunshot at this place and time, which happens to be where the car is.
It's all circumstantial and it's quite the house of cards. If they are all correct everything is fine. When you have documents which routinely has people massaging one or more of the facts then it all falls over like the house of cards it is.
So a driver took a passenger in his car, the passenger was shot, and an AI is claimed to have spotted a gun shot at the right time at the right area.
But how is this relevant? The guy was shot. So we _know_ using human intelligence there must have been a gunshot. So how is that AI relevant?
First I am not a lawyer or providing legal advice. But I do remember that this 1988 article did get the legal department all excited when discussed while drinking at a local after-hours bar/pub. Since it was my copy of Science I got to explain some of the scientific evidence about DNA to the drinking partners. Science magazine is published by the American Association for Advancement of Science. AAAS was founded in 1848 as the American equivalent of the Royal Society.
Abstract
Ensuring the scientific validity of scientific evidence has always posed problems for judges and lawyers largely untrained in science. As recent cases involving the health effects of chemicals and drugs make clear, however, irrational and inconsistent decisions result when courts do not hold expert witnesses to the standards and criteria of their own disciplines. A trend toward more thorough judicial review of scientific claims has developed, and it should be encouraged.
In summary, the article is a good example 'Nullius in verba'*
Evolving legal standards for the admissibility of scientific evidence
B Black
See all authors and affiliations
Science 25 Mar 1988:
Vol. 239, Issue 4847, pp. 1508-1512
DOI: 10.1126/science.3281252
*The Royal Society's motto 'Nullius in verba' is taken to mean 'take nobody's word for it'.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
