It took a day, I had to ponder several things here. I was drawn to an article by the associated press. The article (at https://apnews.com/article/artificial-intelligence-algorithm-technology-police-crime-7e3345485aa668c97606d4b54f9b6220) gives us ‘How AI-powered tech landed man in jail with scant evidence’. Here we have two issues, the first is that AI does not (yet) exist, the second is that AI evidence should not be valid, the rules of evidence are quite clear, so when I see “it came from a clip of noiseless security video showing a car driving through an intersection, and a loud bang picked up by a network of surveillance microphones. Prosecutors said technology powered by a secret algorithm that analysed noises detected by the sensors indicated Williams shot and killed the man.” So for all intent, we might think that the prosecutor was really clever, but as I personally see it, the man needs to be taken behind a bicycle shed and shot in the head, but that is merely my personal view. We might give value to “a secret algorithm”, yet that is merely an approach to not scrutinise the evidence. I have no idea how his defence faltered, but it did.
In Intelligence analyses there are two parts. The first is that every source is unique and as long as they are NOT connected, they can NEVER support one another. Why is that? It comes from a much older setting which is found in “Trust, but verify!” We can accept all kinds of facts handed to us, but verification is where it is at. Verification gives us the larger setting that this source makes a claim and we verified that claim via other sources and we get the same results and conclusions. This is also why actual news needs collaboration from multiple sources, and it is why credibility of these sources matter. It is why witnesses are tested, cleared and processed to give the other party no option to diminish their testimony. It is so for a person and it needs to be more for any device. And whomever relies on “a secret algorithm”, is soon regarded as non-essential weight to any office. You see, the algorithm was programmed. I am not stating that the person was wrong, or did a bad job, but who knows what the brief for the algorithm was? That brief also gives the programmer more (or less) freedom of programming. Then we get the installation and testing of the microphones, it they are out by half an inch, there is every chance that they picked up another shot, perhaps even a muffler bang, who tested that part? Who looked at the map (a GIS speciality) and considered the noise and the event? Now consider for a moment the byline “ShotSpotter equipment overlooks the intersection of South Stony Island Avenue and East 63rd Street in Chicago on Tuesday, Aug. 10, 2021.”, this is all installed on a light-pole, so if any car ever hit it, the pole will be off by several degrees, did the software see that, was that ever considered?
There is a lot more, it is seen in the part “The company’s methods for identifying gunshots aren’t always guided solely by the technology. ShotSpotter employees can, and often do, change the source of sounds picked up by its sensors after listening to audio recordings, introducing the possibility of human bias into the gunshot detection algorithm. Employees can and do modify the location or number of shots fired at the request of police, according to court records. And in the past, city dispatchers or police themselves could also make some of these changes”, so what were the raw collections, what was the distance to the event and what are the specifics of the so called “noiseless security video”, there are a truckload of issues and that is why verification is essential. This is all before we get to “an Associated Press investigation, based on a review of thousands of internal documents, emails, presentations and confidential contracts, along with interviews with dozens of public defenders in communities where ShotSpotter has been deployed, has identified a number of serious flaws in using ShotSpotter as evidentiary support for prosecutors”, it is merely the top of the iceberg, when we consider “classify 14 million sounds in its proprietary database as gunshots or something else”, you think this is trivial, but it is not. You see, this is in part the evidence, 14,000,000 sounds seems impressive, but it is not. You see there are an estimated 72 million handguns in existence, I have no included rifles and other two handed weapons, and if the database of sounds includes mufflers and tire blowouts, that lit is rather slim compared to what is out there. I can see close to half a dozen issues straight of the hockey-stick and whilst people are considering where the puck is (in Pittsburg they call it a biscuit).
So why the hockey reference? The puck moves fast, really fast and plenty of people watching the game lose sight of it in a match, this is no different. Two sources, not connected and well over 50% unverified, how could this man be found guilty? I also have some serious questions for the judge there, but I wonder if it was on his plate, it was on the plate of the prosecutor and as I personally see it, that evidence had no case in court, except perhaps a court officiated by the Marx brothers.
SZo when we get to the end and we see “ShotSpotter CEO Ralph Clark declined to discuss specifics about their use of artificial intelligence, saying it’s “not really relevant.”” Someone needs to take that horse and coach it to the side of the road, what some call AI, is merely machine learning, optionally deeper learning and it makes all the difference. With the amount of human interference (interaction) on the track from the microphone to the court room, those relying on AI are hoping to avoid the setting of bias and programming error, even source comparison errors. I reckon this Ralph Clark is on a slippery slope and with Michael Williams now on the stage where he can claim damages, a decent 8 or 9 figure damage, the 200 cases might represent a massive payout from the Government making the rules of evidence a clear debating point for whomever takes this to the next level and when the government loses a second or third trial it will be up to the Ralph Clark’s of the world to set up a defence perimeter, but I reckon it needs to be a lot more than “a secret algorithm” because at that stage such a defence will not hold water, not by a long shot. It would also help by not hiding behind AI when it cannot be AI, but that is merely a personal observation.