3 ms·
On one hand, it seems that Shotspotter lacks the accuracy to be used in court. On the other hand, it seems like a valuable probabilistic tool to help police per
by s1artibartfast 5y ago
On one hand, it seems that Shotspotter lacks the accuracy to be used in court. On the other hand, it seems like a valuable probabilistic tool to help police perform detective work and collect real admissible evidence. The problem seems to confusing the two.
You can use an Ouija board to help find video evidence, but the video evidence still needs to show something incriminating.
- ChrisKnott 5y agoIn the Williams case, I feel like the key information is how likely ShotSpotter is to miss a gunshot completely. They have video footage that shows the suspect and victim in the suspect's car alone at 11:46. If the victim was shot at 11:46 then this is extremely strong evidence of guilt, especially as it would also show the suspect lied. It seems reasonable to me that an algorithm might mistake a gunshot from inside a car as a firecracker out on the street. The classification aspect is not important. What the defence have to argue is not just that this sound is a firecracker, but the actual shot that killed the victim (which by the suspect's account happened in the same place), was not captured.
- s1artibartfast 5y agoI'm not familiar enough with the case to guess at strategy. The keys seems to be: If police have any evidence besides ShotSpotter to connect the time and place of the shooting with the video of the two of them together.
- ChrisKnott 5y agoI don't they necessarily need other evidence for the time. The article and most of the comments here are focussing on its inaccuracy in geolocation and classification, but it seems to be an accurate recording of loud sounds that ring out across the city. A high false positive rate actually works against the suspect. He needs it to have missed one.
- s1artibartfast 5y agoI think that really depends on the frequency of loud sounds detected. If possible gunshots are detected every 2 minutes, false negatives don't really matter in this case.