The dog got a reward when it signaled the presence of drugs. (before an actual search to confirm the dog's signal) That seems like a great way to train a dog to always signal the presence of drugs.
Anyone who knows anything about dogs would say that drug-sniffing dogs are a constitutional nightmare. They're basically a rubber-stamp. Dogs want to please their handlers. If the handler wants the dog to alert, the dog will alert. The presence or absence of drugs isn't relevant to the dog.
Yes. I train my dogs to give me its paw when I show it a cookie and say "paw". It never takes long for them to just sit down and start waiving their paw around as soon as I pull out the cookie. Or rushing to the door when I get home and doing the same thing even when I didn't present a dog treat.
Actually in several states they're still usable. Some as supporting evidence, others requiring double consent, and some just as evidence. Their time has not completely passed sadly.
Never mind that the government itself still uses them as part of their security clearance process...
> The article for this post is a pretty good source: 100% rate for the dog indicating the presence of drugs is pretty much a rubber stamp.
If you read the article you will find a description that the dog didn't alert the first time around the car, didn't alert the second time around the car, and only alerted after the cop maliciously signaled to the dog that they want an alert. To me that doesn't sound like a bad dog, it sounds like a bad cop.
>Similar patterns abound nationwide, suggesting that Karma's career was not unusual. Lex, a drug detection dog in Illinois, alerted for narcotics 93 percent of the time during roadside sniffs, but was wrong in more than 40 percent of cases. Sella, a drug detection dog in Florida, gave false alerts 53 percent of the time. Bono, a drug detection dog in Virginia, incorrectly indicated the presence of drugs 74 percent of the time.
These are concerning false positive rates, but I don't think anything in the article suggests the extent of what the parent wrote, in a universal sense:
>They're basically a rubber-stamp. Dogs want to please their handlers. If the handler wants the dog to alert, the dog will alert. The presence or absence of drugs isn't relevant to the dog.
Sure, but that's one particular dog. I'm asking about the statistics for all of the drug dogs in the US.
I agree if a particular dog is wrong more than 10 - 20% of the time, the dog and/or the handler should be taken off of the job. And I agree that they probably shouldn't be allowed to be used by police unless they already have probable cause through some separate means.
I think it was seen as me slyly trying to defend the use of drug-sniffing dogs. I'm really not; I was just curious about the data, and wondered how much variation there might be.
It's interesting to see the difference in testing between the US and Europe (or NL at least) - here getting 60% correct is a passing grade (usually 50% is the minimum)
It's tricky because, while dogs can understand fairly complex vocabulary, their ability to reason abstractly about events is less developed. If the dog alerts, you spend two minutes searching, and then you reward the dog, it will have trouble connecting the reward to the alert.
The obvious solution is to train the dog to search things that you know have drugs (or don't) because the trainer set it up that way. You'll need to then repeat this training often, and not always in the same environment. If you don't, eventually the dog will figure out that it will always be rewarded for alerting in the field.
The problem with this, of course, is that it's expensive and cumbersome to do. You need to set up cars/bags/lockers/etc with and without drugs in them in a variety of locations, secure them (lest actual criminals steal the drugs), and bring the dogs to them (lots of time out of the handler's day). Also, even if you do this, you'll still get false positives: dogs are living animals, not machines.
This is hilarious. It's exactly the same problem as the "use AI to identify suspicious people" thing except that everyone freaks out about the former but loves this dog thing because they're fluffy.
If we ever make fuzzy happy fun killbot, I think civil rights might be doomed because there'll be an army of killbot fans who will defend them even if they run internally on an API call to random.org
My dog tried to train me to give him a night time treat.
When I let him out at night, I give him a treat when he comes back in after doing his business outside.
Then he got smart enough to scratch at the door to go out again, then came right back in and sat by the cabinet to wait for his treat. Finally he gave up on the extra treats when he found it didn't work.
Dogs are very good at figuring out the actions they need to do to get a reward.
In the case of Wendy Farris it's actually even more blatant than that as the dog handler actually touched the car, after which the dog looks at the owner and "finds drugs" immediately in exactly that location. Surprise surprise. This is at the end of the second pass.
A clue to the dog about as subtle as a brick in the face. The best possible good faith interpretation I can give to this is spectacular ignorance in understanding how dogs work.
