Human Reviewers Can't Keep Up With Police Bodycam Videos. AI Now Gets the Job 26
Tony Isaac shares a report from NPR: After a decade of explosive growth, body cameras are now standard-issue for most American police as they interact with the public. The vast majority of those millions of hours of video are never watched -- it's just not humanly possible. For academics who study the everyday actions of police, the videos are an ocean of untapped data. Some are now using 'large language model' AI's -- think ChatGPT -- to digest that information and produce new insights. [...] The research found the encounters were more likely to escalate when officers started the stop by giving orders, rather than reasons for the interaction. While academics are using AI from anonymized videos to understand larger processes, some police departments have started using it to help supervise individual officers -- and even rate their performance. An AI system mentioned in the report, called TRULEO, assesses police officers' behavior through automated transcriptions of body camera footage. It'll evaluate both positive and negative conduct during interactions, such as traffic stops, and provide feedback to officers. In addition to flagging issues like swearing or abusive language, the AI can also recognize instances of professionalism.
The Future? (Score:2)
Re: (Score:2)
Re: (Score:2)
An AI reviews the copcam footage and flags incidents for human review.
No cop is disciplined solely because the AI says they're a bad cop.
An AI review is better than no review.
There are lots of bad cops out there, and 10% of the cops cause 90% of the misconduct.
Re:The Future? (Score:4, Insightful)
The bad cops wouldn't be an issue for very long if the good cops didn't look the other way for them. Cops are indoctrinated into an 'us vs. them' mindset that can make even an otherwise fairly decent person help cover up unacceptable things.
Re: (Score:2)
The bad cops wouldn't be an issue for very long if the good cops didn't look the other way
The good cops are less likely to look the other way if they know the video will be reviewed, and they may be disciplined for failing to report misconduct.
Re: (Score:2)
It actually works more the other way - officers wearing body cams have fewer incidents of aggressive offenders because they know their actions will be documented.
It's amazing to hear officers before and after a body cam pilot project. Before they're all "my privacy in the bathroom!" and "I won't be able to use discretion to let people off for little things" and after they're totally sold on the idea.
You know, except for the really bad cops who then find a need for convenient 'malfunctions' of the equipment
Re: (Score:2)
When the so-called "good" officers ignore the actions of the bad actors, they enter into a knowing conspiracy that violates their oath to uphold the law. At that point they are no longer "good", but parties to a coverup. Unless a department is relatively small, has really outstanding leadership, and is just lucky, it's pretty much a lock that the culture of coverup has tainted the entire organization. It hard not to conclude that a significant
Re: (Score:2)
I lie that it has the potential for the carrot AND the stick here as well.
source code or you must aquit! (Score:1)
source code or you must aquit!
moot (Score:2)
The only thing this will do is bring videos of interest to the attention of humans. No source code needed for that.
Good! (Score:3)
Finally a good use for AI! Seriously, of all the shit that AI has been promoted for, this is perhaps the only one where it actually solves an existing problem.
Re: (Score:1)
I'm not sure what problem it solves. If there is use of force (a scuffle), the video is supposed to be reviewed by humans per policy in most police departments.
That means bots would only be useful for watching the trivial encounters, which usually have no reason to be reviewed.
An exception may be if yelling or cussing is detected for a non-scuffle encounter, but that should merely alert the human inspectors to review it, not be used for grading itself.
Using it for general grading has lots of risk. The TFA g
Re: (Score:2)
If there is use of force (a scuffle), the video is supposed to be reviewed by humanshi
What if the scuffle isn't reported?
You're saying that AI reviews are unnecessary as long as policies are followed and all cops are good cops.
What if they're not?
You lack imagination. (Score:1)
Thus, detecting yelling/cussing that didn't lead to physical conflict is the only clear use I can see for police departments so far.
Then you aren't thinking very hard. Here is a list of other videos they'd potentially flag and why:
Shifty body language: Lots of fidgeting, avoiding eye contact, or hiding hands.
Dodging cops: Changing direction or trying to avoid officers.
Too many glances: Constantly watching officers without engaging.
Hiding stuff: Trying to conceal objects in a way that seems off.
Odd group hangouts: People gathering in unusual or secluded spots.
Evasive answers: Giving weird, vague, or inconsistent responses to quest
Re: (Score:3)
The AI is for reviewing cop behavior, not perp behavior.
Seems unnecessary (Score:3)
Review is only important if there is an incident, or randomly to ensure the systems are functioning.
After that, it's a retention policy and storage/retrieval issue: How much video can we afford to keep against the need to use it as evidence, is that enough, and if not can we get more funding?
Oh, and not by cops. Police equipment ought to be stored, issued, tracked, and monitored by an arms-length agency that is prohibited from unmonitored contact with members of the police force. Then there's no awkward questions about whether a cop 'accidentally' destroyed inconvenient evidence.
Re: (Score:2)
Review is only important if there is an incident
The point of TFA is that the reviews can be used to prevent incidents by discovering patterns of behavior that lead to confrontation.
Municipalities pay millions in lawsuits and settlements because of cop misconduct. So they have a clear incentive to reduce it.
90% of misconduct is caused by 10% of the cops. Female cops have far fewer incidents because they try to de-escalate, whereas a male cop often does the reverse.
10% of American families have incidents of domestic abuse. For cop families, it's 40%. Those
Re: (Score:2)
I don't know if review is only important if there is an incident. Reviewing footage maybe how you know there was an incident.
Now I agree that in a case of an incident, you probably want human review rather than AI review.
But even just using an automated system to flag "nothing happening for the next 3 minutes" is very useful already. But you probably can get a lot more out of that. You probably can extract insult rate, drunken behavior, escalations, and things like that automatically. Force-level statistic
Re: (Score:2)
Reviewing footage maybe how you know there was an incident.
Excepting dead people, if the subject of some police contact doesn't know that there has been an incident, then there hasn't been one.
Re: (Score:2)
There is a simple solution. A cop “forgets” turn on the camera or the footage accidentally gets deleted?
Case dismissed.
First things first. (Score:4)
Looking forward to stories in 5 years (Score:2)