Yea I share the same concerns about the "AI", but this sounds like a good thing. It's going through footage that wasn't going to be looked at (because there wasn't a complaint / investigation), and it's flagging things that should be reviewed. It's a positive step

What we should look into for this program is

• how the flags are being set, and what kind of interaction will warrant a flag
• what changes are made to training as a result of this data
• how the privacy is being handled, and where the data is going (ex. Don't use this footage to train some model, especially because not every interaction is out in the public)

Make it publicly accessible. It'll most certainly get watched and problems will be reported to be investigated further.

File a complaint, and you get to view the video. If nobody files a complaint, there is no need to view the video.

Indeed, nobody should be looking at the video unless a complaint is filed.

WE should be able to review it/see it ALL.

We pay these fucks to torture and kill with our tax $.

They should have nothing to hide from us.

Well I mean you could rig the cameras to turn on when the cop gets out of their car to break the footage into specific encounters where the cop had to interact with someone. Identify the files by the date, time, and badge number of the cop the camera is assigned to, and now you've got an easy to search database of footage whenever an incident is reported either by the cop because they had to issue paperwork for it or by whoever they were interacting with because they want to lodge a complaint.

While randomly selecting files not involved in ongoing investigation as potential training material could be helpful, we don't actually HAVE to have an assigned review resource to scan for bad behaviour or relevant material to investigations since in both cases someone is incentivized to start the process that will pull the relevant footage anyways.