Hacker News new | ask | show | jobs
by kwertyoowiyop 390 days ago
This will make murder mystery shows harder to write. Even now they usually put in some line about how they don’t have traffic camera coverage in the critical area, and they ignore getting location data from suspects’ phones.
2 comments

I have always thought that futuristic police state movies and shows underestimate how oppressive a fully capable, automated surveillance apparatus could be.

Movies like Minority Report try to show the surveillance state as being a struggle to overcome, but it is still always too easy. Computers don’t get distracted, scale perfectly, and can run 24-7. You can’t just sneak away with your head down, because the machines would have tracked you into a place, would know exactly who is in every building, and would be able to associate the person exiting a building with the person who went in. They wont forget.

Look at how quickly the cops tracked Luigi Mangione. It’s not clear how much face recognition (as opposed to manual search) contributed to that, but even for a person who wore a mask, all it took was a slip up where he took the mask off in one place.

I am not saying this is a bad thing in the case of a pre-planned murder. But it does make it obvious how hard it might be to evade notice in the future, assuming you are doing it for more legitimate privacy reasons.

If you go into a building and change clothes, they will not remember. Of course, we're assuming that the place you went into does not have cameras accessible to the system. At some point, building codes/permits will start requiring cameras specifically to feed into this system.
If the system sees someone exit a building and that person does not match someone the system saw enter the building, the person could be flagged and tracked as possibly being someone else who entered the building. Once the system accounts for everyone else who entered the building (by seeing when they exit), the system would be able to correlate that the unknown person who exited is the same as the only unaccounted for person who exited
If that building in an apartment building? Someone from an apartment building might not enter before leaving on the same day. How far back in the camera's history does the system look for that person entering?
This is the point... computers can watch 24/7 and never forget or get distracted. The system could look weeks back, and will never forget a person.
I know storage is cheap, but I think you run into storage space issues very rapidly if you want high-definition, 24/7 surveillance of every apartment building stretching back weeks.
Storing video footage from weeks back does not seem feasible. I assume footage would get overwritten after a specific time period.
until someone finds that certain cameras are unavailable, or that something has obstructed the view for a significant enough time to cause reasonable doubt
Plastic surgery clinics would likely have an issue
Changing clothes does nothing against gait recognition.
Change your shoes with a rock/tack in one of them. Tie your laces together. There's all sorts of ways to change your gait
Well, then the system could flag you because it doesn't know who you are. Someone walks out of a building with a gait that doesn't match the gait of anyone known to have entered the building. It would follow the person to the next building and the next, until it figures out who they are via the process of elimination (or if it see you go back to your normal gait). Then, the system would note (and store forever) that you have multiple gaits, and would never fall for the trick again.
you have way more confidence in AI than I do. "never fall for the trick again" bwahahaha. it can't even tell the same answer twice, or tell you the correct answer the first time for everything.
All of which give you an abnormal gait that the system should immediately flag for immediate inspection as someone trying to hide their gait.
In Los Angeles, a firebug sets a morning fire. 4 hours later and 50 miles away, they set another fire. The firebug is apprehended late in the afternoon. Good policework, I wonder how they did that. /s
AI generated deepfakes solve this you can't trust the cameras footage now