3 ms·

What if the target (being monitored by said van) took off and "lost" the van in the sidestreets of US city? The FBI would be criticized for it's ineffective su

What if the target (being monitored by said van) took off and "lost" the van in the sidestreets of US city? The FBI would be criticized for it's ineffective surveillance techniques. The plan is just an upgrade - the laws and controls (or lack thereof) are what is important - not so much the tech.

Retric 11y ago
The FBI can easily follow one suspect, but it's an expensive provision. The issue is when you decrease the costs you can start following thousands of suspects for the same net cost so it takes a much lower standard of evidence to start. Suspect > suspect’s friend > friend of friend and you quickly hit the point where everyone is being watched. Worse large lists quickly become meaningless. Both the 9/11 hijackers and the Boston Marathon bombers where on a list, but that list was simply too large to be useful because 99.9% of everyone on those lists ended up doing nothing. Now consider something like a DNA database. With 100 million random people on it a 99.999% accurate test going to report a false positive every single time and most tests are nowhere near that accurate.
- genericuser 11y ago
  Getting from the admitted targeted surveillance with to 'everyone is being watched' seems like a bit of a slippery slope argument to me. I think, as you point out, that the overwhelming quantity of Data that would be produced by such an approach prevents this by creating a cost/benefit equilibrium for the FBI on the number of people they monitor with any regularity to build their cases.
  - Retric 11y ago
    License plate scanners are actually just scanning everyone so there is no slippery slope involved. Right now we are all being watched. It's a pure cost issue; camera + OCR + database lookup is really cheap so just tag everyone.
    - genericuser 11y ago
      How did we go from the FBI flying aircraft with video cameras to license plate scanners which scan 'everyone'? I ask because there is no mention of license plate scanners in any of the parent comments to this, or the article. Also while I assume by everyone you mean everyone who passes through the field of view of the scanner, and not actually everyone; phrasing things in false absolutes like that does not further your point.
      - Retric 11y ago
        It's hard to directly read a license plate from an aircraft. However that's not necessary as the plates are being recorded as part of a separate system all it takes is continuous footage of a car from aircraft and you are recording everyone. https://www.aclu.org/issues/privacy-technology/location-tracking/automatic-license-plate-readers https://www.aclu.org/issues/privacy-technology/location-trac... Basically, aircraft follows car X, it passes by a scanner, you now know every car with timestamp that passed by the scanner and can look it up on the video. Doing this automatically is just a question of having enough video coverage and some fairly basic video analysis. They don’t even need to do this in real time as you can write the software after the fact. In other words because they already have license plate scanners and are doing OCR + lookup they just need a video of the cars after that point. Note: I doubt they have nearly enough footage to track everyone right now, but the cheaper it is to add one more camera and archive the footage the closer you get to that point. PS: You can even back track cars so if you have a lot of footage of a car that then passes near a scanner you then know its path from wherever the footage starts.
        genericuser 11y ago
        You completely ignored my question to you. I asked how you got to the topic of license plate readers, I know how they work and an explanation of how they work does nothing to answer the question of how you transitioned to that topic.
        Retric 11y ago
- cma 11y ago
  How does more data hurt an investigation, if you are statistically literate? What is the conditional probability on that DNA test when it turns up a match with someone who also knew and was in recent contact with the murdered person and had a motive? Same with watch lists, doesnt all the ineffectiveness go away if they have tiering or triage system?
  - Retric 11y ago
    You’re now basing your case on circumstantial evidence which is considered near worthless for a reason. If you have 5 suspects and then do DNA testing that's solid evidence. Ex: Paternity test. But, if you find someone with DNA testing you need to completely ignore that evidence after that point as 100% of its predictive power has already been used. PS: To show just how easy it is to mess this stuff up. Suppose a murder occurred in NY city and the DB returned 20 people so you look at the closest 3 suspects and aha someone lives just 10 miles from the crime! However, note you already picked someone that was close by so you need to consider the odds that someone who lives close by happens to lives in the area. EX: A high percentage of people living in say NY state also live within 10 miles a crime in NY City.
    - TeMPOraL 11y ago
      That seems to be a textbook case of people failing to apply Bayes' theorem properly, not a fault of tests or technology. They, just like doctors analyzing cancer tests, need to remember to multiply one more number...
      - Retric 11y ago
        It's also the kind of information that's easily dropped or misunderstood by a jury at trial.
  - SomeStupidPoint 11y ago
    > Same with watch lists, doesnt all the ineffectiveness go away if they have tiering or triage system? If properly used, perhaps. However, the standards for government forensics have been... questionable. Generally speaking, government "tests" seem more focused on confirming the prosecutor or investigator's theory than on actually resolving truth about facts. You seem to be assuming a great deal of statistical literacy on the part of people utilizing these database tools, which I'm not sure is a solid assumption.

Retric 11y ago

The FBI can easily follow one suspect, but it's an expensive provision. The issue is when you decrease the costs you can start following thousands of suspects for the same net cost so it takes a much lower standard of evidence to start. Suspect > suspect’s friend > friend of friend and you quickly hit the point where everyone is being watched. Worse large lists quickly become meaningless. Both the 9/11 hijackers and the Boston Marathon bombers where on a list, but that list was simply too large to be useful because 99.9% of everyone on those lists ended up doing nothing. Now consider something like a DNA database. With 100 million random people on it a 99.999% accurate test going to report a false positive every single time and most tests are nowhere near that accurate.

genericuser 11y ago
Getting from the admitted targeted surveillance with to 'everyone is being watched' seems like a bit of a slippery slope argument to me. I think, as you point out, that the overwhelming quantity of Data that would be produced by such an approach prevents this by creating a cost/benefit equilibrium for the FBI on the number of people they monitor with any regularity to build their cases.
- Retric 11y ago
  License plate scanners are actually just scanning everyone so there is no slippery slope involved. Right now we are all being watched. It's a pure cost issue; camera + OCR + database lookup is really cheap so just tag everyone.
  - genericuser 11y ago
    How did we go from the FBI flying aircraft with video cameras to license plate scanners which scan 'everyone'? I ask because there is no mention of license plate scanners in any of the parent comments to this, or the article. Also while I assume by everyone you mean everyone who passes through the field of view of the scanner, and not actually everyone; phrasing things in false absolutes like that does not further your point.
    - Retric 11y ago
      It's hard to directly read a license plate from an aircraft. However that's not necessary as the plates are being recorded as part of a separate system all it takes is continuous footage of a car from aircraft and you are recording everyone. https://www.aclu.org/issues/privacy-technology/location-tracking/automatic-license-plate-readers https://www.aclu.org/issues/privacy-technology/location-trac... Basically, aircraft follows car X, it passes by a scanner, you now know every car with timestamp that passed by the scanner and can look it up on the video. Doing this automatically is just a question of having enough video coverage and some fairly basic video analysis. They don’t even need to do this in real time as you can write the software after the fact. In other words because they already have license plate scanners and are doing OCR + lookup they just need a video of the cars after that point. Note: I doubt they have nearly enough footage to track everyone right now, but the cheaper it is to add one more camera and archive the footage the closer you get to that point. PS: You can even back track cars so if you have a lot of footage of a car that then passes near a scanner you then know its path from wherever the footage starts.
      - genericuser 11y ago
        You completely ignored my question to you. I asked how you got to the topic of license plate readers, I know how they work and an explanation of how they work does nothing to answer the question of how you transitioned to that topic.
        Retric 11y ago
cma 11y ago
How does more data hurt an investigation, if you are statistically literate? What is the conditional probability on that DNA test when it turns up a match with someone who also knew and was in recent contact with the murdered person and had a motive? Same with watch lists, doesnt all the ineffectiveness go away if they have tiering or triage system?
- Retric 11y ago
  You’re now basing your case on circumstantial evidence which is considered near worthless for a reason. If you have 5 suspects and then do DNA testing that's solid evidence. Ex: Paternity test. But, if you find someone with DNA testing you need to completely ignore that evidence after that point as 100% of its predictive power has already been used. PS: To show just how easy it is to mess this stuff up. Suppose a murder occurred in NY city and the DB returned 20 people so you look at the closest 3 suspects and aha someone lives just 10 miles from the crime! However, note you already picked someone that was close by so you need to consider the odds that someone who lives close by happens to lives in the area. EX: A high percentage of people living in say NY state also live within 10 miles a crime in NY City.
  - TeMPOraL 11y ago
    That seems to be a textbook case of people failing to apply Bayes' theorem properly, not a fault of tests or technology. They, just like doctors analyzing cancer tests, need to remember to multiply one more number...
    - Retric 11y ago
      It's also the kind of information that's easily dropped or misunderstood by a jury at trial.
- SomeStupidPoint 11y ago
  > Same with watch lists, doesnt all the ineffectiveness go away if they have tiering or triage system? If properly used, perhaps. However, the standards for government forensics have been... questionable. Generally speaking, government "tests" seem more focused on confirming the prosecutor or investigator's theory than on actually resolving truth about facts. You seem to be assuming a great deal of statistical literacy on the part of people utilizing these database tools, which I'm not sure is a solid assumption.