3 ms·

In addition to the other reasons listed, it also probably wouldn't be unique enough. It seems that their records are keyed off of ONLY: * Birthdate * Last

by mjevans 10y ago

In addition to the other reasons listed, it also probably wouldn't be unique enough. It seems that their records are keyed off of ONLY: * Birthdate * Last Name * First name If I were going to take the bother of changing my name I'd make the middle name, a part I can drop from most non-legal/contractual areas, the actual unique identifier. By which I mean I'd have the "middle name" be a short normal name post-fixed by a base26 encoded hash; I should probably also add 'son' at the end of that... say it's European...

dwighttk 10y ago
It seems like middle initial was also used... she knew the other person was Lisa S Davis but had to figure out what her middle name was.
- mjevans 10y ago
  Let's do some back of the envelope math: 04 bits = Middle initial (It's actually 5, but I imagine the common ones cluster) 10 bits = First Name 10 bits = Last Name 120 bits = 42 adult years (12*10): 3-years ~=1024 slots ~= 10 bits A naive approach might be to add all of these together, except that's unaware of the Birthday Paradox (aka Birthday Problem). https://en.wikipedia.org/wiki/Birthday_problem https://en.wikipedia.org/wiki/Birthday_problem The sets are larger, but even with the addition of those extra 4 bits there's still going to be overlap in a non-trivial population. When is there a "good" chance of a match in all of the fields that matter? What is a "good" chance? What is an acceptable collision? I think most sane people would agree that the final question is the important one. In matters that can derail someone's life and send them to jail an acceptable false positive rate is ZERO. Therefore answering the preceding questions is largely meaningless, we already have example proof of at least one failure, which is enough to declare the system inherently broken.
  - CodesInChaos 10y ago
    120 bits for the birth date is much too big. It's closer to 14 bits (taking your 42 year figure it's log2(42*365)=13.9).
    - mjevans 10y ago
      You're right, I got confused expanding the number of bits, I'd edit my post... but it seems I can't anymore. shrug That makes the birthday paradox / problem a LOT worse as well.

by mjevans 10y ago

dwighttk 10y ago

It seems like middle initial was also used... she knew the other person was Lisa S Davis but had to figure out what her middle name was.

mjevans 10y ago
Let's do some back of the envelope math: 04 bits = Middle initial (It's actually 5, but I imagine the common ones cluster) 10 bits = First Name 10 bits = Last Name 120 bits = 42 adult years (12*10): 3-years ~=1024 slots ~= 10 bits A naive approach might be to add all of these together, except that's unaware of the Birthday Paradox (aka Birthday Problem). https://en.wikipedia.org/wiki/Birthday_problem https://en.wikipedia.org/wiki/Birthday_problem The sets are larger, but even with the addition of those extra 4 bits there's still going to be overlap in a non-trivial population. When is there a "good" chance of a match in all of the fields that matter? What is a "good" chance? What is an acceptable collision? I think most sane people would agree that the final question is the important one. In matters that can derail someone's life and send them to jail an acceptable false positive rate is ZERO. Therefore answering the preceding questions is largely meaningless, we already have example proof of at least one failure, which is enough to declare the system inherently broken.
- CodesInChaos 10y ago
  120 bits for the birth date is much too big. It's closer to 14 bits (taking your 42 year figure it's log2(42*365)=13.9).
  - mjevans 10y ago
    You're right, I got confused expanding the number of bits, I'd edit my post... but it seems I can't anymore. shrug That makes the birthday paradox / problem a LOT worse as well.