3 ms·

It's not everything you type, just bits and pieces stripped of sequence, and most of what's saved is on your device only. They specify that what they send is on

by ewzimm 11y ago

It's not everything you type, just bits and pieces stripped of sequence, and most of what's saved is on your device only. They specify that what they send is only "some," and logically, everything typed into the billion devices they're shooting for would be too much data for even Microsoft to handle. Passwords and the like are probably identified by the type of field they're entered into, as Windows forms and websites have explicit identifiers for passwords, and it would be easy not to record this. I think this analysis is more for things like analyzing the frequency of terms and which characters tend to follow other characters, but I don't have much more information, and I agree that there needs to be more shared about how this works.

orclev 11y ago
So, here's a question then, assuming that's true, why send it? If its been so thoroughly anonymized that nothing useful can be gleaned from it, why bother sending it in the first place? If it is possible to derive useful info from it, that seems to directly contradict the idea that no sensitive information is going to be exposed (there being no realistic way to programmatically identity only sensitive information without some very specific domain knowledge). You can't have it both ways, either the info isn't useful in which case they have reason to transmit it, or else it's exposing potentially sensitive information and they shouldn't be transmitting it.
- ewzimm 11y ago
  I can only speculate, but I would guess it goes something like this: You type "the quick brown fox jumps over the lazy dog." You continue typing other stuff. 30 minutes later, there's a queue of random slices like: { "lazy dog", "chocolate donut", "brown fox" }. That gets encrypted and sent into a big pile of other data for a machine learning algorithm to develop better typing prediction. I would like to know if that's the gist of it or if there are other things happening. Hopefully there will at least be a developer blog or something like that to better explain it.
pdkl95 11y ago
That's what it may be sending today. What about next month after one of the forced automatic updates? If you authorize someone to take your data, that doesn't mean they have to do it right away. > only "some" 99% is still "some". Regardless, since when is "some" keyloging acceptable? > too much data That depends on a lot of factors. Text is small and trivial compared to most stuff on the internet these days. How many kilobytes per day of text does the median windows user even type? It would be even easier for MS to filter out stuff like video games (DirectInput?) to reduce the size. > I think Guessing is useless. What matters is the data going over the wire and what authorized MS to do in the future. > there needs to be more shared about how this works No, there needs to be explicit informed consent form the user.
- ewzimm 11y ago
  I'm just trying to objectively interpret a really small amount of data here, not make a justification for or against any particular policy. You're right that it's important to have informed consent about this kind of thing. That applies to Google, Facebook, or Amazon too. It does seem like it's pretty obvious here when there are so many statements about how Cortana gets to know you, your speech, your schedules, your flights, your interests, etc., and there's a delete button for everything too. Most people probably like all that, but I would like to see more robust privacy settings to easily disable tracking whenever it's desired.
- gambiter 11y ago
  > That's what it may be sending today. What about next month after one of the forced automatic updates? You mean, like this? # apt-get update && apt-get upgrade Granted, running that isn't 'forced', but given that OSS users don't _actually_ analyze the code in their software, it's basically the same thing. There's always a chance of something being slipped in later, no matter what OS you're on. Rampant paranoia and conspiracy theorizing does no one any good.
  - pdkl95 11y ago
    The "forced" you so easily dismiss is entirely the point. In case you haven't noticed, people choose when (and if) they upgrade, for a variety of complex reasons. One of those reasons is to wait and see if the early adopters complain of any issues. The problem of inspecting the code itself is a separate issue entirely, though at least in the area of Free Software that inspection is possiblea, and reproducible (deterministic) will simply the problem. Good luck doing the same with Windows (or any other closed software). > Rampant paranoia and conspiracy theorizing Now you're simply resorting to either insults (and/or wilful ignorance).
mikeash 11y ago
The trouble is that not every place you type a password will necessarily have something identifying it to the OS as a password field. For example, what if you're entering a sudo password into an ssh client, or saving a password into a text file (saved on an encrypted drive, surely) using a text editor?
- ewzimm 11y ago
  I'd like to know too. Maybe they compare typed data to a dictionary and throw away everything that doesn't match so that every slice contains only dictionary words that couldn't be identified as passwords. That's just speculation of course.
  - mikeash 11y ago
    It's hard to see the utility in a bunch of individual unassociated dictionary words being sent back. Maybe there is some that I just don't know about. However, my money is on them being more confident in their scrubbing than is warranted. There have been so many cases in the past where companies have scrubbed and anonymized data, only to find that they missed something critical.
    - ewzimm 11y ago
      Even the frequency of words used might be valuable for typing predicion, but I would guess they would probably group a few words together. I really would love to know how it works.
      - mikeash 11y ago
        Same here. If it has holes, it's important to know about them. If it doesn't, then it would be fascinating to understand how it works.

by ewzimm 11y ago

orclev 11y ago

So, here's a question then, assuming that's true, why send it? If its been so thoroughly anonymized that nothing useful can be gleaned from it, why bother sending it in the first place? If it is possible to derive useful info from it, that seems to directly contradict the idea that no sensitive information is going to be exposed (there being no realistic way to programmatically identity only sensitive information without some very specific domain knowledge). You can't have it both ways, either the info isn't useful in which case they have reason to transmit it, or else it's exposing potentially sensitive information and they shouldn't be transmitting it.

ewzimm 11y ago
I can only speculate, but I would guess it goes something like this: You type "the quick brown fox jumps over the lazy dog." You continue typing other stuff. 30 minutes later, there's a queue of random slices like: { "lazy dog", "chocolate donut", "brown fox" }. That gets encrypted and sent into a big pile of other data for a machine learning algorithm to develop better typing prediction. I would like to know if that's the gist of it or if there are other things happening. Hopefully there will at least be a developer blog or something like that to better explain it.

pdkl95 11y ago

That's what it may be sending today. What about next month after one of the forced automatic updates? If you authorize someone to take your data, that doesn't mean they have to do it right away. > only "some" 99% is still "some". Regardless, since when is "some" keyloging acceptable? > too much data That depends on a lot of factors. Text is small and trivial compared to most stuff on the internet these days. How many kilobytes per day of text does the median windows user even type? It would be even easier for MS to filter out stuff like video games (DirectInput?) to reduce the size. > I think Guessing is useless. What matters is the data going over the wire and what authorized MS to do in the future. > there needs to be more shared about how this works No, there needs to be explicit informed consent form the user.

ewzimm 11y ago
I'm just trying to objectively interpret a really small amount of data here, not make a justification for or against any particular policy. You're right that it's important to have informed consent about this kind of thing. That applies to Google, Facebook, or Amazon too. It does seem like it's pretty obvious here when there are so many statements about how Cortana gets to know you, your speech, your schedules, your flights, your interests, etc., and there's a delete button for everything too. Most people probably like all that, but I would like to see more robust privacy settings to easily disable tracking whenever it's desired.
gambiter 11y ago
> That's what it may be sending today. What about next month after one of the forced automatic updates? You mean, like this? # apt-get update && apt-get upgrade Granted, running that isn't 'forced', but given that OSS users don't _actually_ analyze the code in their software, it's basically the same thing. There's always a chance of something being slipped in later, no matter what OS you're on. Rampant paranoia and conspiracy theorizing does no one any good.
- pdkl95 11y ago
  The "forced" you so easily dismiss is entirely the point. In case you haven't noticed, people choose when (and if) they upgrade, for a variety of complex reasons. One of those reasons is to wait and see if the early adopters complain of any issues. The problem of inspecting the code itself is a separate issue entirely, though at least in the area of Free Software that inspection is possiblea, and reproducible (deterministic) will simply the problem. Good luck doing the same with Windows (or any other closed software). > Rampant paranoia and conspiracy theorizing Now you're simply resorting to either insults (and/or wilful ignorance).

mikeash 11y ago

The trouble is that not every place you type a password will necessarily have something identifying it to the OS as a password field. For example, what if you're entering a sudo password into an ssh client, or saving a password into a text file (saved on an encrypted drive, surely) using a text editor?

ewzimm 11y ago
I'd like to know too. Maybe they compare typed data to a dictionary and throw away everything that doesn't match so that every slice contains only dictionary words that couldn't be identified as passwords. That's just speculation of course.
- mikeash 11y ago
  It's hard to see the utility in a bunch of individual unassociated dictionary words being sent back. Maybe there is some that I just don't know about. However, my money is on them being more confident in their scrubbing than is warranted. There have been so many cases in the past where companies have scrubbed and anonymized data, only to find that they missed something critical.
  - ewzimm 11y ago
    Even the frequency of words used might be valuable for typing predicion, but I would guess they would probably group a few words together. I really would love to know how it works.
    - mikeash 11y ago
      Same here. If it has holes, it's important to know about them. If it doesn't, then it would be fascinating to understand how it works.