33 Bits of Entropy

for any purpose, at any time, provided that following collection
of such location and speed information identifiable to your Vehicle

They store the data tied to your identity. A data breach (quite common these days...) would be a Big Deal. GPS tracks of everywhere you've gone in your car, ever? That's worth quite a bit of money in the right hands.

He gave no evidence that they were not anonymizing the data properly,
he just assumed they were not.

Zipcode, birthday, gender: identifies 87% of Americans[1]. Your (Home,Work) gps tuple? Unique[2][3]. His assumption is quite safe; every "anonymized" dataset that's been released into the public (that I know of) has been de-anonymized. Why would this one be special?

1) http://arstechnica.com/tech-policy/news/2009/09/your-secrets-live-online-in-databases-of-ruin.ars

2) http://crypto.stanford.edu/~pgolle/papers/commute.pdf

3) http://33bits.org/2009/05/13/your-morning-commute-is-unique-on-the-anonymity-of-homework-location-pairs/

EDIT: In response to parent edit and below comments

I have no proof of these, but factoids I believe to be true (so feel free to base a research paper on them :D)

1) To identify commuters: (Highway-Entrance-Location, Average-Highway-Entrance-Time, Highway-Exit-Location, Average-Highway-Exit-Time) -> some derived values: approximate (home,work), average speed, average driving aggression

2) Really, now that I think about it, any dataset where multiple gps tracks (for a single person) are tied…

NoScript, RequestPolicy, Adblock Plus, RefControl, UAControl, Do-not-track cookie-disabled Mozilla Firefox user here.

Regulating Internet technologies to this level of pedantic granularity will ensure that spammers, scammers, crackers and fraudsters have an effective monopoly on privacy-busting technology. The incentive for software to implement correct technological solutions will be taken away if an honesty box Do-not-track approach is considered adequate. I want my browser to have a maximal number of reasons (including a multi billion dollar advertising industry) to address the technical concerns highlighted by Panopticlick[1] or privacy experts that "get it"[2]. These technical problems will be exploited by unwanted parties on the Internet. Exploitation will occur legally in other jurisdictions out of reach of EU laws.

The full text of the directive can be found at §66 on page 20 of Directive 2009/136/EC at [3].

[1] https://panopticlick.eff.org/

[2] http://33bits.org/

[3] http://eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri=OJ:L:2009:337:0011:0036:EN:PDF

IIRC, Larry Detweiler's sock-puppeting attempts on the cypherpunks mailing list were unmasked by other cypherpunks with this technique in the early 1990s.

Detweiler, natively paranoid and geographically isolated in Colorado, had come to the conclusion that most of the prominent cypherpunks posters were sock-puppets of a single person he called "Medusa" — he invented the word "pseudospoofing" to describe the phenomenon, although "sock-puppet" is the more common current term — and resolved to use the same technique in response.

I never read the actual word-frequency analyses, just Detweiler complaining about having had it used against him, which I assume he didn't imagine.

There was a recent paper (two or three years ago?) that claimed that people could usually defeat this kind of linguistic analysis simply by trying to change their writing style. I'm skeptical that this is applicable in general.

http://33bits.org/ is a pretty good blog on the unreasonable difficulty of anonymity in a computerized world.

It's actually pretty debatable whether this year's KDD Cup will really help the science of music recommendation:

http://musicmachinery.com/2011/02/22/is-the-kdd-cup-really-music-recommendation/

Because it's entirely anonymised, not just the users but the artists too -- c.f. Netflix's problems with deanonymization:

http://33bits.org/2010/03/15/open-letter-to-netflix/

This means you can't use any interesting characteristics of the music itself, or the associated metadata, to aid the recommendations. All the interesting domain knowledge is stripped out, which likely means the best solutions still won't work as well as algorithms that use metadata (like Last.fm's) or content analysis (like Pandora's) or both, and certainly won't lead to any particularly interesting insights about what drives people's tastes.

Disclaimer: I work at Last.fm

Hey all,

I'm one of the authors of the Netflix and social network de-anonymization papers referenced, and the author of http://33bits.org. I also had the pleasure of having many talks with law professor Paul Ohm over the technical aspects of his excellent paper (which another commenter has posted a link to.)

Paul's paper is a great example of people in the law community "getting it." (He happens to have a CS degree and is a hobbyist perl hacker!) We need more people like Paul, and we equally also need tech people who understand law and policy. I encourage everyone to give Paul's paper a quick read, gain awareness of the issues such why the current privacy laws are wrongheaded and what needs to be changed, and be on the look-out for ways to change things (as consumers, as tech influencers, as citizens). Cheers.

Arvind Narayanan has made this observation in terms of the difficulty or impossibilities of anonymizing databases:

http://33bits.org/

However, he didn't propose a particular list of 33 questions whose answers are maximally statistically independent. Even with things that are totally uncorrelated, you do get some chance overlap where people happen to have the same answers on a small number of them.

For example, if we took the first 33 bits of the SHA256 of one's HN username, you and I happen to agree in positions 7, 8, 11, 13, 15, 17, 18, 19, 21, 26, 27, 28, 29, 31, and 32. (So we could say that each additional bit of that SHA256 value doesn't actually provide a full additional bit of distinguishing power relative to the previous ones.)