However, I doubt that's a problem in practice. I'd assume these finger printers know what they're doing. It certainly seems so.
How could one make an experiment collecting lots of these finger prints and determine the false positive rate?
The worst part of this? Trying to hide from fingerprinting makes your fingerprint more unique
Didn't seem so in the experiment in the article.
Sure they'll be able to place you in the bucket "tor user", but is that really more narrow than what you'd get without Tor?
No, I don’t care about the one time you downloaded a gentoo iso over tor.
So if I fingerprint you on a site which is using my commercial fingerprint service, then I can sell your hash to other places and tell them all about your browsing habits. The more places run my fingerprinting service, the more data I can collect on you.
The first time I heard about fingerprinting was with EFF's panoptoclick, which stated how many hashes had been generated from visitors, and how many you shared with them.
Any educated person with sufficient math knows the mathematical structure of a hash will never be unique. Its a Galois field, or 'finite field' after all.
The core of this issue is the flawed but convincing belief promoted by an entire industry that if the probability is sufficiently low, its unique, and following these axioms if its unique its an individual person (eyeball).
Under that assumption, all you need to do is collect fields of information that are variable, and group them together such that it yields to a sufficiently low threshold, I think currently that threshold is about 1 per million. Its a great way to defraud advertisers if you think about it.
In my opinion its just flawed thinking but there are some real fanatics out there that subscribe to this.
For example applied probability is used as part of the protocol design when accounting for binary erasure channels in things like cell phones. You shouldn't be able to have communications blocked in only one direction, or altered without it being noticeable, but stingrays may have the ability to do this according to the limited documentation that has been released so far.
Probabilities in general have real problems fundamentally with validity. I think the most common one in use today is the Axiomatic approach, or the Frequentist Approach.
So I guess I'll ask, are you a believer?
The worry would be that the hash is unique to me (i.e. a fingerprint), but I don't see the evidence that it is.