Shouldn't be difficult to build a tool to randomly obfuscate writing you want to conceal. Swapping words/sentences with semantic equivalents should be a rather trivial exercise. If the tool allows the human to revise and tune it, it's even easier.
This was the approach the "Shadow Brokers" took. People loved speculating on their native language and how they were faking their poor English skills when the reality is they obfuscated it.
That might be like saying it's easy to make a Facebook clone in a day.
Any tool that could do it well could translate perfectly between languages I'd guess since it'd need perfect understanding of the writing. Else you introduce leaks.
Are you claiming the author (Muse) said Mr. Satoshi Nakamoto was forced to leave the Illuminati?
If so, smell test is very bad.... but source?
Problem here is, why not link this? You are smell testing way worse by not linking.
Back to the fiction, like Fairy Tales some times there's a (thought) lesson even if it's made up. And even if the supposed lesson is wrong it makes you think.
It assumes a number of things not in evidence, some of which are fairly improbable:
1. That the NSA has access to a sizable chunk of all Internet communications, sufficient to have the "known" communications to match to the "unknown" samples.
a. The story references PRISM, but public reports describe PRISM as only being an implementation of section 702 authority, and probably not bulk (i.e. 100%, or whatever) collection of all email.
b. It also references MUSCULAR, which was bulk interception but, frankly, the way this is written, it comes off as name-dropping, and isn't otherwise a meaningful reference. (It's like when a scifi author says that faster-than-light travel was made possible once scientists perfected hyper-tachyon capture--it sounds cool, but I just made up those words.)
2. That the NSA interception makes it easy to tie existing samples to known identities. This is a massive data mining problem. I could imagine an implementation where someone can (in a computationally expensive manner) rank writing samples by similarity to a target; an analyst might then manually pore over similar samples for clues to identity. So in this case, "Satoshi" writing might be very close to emails sent by some.real.name@example.com, which spills the beans quite quickly, but there's no reason to assume that is so.
3. That stylometrics are sufficiently advanced to distinguish between 1Bn English Internet users. I'm not well-read on this aspect, but it seems like the most improbable part; see e.g. https://www.researchgate.net/profile/Ahmed_Abbasi4/publicati.... The few papers I just dug up abstracts for seem to suggest error rates in the range of 5-15% among a limited number (~100) of authors, so the idea that the NSA could attribute writing out of 1B potential authors is pretty far-fetched.
I assumed the point of this story was to use fiction to illustrate something that's sort of hypothetically plausible, and to illustrate the privacy risks such a scenario poses. But I was annoyed by the author's claim to be writing fact. :)
I'd assume they had a (small) pool of candidates to start with and finding the author of texts in the group of authors is very far from identifying every person on Earth by linguistic style (as the article states)
13 comments
[ 3.1 ms ] story [ 16.5 ms ] threadAny tool that could do it well could translate perfectly between languages I'd guess since it'd need perfect understanding of the writing. Else you introduce leaks.
Why is this even here?
There is nothing in it that would be outside of the capabilities of the NSA.
IMO I tend to believe it is not fiction, without even looking at the authors resume.
It passes many smell tests that a lot of HN articles clearly fail.
If declared as fiction, not if stated as a fact. Right now that piece is as fake news as any other fake news.
BTW Mr. Muse also omits how once he was identified by the NSA, Mr. Satoshi Nakamoto was forced to leave the Illuminati ...
If so, smell test is very bad.... but source?
Problem here is, why not link this? You are smell testing way worse by not linking.
Back to the fiction, like Fairy Tales some times there's a (thought) lesson even if it's made up. And even if the supposed lesson is wrong it makes you think.
Naah, I am claiming that my sources maybe told me that (and they also maybe told me that this piece of info was shared with Mr.Muse).
To cite (maybe) Isaac Asimov:
https://en.wikiquote.org/wiki/Absurdism
"Those people who think they know everything are a great annoyance to those of us who do."
1. That the NSA has access to a sizable chunk of all Internet communications, sufficient to have the "known" communications to match to the "unknown" samples.
a. The story references PRISM, but public reports describe PRISM as only being an implementation of section 702 authority, and probably not bulk (i.e. 100%, or whatever) collection of all email.
b. It also references MUSCULAR, which was bulk interception but, frankly, the way this is written, it comes off as name-dropping, and isn't otherwise a meaningful reference. (It's like when a scifi author says that faster-than-light travel was made possible once scientists perfected hyper-tachyon capture--it sounds cool, but I just made up those words.)
2. That the NSA interception makes it easy to tie existing samples to known identities. This is a massive data mining problem. I could imagine an implementation where someone can (in a computationally expensive manner) rank writing samples by similarity to a target; an analyst might then manually pore over similar samples for clues to identity. So in this case, "Satoshi" writing might be very close to emails sent by some.real.name@example.com, which spills the beans quite quickly, but there's no reason to assume that is so.
3. That stylometrics are sufficiently advanced to distinguish between 1Bn English Internet users. I'm not well-read on this aspect, but it seems like the most improbable part; see e.g. https://www.researchgate.net/profile/Ahmed_Abbasi4/publicati.... The few papers I just dug up abstracts for seem to suggest error rates in the range of 5-15% among a limited number (~100) of authors, so the idea that the NSA could attribute writing out of 1B potential authors is pretty far-fetched.
I assumed the point of this story was to use fiction to illustrate something that's sort of hypothetically plausible, and to illustrate the privacy risks such a scenario poses. But I was annoyed by the author's claim to be writing fact. :)
I've always thought this as well. It really makes the most sense.