Uncle Sam to unmask anonymous writers using AI The US intelligence community has launched a program to develop artificial intelligence that can determine authorship of anonymous writing while also disguising an author's identity by subtly altering their words. The Human Interpretable Attribution of Text Using Underlying Structure (HIATUS) program from the Intelligence …
I believe you're thinking of the FBI's COINTELPRO operation.
Yes, there's a real tendency for intelligence programs of all sorts to devolve into circle-jerks, and automating them just makes that more probable.
We already have zombie botnets breaking into systems vulnerable to older exploits and distributing malware, with no one at the helm. That's likely to include some of the "we paid the ransom but received no key" ransomware infections – those BTC are probably just lost forever, in wallets that no one has the key for any longer. The same will happen with these systems, because they're pretty trivial to develop. So it's cheaper for adversaries to inject more noise than it is for observers to extract signal.
Hmm my cynical side wonders whether the goal of the intelligence services wanting an AI that is able to identify anonymous writers text is less about attributing text to a particular author and more about trying to identify whistleblowers or others that speak out against the government anonymously.
Yes, this is what it will mainly be used for. Someone is posting stuff somewhere on the Internet that some government wants to stop. They'll probably have a list of suspects but can't pin it down to a particular individual.
By running these posts through AI analysis and comparing them to existing samples from their suspects they'll be able to figure out who it is with a high enough degree of probability to make it worthwhile picking him up for interrogation, or just "disappearing" him.
To make this work you need a body of existing samples to compare to. Fortuitously for those seeking to apply this technology, the sorts of people most likely to be targets are also people who tend to write for a living - journalists, academics, etc.
AI as a means of enabling pervasive surveillance is just getting started and we can expect lots of stuff like this to appear more in future.
It's a terrible thing, and you've just proven why.
People need to be able to challenge the false narrative that there was a terrorist insurrection at the Capitol, without fear of the corrupt FBI* turning up heavily armed at 5am and scaring their kids.
You have however made me laugh by posting that anonymously.
*Since I'll no doubt get challenged on that point, see https://www.nbcnews.com/politics/national-security/two-4-warrants-letting-fbi-spy-ex-trump-aide-carter-n1121406 or, if you want something from just the last week, https://www.latimes.com/california/story/2022-09-23/fbi-beverly-hills-safe-deposit-box-raid-forfeiture-judge
Scroll to the next paragraph for what I really wrote.
The method you describe is great for hiding source text in programs created here, but doing so requires a lot of negotiation with other needs. The interpretation of these ideas does not seem to be easy to understand, and the people who propose them may not have a clear understanding of what they mean. Readers who don't know writers who do this are easily overlooked because they don't know how to write a series or how to read the news. As for author privacy, it only works if the author does not use offline translation software. Anyone with access to Google Docs has a unique fingerprint because they can identify someone who uses Google to translate. Unless Google decides not to release this data to the government (good luck), it's a more confusing combination than natural language.
The method you describe would work very well at disguising the source of text from the program being built here, but in accomplishing that goal, it would end up compromising too much on other requirements. I'm guessing the translated version of this comment wasn't easy to understand, and someone who produces this result is likely not to be clearly understood. Readers unaware that the writer has done this could easily dismiss them as being unable to write coherently, which might cause the information to go unread. For the privacy of the writer, this also fails, at least unless the writer only uses offline translation software (having tried some of the options there, I can say with confidence that this would be even harder to read if they do). Anyone with access to Google's logs can identify the connection that performed these translations and it's Google, so they have a bunch of fingerprinting techniques. Unless Google decides not to give that information to governments (good luck), it could end up being a stronger piece of evidence than the original fuzzy match from language alone.
runs their source text through a chain of different languages via Google Translate
Even if that produced useful output, it's a highly awkward and inefficient way of disguising the features that a typical DL-style stack of CNNs would use to identify authorship. Just create a model of a style that's significantly far from yours – by analyzing samples, or using a GAN or similar to create a synthetic one – and use that to post-process your output.
That's a sufficiently easy exercise that I imagine it will relatively soon be available as a "filter" for all sorts of smartphone text-communication apps and the like – if it isn't already. An advanced CS undergrad could cook one up.
Analysis of my posts would indicate British education to a particular level and in what subjects - with a reasonable age grouping. Geographical location and employer would also be indicated.
I use Anonymous so that my posts are judged on their own merit - without biasing people to a favourable or unfavourable reaction by knowing my El Reg identity used for other posts.
Using AI to attribute anonymous texts to known authors is something that academics have been working on for years for historic texts. This in turn is an improvement on the older manual methods. A typical application would be for example to try to attribute an anonymous poem to a known author (e.g. Shakespeare, or one of the Greek classics).
If they are doing anything new in the project being described in the story, I suspect it will be in terms of trying to do this in a way which was more suitable for mass surveillance, or to try to do it in cases where the author was trying to conceal his writing style.
This last point, uncovering authors who were trying to conceal their style, may be why they are doing joint a anonymize/de-anonymize project. By pitting them against each other they can get ahead and stay ahead of the AI arms race when it comes to de-anonymizing people.
Huo lung beffure-a peuple-a stert feeltering zeeur cumments thruough zee-a inchefferizer? Bork Bork Bork!
Here again going to say that Wikipedians have had to do this for years, to unmask manipulators. I'm thinking of one stock market analyst who skewed articles for years, finally unmasked in 2008, but persisting with multiple identities past 2011. Having fooled the Wikipedia Lares completely, a dozen plus people put together a 'profile' that included language usage in order to single out all the false identities and get the guy blocked permanently.
A large number of Wikipedians did not like this. They wanted every 'user' judged separately, without any prejudices, and in the best possible light in each case (userid). It took over a year to muster a convinced majority that this was one person with an incompatible agenda. By the end of that process many of the people instrumental in revealing the problem had left Wikipedia, tirelessly lambasted by those with high morals. The good guys were tainted beasts, they said.
The intersection of public and private identities will always present conundrums. But idealism must be tempered by reality. Some people do not like this. They want to treasure -- privilege -- their perfect dream of how people should be. A 'perfect' Ponzi scheme, that is.
Language analysis for identification is another tool needed in the real world of web anonymity. I've seen it absolutely required to stop shysters. What tools will you scorn while protecting yourself?
Blimeynthat's true. I try to respond here like one who lives in Ol Blighty, mainly so the proles will better understand me typing and that. And, I try to channel Sir Pterry quite often. Other sites I post to I will use different mannerisms that are a better fit for the site. I doubt that an AI would be able to tie me different nom de plumes to the same lerson, at least in my lifetime. But then, I's gettin old too!
uh8vC1yBot27MJCXi3ada5KHMJi1mx8ZqRC32Ra38P8zGLM9Ob6NgDUvYXCvMxaJGtejMhQtiFKb
8fYPIPuVatiNCRSdC7Ebgte7MP6BInYhENaFYhYv8VcV8TKL2BcjyZYhe58F8JYl4HoT0rKxkVYB
0nMFU5GJUZ2BWXOJYXeNOhs9gdAFoNgvaJ8vkhIHOJy5M5qVGpGFqF0TOVMLAxG1K3ClMHCJM3O1
gLebAZQNMpYF4fmR2xKDo9E3ERo5uvqZqxaduhKfeNApYNA96lYlwhKDGVUlMVcRqPatqt0ByVM5
Qbq3oFylQ3UHKpAxoRCzg3EZS1AzYnkjQ18pezGBGNEp6haLiLqNabMl0RYfYT2tsVi56fMFUL0J
gDa7SrEvCliVINEBaxmVoLw3ch4PgR2NkV4P
Yes, there is, at least for those people you want to see anything. You realize that this surveillance only is relevant if they have a copy of your messaging. If you encrypted it for private consumption, you're already fine. In fact if you didn't encrypt it but sent it for private consumption, you're still fine (either they didn't intercept it and don't have a copy or they did and know where it came from). This software would be used on something you released publicly, and encrypting something you publish just makes it difficult for anyone to read it.
Summary: encryption and language-anonymization are for different things. Having one when you need the other is not helpful.
McKinnon said that part of what HIATUS is doing is trying to demystify some of the unknowns around neural language models (the focus of HIATUS' efforts), which he said work well but are essentially black boxes that function without their developers knowing why they make a particular decision.
The Intelligence Community,... and there's a relatively recent and novel noble sounding, misleading secrecy shrouded euphemism for a whole Pandora's Box of private renegade rogue pirates and wannabe government spook sub-contractors betatesting their various versions and motley collections of comely attractive addictive wares out there in the ethereal spaces which virtually command and remotely control practically everything, for both the greater public good and its rivals, if ever there was and is such an IoT thing ..... and IARPA type enterprises would do well to realise that not all functioning black box developers are ignorant of the whys and the therefores of their activities, and neither invariably also always too, those of their rivals/contemporaries ..... the official opposition and/or market competition.
Bag one of those to support your team with access to ground breaking and Earth shattering product, and it's a veritable licence to print money 24/7/365, given what can then be so easily done.
:-) And quite perfect for that and those stuck in a rut and having a particularly and peculiarly bad present time requiring an unprecedented unscripted intervention on whatever scale is necessary.
PS .... Can the Conservative Party and its Members in Parliament and Offices of State and in a Cabinet of collective responsibility be held liable and accountable for the catastrophic losses they have inflicted upon the nation they are paid, with an additional dessert dish of expenses, to server, and which they were warned were possible and likely if following the path the Chancellor of the Exchequer took via his "mini-budget" statement to the House of Commons and the nation on Friday 23 September 2022 ..... for no meaningful punitive penalty guarantees much more of the same, and possibly worse, time and time again, and that is surely serial madness both confirmed and compounded, which is a very strange default condition to accept in public positions of national and/or international and/or universal leadership/conservatorship ?
@amanfromMars_1
Quote: "...collective responsibility ... liable and accountable ..."
To which planet does this quote apply? The last ministerial resignation that I can remember is Profumo, and that was 1963!!!
Since then there has been no evidence at all of ANY RESPONSIBILITY AT ALL from members of the UK cabinet!
Fortnums and Mason bags stuffed with cash; PPI contracts for friends; cosy director jobs before (and after) holding cabinet office.....the beat goes on....
Yes, well that was the whole point of the question, AC.
Whenever one can clearly see them for what they really are, what sort of a pussy allows them to lead and feather their nests and ingratiate themselves with like-minded peers and wannabe groupies?
And as for those old oxymorons of Military Intelligence and Secret Intelligence Services supposedly tasked with providing State Security, what valid conclusion is one to make of their active government support and non invasive disruptive engagement?
Pussies’R’They or Us?
However, things have been somewhat fundamentally changed of late and much to the consternation of status quo facilities and utilities. Progress indeed, but certainly not as was expected or diligently planned for to be under extant forces command and control/conventional traditional ways and means and memes of systems administration and attractive reward.
Quotes from some googled instances:
1972 — R Maudling, Brit Home Sec and Deputy Prime Min, resigns July 18 because of his past involvement with bankrupt internatl architecture company.
John Stonehouse, the former Labour minister, who was convicted in 1976 for theft and false pretences.
David Mellor was a Conservative member of the UK Parliament [...] resigned in Sep 1992 following a [sex] scandal in the tabloid press.
[1994] Jonathan Aitken yesterday was stripped of the last vestige of official respectability when he lost his membership of the Privy Council and, with it, the title the Right Honourable.
Since then honourable resignations have become like hen's teeth. Not because of a lack of cause - but a general increase in politicians' hubris and immorality.
Funnily enough, I knew this comment was written by you, amanfromMars 1, without even finishing your first sentence or reading your username beforehand. Didn't even need an AI to do it!
You're gonna have update your writing style if you want to avoid detection in the future!
Thanks for that good news, Tech Cadet, which makes it extremely difficult, and ideally practically impossible, for any virtual smears and suspect questionable ACTivity by a wayward third party being successfully applied and believed attributable to one via any sort of malicious abuse and flagrant misuse of one's identity/moniker/handle/name/being.
Have an upvote for taking the time to share the good news. :-)
There are a number of hoops, trindflo, to be presently jumped through if a foreigner has anything interesting worth investing in to share either first or exclusively with Uncle Sam and allies who be not as plantation slaves, for all are bound to be worrying about security arrangements being adequately robust and failsafe energised.
For example ......
amanfromMars [2209301925] .... agrees on https://www.nationaldefensemagazine.org/articles/2022/9/30/attracting-tech-talent-a-national-security-issue-says-white-house-official[Thank you. Your comment will be displayed soon after reviewing.]
Something very similar to the difficulty discussed here is shared with Stew Magnuson in a reply to his 9/29/2022 report "JUST IN: AUKUS Capacity Shortfall ‘Comes as No Surprise,’ First Sea Lord Says”. Recognising points of failure is one thing, resolving and remedying them is quite another.
Re "“I think the U.S. needs to reform the way in which it thinks about collaboration,” he [Vice Adm. Martin Connell, the Royal Navy’s second sea lord] said."Whenever most, if not all, US Requests for Sensitive Information and Military Proposal Tenders state that they are only open to and available for US citizens, and most, if not all, usually are, does such practically guarantee one severely disadvantaged in a rapidly agile, worldwide encompassing discipline that seeks to provide sensitive and secretive solutions to a rabid global marketplace populated in foreign lands with mirroring requirements chasing similarly leading and/or overwhelming advantage.
Failure to remove that blockage/blockade virtually directs all possible positive creative alien help elsewhere foreign leaving the home teams behind in a future to struggle with the consequences of the disadvantage they have created and find themselves battling against.
Ps ..... That default National Defense Magazine acknowledgment, Thank you. Your comment will be displayed soon after reviewing, is certainly misleading and should read, if the truth needs to be told ....[Thank you, but be aware, your comment may not be displayed for viewing after reviewing]
To imagine though that such a pronouncement and discovery suggests the end of a comment journey rather than its continuation onward with the opening up of any number of tightly closed doors for its content to be further examined and pondered over would be surely naive and certainly most unlikely in magic circles exercising both ancient and postmodern secret intelligence services.
Damn those terrorists and child abusers... If it wasn't for them, we would still have some semblance of privacy!
But what other options are there than to unmask everyone, and be able to track and trace everyone's thoughts at all times?
Thank goodness for our military and police that are able to direct their hefty budgets onto the people they are protecting to uncover the criminal thinking. Still, they aren't having that much success in making arrests are they? I'm starting to think that the "protection" we receive might not be for our own good....
Who are the terrorists and child abusers again?!
I view this as yet another example that the US DOD cannot harbor the idea of defense when offense is an option. This aspect of the project makes it seem like it is not about stopping anonymous trolling, but rather about owning the weapon.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
