So, the US Secret Service has issued a requirement for software that can detect sarcasm in tweets. And lo, there was much rejoicing in the land, especially among post-doc researchers looking for grant money to pursue research in algorithmic applications of semiotics with a side-line in heuristic knowledge processing and associative networks. And Agent Smith scowled furiously, and was perplexed.
Background: the US Secret Service has two main jobs (three, if you include persecuting Role Playing Game companies, but let's leave Steve Jackson out of this for the time being): combating currency counterfeiting, and protecting the President Of The United States, an office that for some reason seems to attract the armed attentions of the deranged, damaged, and just plain homicidal the way my wardrobe attracts moths. Traditionally, the job of protecting the POTUS could be made a good deal simpler by (a) listening for lunatics with guns uttering death threats, and (b) sending a couple of nice fellows in dark suits and dark glasses to have a polite conversation with the aforementioned lunatic and convey the impression that their displeasure would be made extremely clear should words ever be translated into deeds. But then the Internet happened, and it just so happened to coincide with a flowering of highly politicized and canalized news media channels such that at any given time, whoever is POTUS, around 10% of the US population are convinced that they're a baby-eating lizard-alien in a fleshsuit who is plotting to bring about the downfall of civilization, rather than a middle-aged male politician in a business suit.
Well now, here's the thing: automating sarcasm detection is easy. It's so easy they teach it in first year computer science courses; it's an obvious application of AI. (You just get your Turing-test-passing AI that understands all the shared assumptions and social conventions that human-human conversation rely on to identify those statements that explicitly contradict beliefs that the conversationalist implicitly holds. So if I say "it's easy to earn a living as a novelist" and the AI knows that most novelists don't believe this and that I am a member of the set of all novelists, the AI can infer that I am being sarcastic. Or I'm an outlier. Or I'm trying to impress a date. Or I'm secretly plotting to assassinate the POTUS.)
Of course, we in the real world know that shaved apes like us never saw a system we didn't want to game. So in the event that sarcasm detectors ever get a false positive rate of less than 99% (or a false negative rate of less than 1%) I predict that everybody will start deploying sarcasm as a standard conversational gambit on the internet. Trolling the secret service will become a competitive sport, the goal being to not receive a visit from the SS in response to your totally serious threat to kill the resident of 1600 Pennsylvania Avenue. Al Qaida terrrrst training camps will hold tutorials on metonymy, aggressive irony, cynical detachment, and sarcasm as a camouflage tactic for suicide bombers. Post-modernist pranks will draw down the full might of law enforcement by mistake, while actual death threats go encoded as LOLCat macros. Any attempt to algorithmically detect sarcasm will fail because sarcasm is self-referential and the awareness that a sarcasm detector may be in use will change the intent behind the message.
Indeed, a successful sarcasm detector implies not only an eerily functional human consciousness emulation and a metric fuckton of encoded knowledge about human cultural relationships, but the ability to engage in primate social interaction with sufficient agility to tell when a primate means something, and when a primate is signalling an implicit negation of meaning. Which in turn means the sarcasm detector requires a theory of mind. Hello, singularity! And while I'm at it, can I have a pony? And the moon on a stick, too. KTHX.
I give it thirty years and a $10Bn budget, tops. Then POTUS can sleep easy, knowing that the Secret Service are onto those pesky sarcastic twitterers who think it's funny to waste their time by cracking jokes about a very un-jokeworthy subject. (Hey, did you hear the one about the convention for presidential assassins ...? No? Me neither. Okay: how about, how many presidential assassins does it take to change a lightbulb?)
Or they could just ban sarcasm on the internet.
Yes, I really think that could work.
The joyful thing is the number of programmers — bright programmers — whose position on the Aspergers/Autism spectrum makes them rather unobservant of sarcasm in the first place. There's nothing quite like trying to build systems to recognise something one frequently doesn't recognise oneself.
(I did say 'them'. cough Obviously I'm never myself socially obtuse.)
Hmm, solve this problem and you may have alleviated another burden of the awkward geek squad.
Obligatory Piranha Brothers reference:
http://www.youtube.com/watch?v=-fNvi6xG-5Y
"He used... sarcasm. He knew all the tricks, dramatic irony, metaphor, bathos, puns, parody, litotes and... satire. He was vicious."
Some thoughts.
First, even ignoring the population on the Autism spectrum, the Twitter medium is notorious for its briefness hiding the writer's intent. Even with people you know personally with the full context of their timeline, you will sometimes be unsure how serious a statement is. And for out of context retweets of people you never heard of, forget about it.
Also, I can't see this actually being used, due to the paranoia of law enforcement. Either the analysis will be provided as a guideline, in which case agents will probably just ignore it, or if it is used as a filter it will be turned off after the first false positive.
Lastly, I would like to get my submission in early. def sarcasmdetector(tweet): if tweet.find('sarcasm') >= 0: return True else: return False
Charlie said "I predict that everybody will start deploying sarcasm as a standard conversational gambit on the internet."
Didn't this happen about 15 years ago?
How about a different explanation, namely that Secret Service knows all this perfectly well, but needs to cover its ass?
Been there, laughed at that.
Yes, but this time the Americans will finally start to understand how it works and get properly involved.
Right now, just off the top of my head, I can think of two promising approaches to this problem:-
Don't try to identify the sarcasm as such. Use the negative approach for identifying weeds, i.e. a weed is anything you didn't positively want growing. In the same way, anything you didn't take straight is sarcasm.
Don't raise the bridge, lower the water. Don't try to identify sarcasm yourself, train people to flag their own efforts at sarcasm with readily identifiable markers you can use as proxies just as there are readily identifiable markers you can use as proxies to tell when a politician is lying. People who don't use the markers the way they are being trained to use them will, ipso facto, be making statements which are highly significant at a meta level - and so, not sarcasm, and so, they will be people who are worth editing out of the discourse in a way that provides reinforcement for the training of the remainder.
I will leave it as an exercise for the reader to determine which, if either, of those approaches uses sarcasm, and if so where and how.
You've seen the level of sarcasm the typical right-wing nutter uses, Charlie. It's probably one of the easier natural language processing problems left on the table. I wouldn't even call it artificial intelligence, since the TRWN can be simulated by a bot on a Apple ][ with a program written in Integer Basic (e.g., Jordan).
What makes you think TRWNs are the real target?
I think the real target is the autonomous exercise of professional judgement.
I've been making anonymous online death threats against the president of the USA for years now. I've yet to receive a single visit from the SS, or even a nastygram. I'm really disappointed actually. Maybe I should switch to non-anonymous threats.
I'm going to kill the president of the USA in three weeks time, at 3 o'clock, on a Thursday.
There. Let's see who comes knocking.
Maybe I should switch to currency counterfeiting instead (except that's likely to attract the attention of whichever government, so...).
Ahem.
Just how many people who've had a pop at PotUS have announced it, nice and obvious like, in an easily monitorable text form, well before they went to it?
Maybe a smarter move would be to put the money into mental health services, such that el nutter, whoever he's targeting (and I'd suggest it's relatively easy to replace a figurehead) gets the help they need BEFORE they hurt anyone.
Oh, and how about making sure said nutter can't get his hands on ANY guns?
Just a thought, from sanityland.
Wait, I think I just outed myself as one of the set {deranged,damaged,just plain homicidal}. I didn't mean to do that.
Do serious loons actually need detecting that way? Long rants about ideology and revenge seem much more the norm and easier to detect. As we just saw, the problem is getting a competent investigation done on loons raving in plain sight.
If only Dave and Frank had used sarcasm, the outcome of the Jupiter mission might have been so very different!
Qaida terrrrst training camps will hold tutorials on metonymy, aggressive irony, cynical detachment, and sarcasm as a camouflage...
Now wondering how sarcasm works in other languages, and whether it translates to and from them. Will translating a message from Arabic or Pashto still have a sarcastic character to it, or will it be lost in translation? And will the software have to be able to translate colloquial forms of the languages?
I think we all know that sarcasm isn't always obvious on the internet, even without translating.
If only Dave and Frank had used sarcasm, the outcome of the Jupiter mission might have been so very different!
"HAL, umm, do you think you could open the Pod Bay doors?"
"Dave, I don't think. I know."
"Yeah, I don't think you know either."
As I get older, I often get the feeling that it is more and more obvious that Douglas Adams wasn't so much making things up as predicting the future state of the world.
Numbers, Charlie, and past tendency.
I think the real target is the autonomous exercise of professional judgement.
Isn't that always the real target of a bureaucratic activity? Bureaucracies exist to control things, and interpret flexibility as poor control.
Health care is one o' them ENTITLEMENTS that libruls keep forcing down the throats of the Amurrican people. Cain't have that. It would raise taxes.
And every man in the U.S. of A. -- even women -- have the right to bear arms. God wrote it in the Constitution. The day a man can't carry a loaded semi-automatic into a Chipotle or Chili's or Sonic is the day the Communists won. Or the terrorists. Or gay marriage. Whoever we're supposed to hate now.
Thoughts from crazyland, a.k.a. the Republic of Texas in the United States of America (for now)
I'm not sure it's entirely stupid. People clearly DO post credible threats on social media, and they DO get drowned out by bogus ones.
If software can help filter out the silly ones even a little bit I can't see how that's a bad thing. Ok not sarcasm. But for example filtering out retweets from someone else that has already been looked at.
it seems to me to be a case where even a very very bad solution is still slightly helpful.
I'm sure Elon Musk could put a pony on a stick on the moon for less that this would cost.
I would think a people-oriented solution would be better. A tip line where people can report plausible threats would do probably do better than any algorithm, because of the humans in the loop.
How about this one: "Umm, NSA. We know the bad guys use the Internet to pass messages. We want you to find them, monitor them and generally mess them up. BUT, you cannot monitor innocent people, breach their privacy, or gather metadata about overall Internet traffic. Please have that done by Monday, OK?"
Signed, Bill Lumbergh Director, NSA
Slight clarification, while I suppose this has applications for the USSS's presidential protection mission, this likely is more related to the Electronic Crimes Task Forces (ECTF), which were added to the USSS's criminal investigation mandate (grown from its original counterfeiting mandate) by the USA-PATRIOT Act.
But then you've got to figure out which are plausible tips.
People-oriented solution? That's been shown to work quite effectively so far. Actual human beings always detect sarcasm, and respond appropriately.
http://www.popehat.com/2013/07/11/the-first-amendment-protects-satire-and-rhetoric-lol-jk/
I'm sure PETA would have something trenchant to say if he did so, though ...
I'm sure Google would be interested in building the technology into Glass.
On a more serious note, you probably could do something by parsing people's social graph, because the same technique has been used with reasonable success against phone data.
Rather than looking for connections (which you might do to identify a criminal gang or terrorist cell) I wonder if you could look for absences?
i.e. your presidential assasins, and right wing bomb makers all tend to be 'loners' in a particular sense. No one ever says 'He was a really grounded sociable guy'.
If I was writing a near future crime/spy/thriller book, I think a pretty interesting area would be how social media will start to interact with undercover work on both sides - does it make creating a fake identity easier, or harder?
Are there warehouses of people working on fake identities now, to be deployed 5 years down the line, hackers retroactively creating abandoned Geocities pages.
Our Gracious Host noted:
"[W]hoever is POTUS, around 10% of the US population are convinced that they're a baby-eating lizard-alien in a fleshsuit who is plotting to bring about the downfall of civilization, rather than a middle-aged male politician in a business suit."
That would have been an improvement from noon UCT-5 20 Jan 2001 through noon UCT-5 20 Jan 2009. Or, for that matter, the corresponding span twenty years earlier; in both timespans, the lizard-alien would have done less damage to civilization.
I set forth the above comment as a test for any automated sarcasm detection system, parallel to the perpetual liar who claims to be lying...
I went to a presentation from a sentiment analysis business last year. They were discussing how they were using it on the firehose feed of everything from twitter. Rather than assassinations, they were more interested in marketing, branding, and the impact on stock prices (there are apparently some iiiiiiinteresting correlations).
The thing that entertained me was they said they had to special-case content coming from the UK — because our levels of sarcasm/irony were skewing their results.
I think it was the most patriotic moment of my life <sniff> ;-)
Oh - and for those who are interested there's quite a bit of active research on detecting sarcasm in the sentiment analysis field. for example:
(by no means a complete list)
Not my field — but there are plenty of folk already poking at this in the commercial sector. I agree that good-as-human is gonna need strong AI. Good-enough-to-be-useful on the other hand, not so sure.
"...whoever is POTUS, around 10% of the US population are convinced that they're a baby-eating lizard-alien in a fleshsuit who is plotting to bring about the downfall of civilization, rather than a middle-aged male politician in a business suit."
There's a difference?
I can see how my comment might be seen as ironic, sarcastic, or funny, but research into the changes of brain structure shows that being in power turns you bad; as in the decisions you make affects the way you think about future decisions.
Proving the old adage that power corrupts etc.
Lounge suited lizards are funny though, right? ;-)
Amusing to see this the day I publish this article on how big data-based neural networks are showing intriguing linguistic transfer learning effects: teach one English, it learns Chinese better, and French even better.
It turns out that there are deep grammatical elements that are being exposed by the neural nets, where (for example) the vector between "man" and "woman" is the same as "king" and "queen", or "boy" and "girl". Seems Chomsky was right after all, so there is scope for some form of sarcasm detector in the state of the art.
That's abstracting from a large body of people, though, something you can in principle approach through statistics.
Finding an individual crazed assassin, much harder; they're an individual, we're not looking for a present trend, we're looking for an event in the future.
It's one of those things where I suspect you don't so much need Strong AI to do it as you need Strong AI to figure out how to do it. (And then you need another Strong AI with a different people-wrangling specialization to convince the legal system that, no, really, it matters what shade of red clown noses are.)
Hm, the theory behind this seems to be that people using sarcasm pose no threat to the US government. Probably true.
This seems like another massive effort to prove that black swans (aka lone wolf terrorists) are predictable. I guess that means that the NSA's favorite excuse for its behavior is wearing a bit thin, and they want new cover?
I, for one, look forward to it. Once we've solved the very narrow and tractable problem of figuring out which random person (and I do mean random, even within the universe of paranoid schizophrenic delusionary gun owners) actually decides to shoot at someone rather than chilling or seeking help, we can get to the rather more expensive and equally random problem of stopping accidental fires and semi-accidental arson. That should be almost as easy.
Yeah. Right.
Your fears about Al Qaeda irony camps are completely unfounded. The actual Statement of Work ("RFP-FBOSocialMediaToolsFBOcombinedSyn-_-Sol") lists the actual requirement as:
"Ability to detect sarcasm and false positives"
so not only will the sarcasm be detected, but the false positives will be, too.
Then again, it also has "Compatibility with Internet Explorer 8" so I'm not sure the whole thing isn't just an elaborate troll. Nobody still uses a 5-year-old web browser, do they?
Well, IE 8 is the minimum supported IE on extended-support government XP systems.
I especially liked: •Ability to create custom reports without involving IT specialists; and •Ability to search online content in multiple languages.
They want natural language processing too, and ID-10-T proof reporting.
it seems to me to be a case where even a very very bad solution is still slightly helpful.
Yes. I suspect they are doing this to see what it would cost to eliminate 90% or 99% of the fluff they see. As the price of having real humans read all of these AND make a rational decision on what to do with each one is likely headed to the GDP of the nation.
A government contract I was involved in a couple years ago wanted compatibility with IE six, though I think they eventually agreed to relax that.
What? No reference to Poe's Law?
I especially liked: •Ability to create custom reports without involving IT specialists; and
Apparently you've never seen some of the "user doable" crap that government contracts wind up creating.
Or you could just take the money it would cost to visit Disney for a few days and fly up to somewhere in the middle of the country / northern half where you might not feel like you are walking inside of a monster pressure cooker. Just check out the weather reports a day or two earlier and pick a city.
Note I grew up near where the Mississippi and Ohio and 2 other rivers merged and I don't mind this kind of weather. Well not as much as most do. But given where Charlie lives he might feel he's in a tub of hot water with his diving mask missing.
[/sarcasm off] [/mostly]
Duuuuh.
Totally wrong post. Sorry. Moderators move it if you want. :(
Until this wondrous feature can be properly implemented, I would like to propose a simpler solution. What the government needs is a special tip line to which anyone may anonymously submit texts that are suspected of containing sarcasm.
Now, wouldn't it be so much easier if we used the Ge'ez script (as implemented by Amharic, the main language of ethiopia, and Tigrinya, the main language of Eritrea) and had an actual punctuation mark - ¡ - which goes at the end of a sentence to mark sarcastic content?
Incidentally, Edinburgh is spelt "ኤዲንቦሮ" in Amharic.
They don't have to be professional spies or trained linguists. Why not use Mechanical Turk or its equivalent to determine possible sarcasm in suspect statements? Two cents per test for the best natural-language processing systems available using native-language speakers only depending on the region of interest? Got to be cheaper, and with enough oversight and cross-checking a lot more accurate out-of-the-box than training up a neural-network or populating a database.
applause
(Wait, do I mean that, or am I being sarcastic? I am so confused.)
Sigh ... Another academic field militarized. Now we're going to have to start training combat rhetoricians.
Is this an attempt by the Americans to retrospectively arrest Charlie Brooker via his wikipedia page?
"I ended a Screen Burn column by recycling a very old tasteless joke ("John Wilkes Booth, Lee Harvey Oswald, John Hinckley, Jr. – where are you now that we need you?"), and within minutes half the internet seemed convinced The Guardian was officially calling for George W Bush's assassination. My inbox overflowed with blood-curdling death threats, and it was all very unfunny indeed – a bit like recounting a rude joke at a dinner party, only to be told you hadn't recounted a joke at all, but molested the host's children, and suddenly everyone was punching you and you weren't going to get any pudding. I've had better weekends."
When I think about it, this kind of makes sense.
Right now they aren't supposed to leave any stone unturned. They get more internet garbage than they can handle, and every one they ignore is officially dereliction of duty for them.
But if they get software that claims it can tell which ones not to bother with, then they're off the hook. Say for example the software says to discount every random internet threat. Then they can ignore all that and concentrate on whatever they think their real job is.
And if somebody who made random threats on the internet then actually does attack the President? They can blame whoever sold them the software. They can then buy new software from somebody else which has fixed the flaws in the original. Problem solved.
They don't need software which does what they say they want. They only need software that lets them ignore stuff they want to ignore.
Hm, the theory behind this seems to be that people using sarcasm pose no threat to the US government. Probably true.
Obey the government!
Hug the president!
Happiness to the military industrial complex!
paws4thot wrote: 'Charlie said "I predict that everybody will start deploying sarcasm as a standard conversational gambit on the internet."
Didn't this happen about 15 years ago?'
35 plus years ago. The early email and earliest 1979 Usenet clearly contained satire, biting wit, and the beginnings of modern Snark.
Postel and his ilk have much to answer for.
I enjoy the comment section here because it can usually be read without a weapons-grade internationalized snark detector. :-)
Automated sarcasm detection 'only' has to perform roughly as well as ordinary trained people to be valuable.
Related, for OGH, first time I've seen "two wetsuits" in casual writing, spotted in a snark-(very-)heavy blog's comments: "Two wetsuits, man. This guy has ammosexual-wetsuit-abuser written all over him."
A gate stuck on 1 is about 99% accurate detecting sarcasm and snark on the intertubes...
Surely it would be simpler for the USG to ban sarcasm from the internet?
Actually, there are two other Pirhana Brothers clips, giving the full story, rather than just "Doug" ... For full, erm, "enjoyment" PART ONE and PERT TWO Ah, it all comes flooding back. For US & other furriners, the Pirhana Bros were loosley based on the notorious Kray gang .....
That'll be Epimenides' Cretan friend who said "So all Cretans are sarcastic? Yeah, right!"?
THAT only make sense to US residents. I had to look it up & still wasn't much wiser ....
Interesting. I was recently told that "Obamacare" has had a very interesting (beneficial) side-effect. The US job/labour market has freed uo, enormously. Why? Because, before people didn't dare try for a new job, because they didn't get healthcare for n years, if at all, aftre moving. Now, everyone has to have it & the problem has evaporated ...... Anyone else know any more about this, closer to the source?
combat rhetoricians As in "Zen & the Art of Motorcycle Maintenance", you mean?
I'm not sure it directly answers your question, but in a fandom of which I'm a member, probably a lapsed member now to be honest, one of the more prolific and beloved authors of fairly adult fanfic (not necessarily pornographic, but more realistically treating characters in their 30's like they were in their 30's at least) turned out to be an undercover cop.
The fandom shrugged and said "but we like your writing" pretty much (although there were, inevitably a wide number of other reactions, but that was the biggest, by far). Her bosses were far less accepting although they didn't officially go beyond 'risking her undercover work' as it was reported. (There's a lot of suspicion the fact she was writing lesbian fanfic didn't help her case but there's no proof.)
That said, from the other side, I bet the agencies that create fake IDs have got good at creating fake social media accounts too, and probably have a little army of people behind the scenes that maintain cover IDs when they're not in active use. It takes longer, you don't just rustle up a driver's license and a passport and whatever else any more, not if you want it to last more than an hour, but you can probably have someone tweeting and posting to their wall with faked geolocation from wherever maintaining 50-100 accounts as their full time job once they're set up.
And to steer it back towards the main topic - if they ever get an algorithm to pass things as "OK because it's sarcasm not real" I wonder how long before it's reverse engineered and the actual assassins that communicate this way are running their communications through the function addSarcasm(string msg) to foil the detector. I'd give it a month tops before there's one that's >90% effective.
A potential solution from an early episode of The Big Bang Theory.
Leonard: For God's sake, Sheldon, do I have to hold up a sarcasm sign every time I open my mouth?
Sheldon (intrigued): You have a sarcasm sign?
Class Wood? Class Trees? Class Forest? Class Plantation? Class Arboretum? Class Lone Pine? Claas Orchard?
Please distiguish. without overlapping .... Perhaps not.
Combat rhetoricians can only change the past. We'll need combat Incantors to change the future. (with a nod to Anathem).
Hey, there's already a perfectly fine soution out there! Everything inside <sarcasm> </sarcasm> tags is sarcasm, everything out it is not! Or, for full posts, if there's a tounge-in-cheek smiley at the end of the post :-p
... and your first paragraph already disproofs itself!
The Congressional Budget Office released a report claiming that the ACA would reduce the number of hours worked by the equivalent of 2.5 million full-time jobs, due to full-timers moving to part-time or self-employment now that their health insurance isn't dependent on having a full-time job (known as "job lock").
Republicans immediately proclaimed Obamacare was a job-killer. "We'll lose 2.5 million jobs!" They probably voted to repeal the ACA three more times in the same day.
Then again, their (and our) corporate masters want job-lock. They want a captive work-force that will work insane hours for just enough pay, because if the white-collar workers get uppity they can fire them and hire new ones just like them. It's also why H1-B visas are so popular: why pay an American citizen a living wage when you can get Indians/Chinese/whatever to work for less and keep them dependent on you for their visa? It's like coal miners and assembly line workers a century ago, except white collar workers are too timid to unionize.
Veering dangerously close to the topic, though: I wonder how much the sudden need for automatic sarcasm detectors is driven by the volume of data the TLAs have to sift through, and how much is a desire to get rid of annoying humans with their health benefits and consciences and need to sleep.
Granted. I was assuming a certain power of syndication, plus talking about this puts me in the mind of the era when I was participating, which was the late 1980s/early 1990s.
Hm, the theory behind this seems to be that people using sarcasm pose no threat to the US government.
I think you're confusing the gubbirmint with the POTUS.
The gubbirmint consists of:- 1) The executive, which for the Yousay includes the POTUS and the vice-POTUS (and various secretaries of state, the US attorny general...). 2) The legislature, aka Congress in the Yousay. 3) The judiciary.
The "lone gunman" (aja JFK (ref to Red Dwarf, episode Tikka to Ride)) constitutes an actual threat to the POTUS, since he has struck at least 6 times, and been successful 4 of them. He's unlikely to constitute a threat to the gubbirmint as a whole, if only because he'll probably not can carry the firearms and the requesite several thousand rounds of ammunition.
A "bomb plot" could conceivably "take out" the legislature (and possibly head of state; see under "Gunpowder Polt") using conventional explosives. Given the skills to assemble something like the Tzar Bomba, it might manage the executive, legislature and most of the judiciary, but looking for people acquiring large quantities of weapons grade actinides seems like a more promising approach to detecting this attack than looking for people using (or maybe not using) weaponised levels of irony, snark and sarcasm!
I understood the original comment just fine thank you kindly. Being a regular viewer of CSI and NCIS probably helped there.
A simpler strategy would be to identify individuals who've lost their sense of humor. Simply hire Google to flash random/various jokes on screen with every search, or intermittently*, and if the keyboard detector doesn't register a physical reaction such as guffaw, twitter, etc. - send in the SS. Interesting: stand-up comedians pushing the boundaries of national security/intelligence technology.
Hey, I've got it! Monty Python's 'Hungarian Translation Book' mixed in with 'Ministry of Silly Walks'.
"You're in a desert, walking along in the sand, when all of a sudden you look down..."
Oh wait, wrong scenario
El - I think faking up a Twitter identity would be relatively easy, but a Facebook identity is more interesting.
You wouldn't just be faking the primary identity, and maybe a family, but a whole history (childhood friends, extended family, ex-work colleages you never unfriended after leaving) that is suddenly a lot easier to check up.
We know how easy it is for people to get themselves into a social network - target mutual friends who appear to accept any request, then convince someone you have many mutual friend - which seems to be a large part of how undercover has always worked.
But it's now way easier to cross-check those claims.
"Sarcasm? What's sarcasm?"
"You know what a joke is? It's the same thing."
"Joke? You want to see a joke?"
BLAM BLAM
"And that's a punch line."
A simpler strategy would be to identify individuals who've lost their sense of humor. Simply hire Google to flash random/various jokes on screen with every search, or intermittently*,
Now, THAT would cause me to loose my sense of humor in no time.
A simpler strategy would be to identify individuals who've lost their sense of humor. Simply hire Google to flash random/various jokes on screen with every search, or intermittently
Well now, you could easily prove that I've lost my SoH that way. Simply flash up a succession of Bernard Manning, John Bishop and Peter Kay material!
The point I wanted to make is who the Secret Service deems most dangerous:
My personal view is that people who don't use sarcasm are the most dangerous, since they don't have a sense of humor and are probably literal-minded. Even if they say good things about the government today, they might get disappointed tomorrow.
Yeah, and my point was that you need a very large seed base of material because I deliberately picked 3 "comedians" whom most of the the readership I'm pretty sure will find at least one of somewhat less funny than root canal surgery without anaesthetic.
You do wonder if around the time the Arpanet tenders where being let if their where letters in Astounding - pooh pohing the idea on a self healing network :-)
"Agencies which create fake IDs" ... that'd be the endangered ones this decade, right?
Passports have mandatory biometric identifiers these days. It's probably only a short matter of time before the biometrics include a DNA fingerprint, too -- not just fingerprints or facial bone structure. And many countries are now retaining the machine-readable passport details automatically whenever anyone crosses a border. This poses a huge headache to intel agencies because a human agent who enters a country using one set of ID can't then re-enter the same country using different ID with the same biometrics -- they become trivially identifiable as a ringer. Indeed, it has been suggested that this process may be on the edge of destroying the utility of human spies who aren't long-duration resident agents; your jet-setting agent gets just one shot at each unfriendly border before they're on the record and can't go back under another name.
Wasn't one of the few interesting things to come out of the tragedy in Malaysia that few airports actually check all the security features on '90s vintage passports, let alone experimental ones like fingerprint scanners? If there are plenty of people flying between countries with passports which were reported stolen in another country a year before, I doubt that people with skills and resources are finding it impossible to get around. The current fad for security theatre won't last forever.
I've been to Malaysia. They're very proud of their monorail and skyscrapers and shopping malls and electronics factories, but your reference point for how their government works should be a Pacrim, majority muslim version of Mexico.
In contrast, try entering the UK or USA without having your fingerprints scanned (at least with a new passport) or at least a stern look from a border agency person and having your passport details electronically captured.
This isn't about "security" against terrorism, it's about reducing the free movement of labour. (See also Lenin's theory of imperialism. Yes, I think we can agree that Lenin was totally crap as an administrator, but as a theoretician he was a pretty sharp mind -- in the context of his times.)
Oh, and how about making sure said nutter can't get his hands on ANY guns?
Do remember that throughout history the most common assassin of the head of a government was a high official in that same government. That may not match past US history, but we aren't the same country we were a couple of decades ago, so the past may not be a reliable guide to the future. As popular dissatisfaction with the government rises, a coup d'etat becomes more likely.
@20:
And every man in the U.S. of A. -- even women -- have the right to bear arms.Damn straight! You never known when you might run into evil alien pod people. Or Canadians. They're pretty much interchangeable.
Please don't make invidious comparisons, Charlie. As we all know, como México no hay dos.
This isn't about "security" against terrorism, it's about reducing the free movement of labour.
The TSA is not, generally, deliberately and consciously evil. They're just a bureaucracy. They've been tasked with a nigh impossible job, and they're under pressure to take actions that could semiplausibly be called doing the job. Adding some biometrics to existing checkpoints meets that criterion in a reasonably easy and cheap manner, so they do it. Securing the border doesn't, so they don't.
Passports have mandatory biometric identifiers these days. It's probably only a short matter of time before the biometrics include a DNA fingerprint, too -- not just fingerprints or facial bone structure. And many countries are now retaining the machine-readable passport details automatically whenever anyone crosses a border. This poses a huge headache to intel agencies because a human agent who enters a country using one set of ID can't then re-enter the same country using different ID with the same biometrics -- they become trivially identifiable as a ringer. Indeed, it has been suggested that this process may be on the edge of destroying the utility of human spies who aren't long-duration resident agents; your jet-setting agent gets just one shot at each unfriendly border before they're on the record and can't go back under another name.
Perhaps this is another motivation for the Anglosphere governments cooperating to ensure that no foreign computer system is ever safe from remote penetration. The collateral damage is that domestic computer systems must remain insecure too, because improved domestic security would spread internationally, and national security types prioritize offense over defense. As I wrote elsewhere recently, both the US Department of Defense and the National Security Agency seem to have permanently confused "Keeping Foreigners in Danger" with "Keeping America Safe."
The US government has evolved elaborate ritual displays of outrage about Chinese hackers stealing secrets from American computers. The actual behavior of law enforcement and security agencies shows that they prefer pervasive information insecurity over pervasive information security. Better that ten domestic systems remain vulnerable to remote exploitation than one foreign system remain secure.
Sounds like you and Josh Green Allen (@fireland) could collaborate:
https://web.archive.org/web/20081012085146/http://wiretapfollies.com/
Didn't the same article state that most rich countries were equally lax about eg. checking passports against the Interpol database of stolen ones? I don't recall anything more than my passport being checked when I entered the US or UK, although the Brits may have done a fingerprint scan. The passport could contain 10 GB of biometrics, but that doesn't matter if nobody checks them. If I recall, fingerprint readers aren't as reliable as people thought they were before computer analysis of large data sets, and every metric which they actually check costs time and annoyed travellers.
Google is turning up a number of news articles, the detailed ones often quoting a Clive Williams of Macquarie University.
"You're in a desert, walking along in the sand, when all of a sudden you look down..."
A piano falls on your head. You're dead.
I'm not sure it is an invidious comparison, Noel. Per-capita PPP-adjusted GDP is in much the same ball-park. Gini coefficient isn't too far out of whack -- both skew high. Both Mexico and Malaysia have rich centers and enormous concentrations of wealth but relatively deprived rural backwaters, issues with endemic political corruption, and a very rich/powerful neighbour who skews the economy. (Admittedly Singapore isn't the United States, but as a proxy for "global trade superpower on the doorstep" it doesn't do badly.)
I think this is doable.
The key point is this. You are all looking at this as an AI problem. This is not an AI problem. Identification of general sarcasm from a single tweet is impossible, but what the Secret Service would genuinely want is classification of threats from an user's tweet record and the people he associates with. Some general thoughts that would go towards a solution:
Keyword analysis of the person's tweet record would get you a lot of the way. You can infer the person's political alignment, degree of preparation for a threat, physical location and so on. If someone makes a bomb threat against a politician he is opposed to, and talks about fertiliser suppliers, and is living where the guy is due to speak, then that makes you more worried than someone whose next tweets are about the next episode of My Little Pony.
Don't detect sarcasm, detect reactions to sarcasm. If it's a joke, the person's twitter followers would likely react to it as a joke. 'Haha', retweets, and so on. Serious threats are not likely retweeted widely, and if they are replied to with statements/users that are also redflagged as per #1, you'd really get worried.
Use the broader tweet record. Tweets that have been said by millions of people should be given a reduced threat. Only way to be realistic in terms of investigative effort.
There aren't actually that many effective ways to murder a politician. That means looking for 'realistic' threats is fairly trivial through application of a whitelist.
Don't give a yes/no answer. I think an actual useful tool for an analyst would be something that doesn't say 'this is sarcastic', but gives an useful user interface that gives relevant past and previous tweet excerpts, and a button to pass this on for deeper analysis. A time saving tool.
Twit-twoo! That’s about all I have to say about tweeting.
However, I am concerned about all our internet data being run through some kind of Voigt-Kampff empathy testing (PKD) software. Bad enough that we’re getting full body x-ray scans before boarding planes … next we’ll be sticking our eyeball into a retina scanner and asked to, “Describe in single words, only the good things that come into your mind about your mother.” –BLADE RUNNER
Actually, that would still be some kind of AI, just not what most people think of, e.g. able to pass a Turing test. Come to think about it, I guess quite a few people would fail at a Turing test, but whatever...
Thing is, if you have a big enough database of a guy/gal's tweets, you could create a profile, and I'm not that sure sarcasm would show with word usage, syntax etc. I'm also not that sure one could easily fake said signature, just like shaved apes usually fail at creating random numbers.
Yes, I'm still dreaming about that psychosis detector, why you're asking?
As for all the talk about autism spectrum and sarcasm (DSM screwed Asperger's) I'm not that sure it's a clearcut negative correlation, especially if you look at detecting and generating sarcastic content.
To use a not officially diagnosed example, both my dad and me are notorious for often missing sarcasm; OTOH, we are also notorious for our cynicism, which seems to be somewhat related to sarcasm. Don't ask my mother about it... ;)
You are exagerating; we are not x-rayed at airports, we get microwaves, exactly the other side of the optical spectrum; you see, x-rays are not that practical for finding things ON humans, not IN humans.
Hi, Charlie.
I was testing the sarcasm detector. Are you sure you're not a computer program?
Apologies! I couldn't resist. The comparison isn't invidious and few Mexicans would take it as such.
But my response, I think, make the point about DOD's fever dream. Unless you knew that "como México no hay dos" has become an ironic catchphrase in that country, the sarcasm is impossible to catch.
You're not quite right. Yes, in the UK and EU the body scanners used are millimeter wave scanners that run on far infrared/near microwave wavelengths. But the USA rolled out some highly dubious backscatter X-ray machines at much the same time. These are banned in the EU due to safety concerns (or rather: due to lack of confirmation of their safety), and due to privacy concerns the DHS began redeploying them from major hubs to less-traveled airports and replacing them with millimeter wave scanners a year or so ago.
Unless you knew that "como México no hay dos" has become an ironic catchphrase in that country, the sarcasm is impossible to catch.
Alas, I didn't get it because I don't speak Latin! (Not even Latin American. Let alone something exotic and foreign like Spanish.)
Damn, there was I trying to sound sarcastic, and then reality has to ruin the joke...
(originally, I tried to emulate the US military press release denying the use of napalm, since the mixture in the Mk77 employed kerosene, not gasoline
http://en.wikipedia.org/wiki/Mark_77_bomb
plus the idea that x-ray were just not that good at doing a strip search, backscatter x-ray is somewhat new to me, thnx)
What you're describing is pretty close to verbatim coding software which has been used by marketing researchers for at least 10 years. There is also software/ algorithms used to analyze speeches by politicians, etc. to assess how emotionally loaded/skewed the speeches were. These two examples use very different rules even though both strive to get at meaning and intent.
One thing that both approaches can and do miss is idiom - this is a problem because idiom can change quite rapidly. Adolescent through early 20's age groups are noted for being idiom sources, and why idioms are often reliable 'age markers', and of socioeconomic status too in some countries. Aggregating the words themselves is easily done; understanding what these specific words mean at that particular moment in time (context) is another matter. Same goes for trying to analyze 'foreign' languages. (Key-word analysis would have to also recognize idiom, detect whether it's the user's native language, etc.)
As for analyzing tweets and other types of social media; this too is already being done by many marketing research/advertising agencies.
Example -- A couple of years ago, I was at a social media presentation (topic was tweets, in fact) where the presenter mentioned being overwhelmed by how much the English (vs., the French or Germans) love their tea! It hadn't dawned on this presenter to consider that the English tweets '(not) my cuppa tea' actually meant (disapproval) approval of some other subject.
OGH: "Agencies which create fake IDs" ... that'd be the endangered ones this decade, right?
Passports have mandatory biometric identifiers these days. It's probably only a short matter of time before the biometrics include a DNA fingerprint, too -- not just fingerprints or facial bone structure. And many countries are now retaining the machine-readable passport details automatically whenever anyone crosses a border. This poses a huge headache to intel agencies because a human agent who enters a country using one set of ID can't then re-enter the same country using different ID with the same biometrics -- they become trivially identifiable as a ringer. Indeed, it has been suggested that this process may be on the edge of destroying the utility of human spies who aren't long-duration resident agents; your jet-setting agent gets just one shot at each unfriendly border before they're on the record and can't go back under another name.
Yeah, not really.
Biometrics are not a precise, mathematical science. Human bodies when measured multiple times give different values, and they evolve over time. Real biometrics give you a statistical assurance that it's unlikely that the person in front of you doesn't match the profile, not assurance that they're a 1:1 match with some sort of encoded profile.
The profiles therefore have error bars.
What that means is that there are reasonably large sets of possible encoded profiles which can match the same physical human being. Which makes this a math game for spies; how many permutations on profiles can we create which match Spy X's fingerprints, DNA, facial structure, voiceprint, irisprint, etc.
And, how much variation in the encoded profile will a nations' passport entry database accept as "different" before it flags a "hey, this profile might match the guy from six months ago, but isn't identical...".
Trivial profile-non-matches, simply diddling the last few bits of the fingerprint and facial data, are already entirely practical. And currently, virtually nobody is watching the near-matches in the databases, so tolerance for how many bits you have to diddle before nobody picks it up is pretty large.
When we reach the point that we're DNA sequencing people in real time, at least for looking for key markers, that's a different thing.
You ordered their jobs wrong. Their first job was to prevent counterfeiting. The second was protection of foreign dignitaries. It wasn't until much later that they were the official bodyguards of the President. Up until President McKinley was assassinated, nobody in particular was responsible for protecting POTUS; since the Secret Service already had experience in "bodyguard" type duties, they won responsibility for protecting the President.
But I totally agree, without any sarcasm whatsoever, that this is an important thing for the SS to be pursuing, and I know absolutely that they will be very successful in their automatic sarcasm detector. I mean, this is only slightly harder than stopping an assassin who is willing to trade their life for their target's. Of course they'll find this task something any 1st year undergraduate student could do. Pff, it's so easy I could write the code for it in the margin, if I could figure out the HTML tag for a margin note.
All this talk about satire is all well and good, but the real money in conversational resources lies in irony. For example, half the conversations in China are based on irony mined from Western Australia.
When we reach the point that we're DNA sequencing people in real time, at least for looking for key markers, that's a different thing.
We're probably a lot closer to that than most people think: the increased speed of genome sequencing turns out to be amenable to Moore's Law, and unless we hit the wall within the next five years, we'll be seeing complete genome sequences for everyone in the developed world performed as part of routine healthcare screening by the end of this decade. The bottleneck rightnow AIUI is matching up the fragments to create the actual map, and that, too, is amenable to massively parallel search hardware -- in fact, it's a classic cloud computing task. Finally, a human genome isn't that big, and the cost of adding the FLASH storage to a next generation passport sufficient to hold it is already sub-$10 and dropping.
Your picture for a border crossing in 2024 is: present your passport. About 4Gb of sequence data is dumped out and retained when you enter your country, and you give a machine a cheek swab or skin scraping. (Under the eyes of the border guards.) You then enter. While you're doing your business in the country you just entered, they run a few quick sequences from your sample and see if they match (a) the genome map in your passport, and (b) any other retained passport genome maps. If it looks oddly good, your sample is singled out for a full sequence run (taking less than a day, at a cost of less than $100). Then they go to town.
If it turns out you're the identical twin of three other guys, then you might have some explaining to do when you try to leave ...
I don't think there's enough margin for your proof here....
If it turns out you're the identical twin of three other guys, then you might have some explaining to do when you try to leave ...
My first wife's hobby was breaking into US nuclear bases in the UK.
She was one of triplets.
And though she and her sisters always insisted they were not identical, when they actually had genetic matching done, it turned out they were identical after all, which would rather screw my ex sisters in law when it came to matching.
(I think there was a recent case where a pair of identical twins were both sentenced when the (DNA and other) evidence implicated one of them, and neither would confess to allow the other free.)
AIUI, for ID purposes full genome sequencing is far more than is necessary. You just look for those key markers, the genes that tend to differ most rather than the ones that we tend to share with the rest of the animal kingdom let alone the rest of humanity. For that, you use prepared receptors that match/don't match each marker gene. While the advances in routine full genome sequencing will surely help, I suspect we already have the technology to do DNA matching fast enough for border control, if there was sufficient will.
Can anybody point to any evidence that the various TLAs presently conduct post hoc matching and analysis on the biometric data captured at border crossings?
Absent such evidence, the default assumption must surely be that the value of said data capture lies not in the data but rather in that the process provides work for bureaucrats and money for politically connected vendors.
If it turns out you're the identical twin of three other guys, then you might have some explaining to do when you try to leave ...
Or if you turn out to be a Human Chimera, with different genomes in various part of your body.
There's still a bottleneck getting the DNA out of the cells and amplified to find the markers.
There's obviously an easier way to do it, since bloodhounds can distinguish between identical twins without using genome analysis.
Anyway, there's also an easy way to spoof genome analysis: you insert the DNA fingerprint data for the false identity into a bunch of hacked oral bacteria, and make sure your spy has them in his breakfast yogurt/mouthwash and skin lotion before he walks through the checkpoint. If you've done it right, the bacterial DNA false positives will overwhelm the spy's DNA from the skin cells, and the spy will sail security through with a false positive DNA match.
This technique doesn't work great for full genome reads, the anomalous marker sequences stand out like a sore thumb. But for a basic 13ish markers scan (by far the bulk of today's scans) it works great.
I think what the idea is, is that while they don't check all the time they check sometimes proofing your agent against any of the various checks is possible but as you know there are quite a large number of checks that might be performed you have to proof against all of them. For the TLAs it imposes higher costs to force a hostile agent to have to take precautions against a wide range of tests. The TLAs work not so much by catching the bad guys but by increasing their costs and inconvenience. It is probably going to catch more drug mules than terrorists, if only because the smugglers make repeated trips so have more opportunity to have a bit of bad luck.
I am reminded that the basic fingerprint system uses an index to find a small group of possible matches which can be then examined in detail. This was all before computers, and a computer-based scan might allow a different level of indexing, but it's the detailed match that costs significant money.
But the costs are dropping. And an intelligence agency might not bother with the sort of detail, training and experience included, that a court might expect.
What about the person who's just had a bone marrow/peripheral stem cell transplant and his allogeneic marrow/stem cell donor is on the SS's hit list? Would current gene testing using a cheek swab sample return a 'false positive' ID?
No, the cheek swab is checking skin cells which are unaffected. A bone marrow transplant would change the white blood cells.
Someone I know had such a transplant, and I believe that one of the tests used to detect/confirm extent of engraftment was a cheek swab test. Blood work definitely comprised the large majority of tests, but there was also a buccal swab test.
In my friend's case, the stem cell transplant also showed up as a 'gender' change on some of the tests. Very weird, and I recall our discussing scenarios where this level/type of testing might become an issue when traveling, certain types of jobs, etc.
Going back to the original subject (!) Perhaps I should quote a thriller ( & supposedly SF ) writer, who has coined, or re-cycled, I'm not sure which ... Three Laws of Security: 1: The desire of those in charge of security, for information, will continue in a straight line, with no limit, short of the heat-dath of the universe. 2: The willingness (& ability) of their superiors [ The guvmint ] to supply them with the monies needed to accomplish their tasks, has very definite limits. 3: This means that the infoprmation collected & assembled by the security authorities inevitably overwhelms their capacity to analyze that information.
Thus, they are, effectively suffocated by their own insecurity.
Whitest kids you know touched on this: https://www.youtube.com/watch?v=v4z09el30f8
Security?
Ho Hum, as of today from a local newspaper that is available over the internet ...BUT should I be disclosing this? ....
" But his latest request for information has led to Defra invoking a regulation which states disclosure would “adversely affect international relations, defence, national security or public safety”.
Mr Latimer requested an update from the government department over information it provided to the European Court of Justice surrounding a breach of a water treatment directive at Whitburn.
And the 71-year-old, who runs the Latimers Seafood Deli in Bents Road, Whitburn, has described the response as “comical”.
He said: “Defra say to disclose the information I am requesting would adversely affect international relations, defence, national security or public safety, and then they go further, stating disclosure would adversely affect the course of justice.
“This is quite comical and makes one wonder why they are withholding the information when all I am doing is asking about a sewerage system.”
Mr Latimer insists disclosure is in the “public interest”, adding: “Withholding information discriminates against the public interest. The public need to rely on Defra to have the ability, as a public authority, to conduct an inquiry in a fair and honest manner.”
In response Ed Beard, head of waste water treatment with Defra, said it was currently in negotiations with the European Commission to rectify the water treatment breach, and claimed disclosure of the information requested would “adversely affect the UK’s relations with the European Commission”."
http://www.shieldsgazette.com/news/local-news/campaigner-laughs-off-security-threat-snub-1-6664476
Now IF that is the outfall site that I'm thinking of then its a charming little cove between steep, limestone, cliffs and its not very far away from an out to sea dumping site for wastes dredged from the tyne side shipyards ...
"CONTROVERSIAL plans to dump a staggering 60,000 cubic metres of dredged waste off a coastal beauty spot has sparked an environmental row.
"The material, which contains contaminants above levels normally accepted for sea disposal, comes from the dredging of the Tyne – and will be disposed of just off Whitburn."
http://www.sunderlandecho.com/news/local/all-news/tyne-waste-dump-row-1-1117891
But doubtless that info is ever so TOP SECRET too ..as is any info that you might possess on environmental polution nearabouts of your neighbourhood.
Oh, that off shore dumping site? Apparently it is over a NATURAL hole in the sea bed and it has been capped with clay ..though local skin divers told me a few years ago that that clay is nowhere near as thick as was claimed.
Mind you you have to question the sanity of people who pluge arround in the depths of the sea sticking metal probes into sites that are reputed to be sealing off highly toxic metalic wastes.
Ah well, I expect that it's all right really. What could possibly go wrong?
We need a few marine biology/environmental science grad students who need 'just one more paper/project'. If they conducted the same study/analysis at multiple points along this coast line, it would provide good comparative and benchmarking information. Crowd-sourcing might work for this.
The whole idea is so stupid in so many ways that is practically violates the Pauli Exclusion Principle for ridiculousness. 's Absurdity #1: Secret service asks for a hard AI able to parse human language with human-level subtlety, which, if anyone could create it, would transform the world far more than they could possibly imagine. But of course there's not the slightest indication that such a linguistic hard AI is even remotely possible, and a great deal of circumstantial evidence that it's a pipe dream. (I.e., simple and obvious conundra like: "Mary saw the puppy in the window and wanted it. Which did Mary want, the puppy or the window?" Which no computer program can answer today.)
Absurdity #2: Even if they get this ludicrous linguistic hard AI, it would drown in so many petabytes of surveillance that it's still futile.
Absurdity #3: The NSA today claims that its surveillance systems are so complex that it can't prevent 'em from deleting older surveillance data in order to comply with privacy requests. This suggests that the NSA has built a system so complicated that it's run out of control, operated by algorithms no one understands and no one can make useful changes in because random tweaks would crash the system. The analogy here? High-speed financial trading which runs amok based on algorithms whose consequences humans can't predict, and which consequently periodically crashes the planet's global financial system.
Kirkpatrick Sale wrote a great book about these kinds of complex systems that exhibit unpredicted uncontrollable behavior: Human Scale.
Absurdity #4: As Bruce Sterling pointed out, there are multiples of intelligence. What if we succeed in creating a true AI, but it's really good as kinesthetic intelligence (dancing) but it's terrible at verbal intelligence (can't explain why it does what it does)?
The whole thing is a classic example of a degenerating researching project, like alchemy. More and more resources get sucked into the failed research project, producing increasingly absurd results, until the whole thing implodes in a shower of embarrassment and universal ridicule.
I'm late to the conversation and this isn't sarcasm, but ISIS communicating through twitter with an app called The Dawn of Glad Tidings seems pretty close, even if in the inverse. Perhaps that is the effect of trying to identify sarcasm and threats: the real threats are going to try to sneak in as innocuously as possible.
Has this been posted yet?
The FBI tried to compile a list of internet slang terms, only to end up with severe EOF (Egg On Face):
http://www.washingtonpost.com/news/the-intersect/wp/2014/06/17/the-fbi-maintains-an-83-page-glossary-of-internet-slang-and-it-is-hilariously-frighteningly-out-of-touch/
I wonder what their entry for "We are aware of all internet traditions" looks like. Which level of irony escaped them?
The FB are USian, and government, and law enforcement, so I'd say "all levels of irony, including ones not even known to the rest of us".
"Even if they get this ludicrous linguistic hard AI, it would drown in so many petabytes of surveillance that it's still futile."
Yes, but they don't need something that works. They need something they can claim they believe will work.
If they get an assassination attempt that they did not predict, that had internet messages they should have predicted from, that will look bad after it happens provided the public hears about it.
But right now they look bad for getting buried in petabytes of surveillance.
If they can just get a good excuse to ignore that stuff they will be off the hook until they get caught with a false negative.
So they need somebody to sell them a product that gives them a small manageable stream of false positives. Then they can assign people to look at that blather as a punishment detail, and they're happy. If it actually worked, that would be even better. But they don't need it to work, they only need to be able to say they don't know it doesn't work.
This article seems to fit this thread: The FBI maintains an 83-page glossary of Internet slang. And it is hilariously, frighteningly out of touch.
Apparently somebody in the FBI thinks that any acronym used on Twitter must be in common usage.