DARPA Hopes To Develop an AI Tool That Can Detect Deepfakes (nextgov.com) 79
America's Defense Department "is looking to build tools that can quickly detect deepfakes and other manipulated media amid the growing threat of 'large-scale, automated disinformation attacks,'" reports Nextgov:
The Defense Advanced Research Projects Agency on Tuesday announced it would host a proposers day for an upcoming initiative focused on curbing the spread of malicious deepfakes, shockingly realistic but forged images, audio and videos generated by artificial intelligence. Under the Semantic Forensics program, or SemaFor, researchers aim to help computers use common sense and logical reasoning to detect manipulated media.
As global adversaries enhance their technological capabilities, deepfakes and other advanced disinformation tactics are becoming a top concern for the national security community... Industry has started developing tech that use statistical methods to determine if a video or image has been manipulated, but existing tools "are quickly becoming insufficient" as manipulation techniques continue to advance, according to DARPA. "Detection techniques that rely on statistical fingerprints can often be fooled with limited additional resources," officials said in a post on FedBizOpps...
Beyond simply detecting errors, officials also want the tools to attribute the media to different groups and determine whether the content was manipulated for nefarious purposes. Using that information, the tech would flag posts for human review. "A comprehensive suite of semantic inconsistency detectors would dramatically increase the burden on media falsifiers, requiring the creators of falsified media to get every semantic detail correct, while defenders only need to find one, or a very few, inconsistencies," DARPA officials said.
But that's easier said than done. Today, even the most advanced machine intelligence platforms have a tough time understanding the world beyond their training data.
As global adversaries enhance their technological capabilities, deepfakes and other advanced disinformation tactics are becoming a top concern for the national security community... Industry has started developing tech that use statistical methods to determine if a video or image has been manipulated, but existing tools "are quickly becoming insufficient" as manipulation techniques continue to advance, according to DARPA. "Detection techniques that rely on statistical fingerprints can often be fooled with limited additional resources," officials said in a post on FedBizOpps...
Beyond simply detecting errors, officials also want the tools to attribute the media to different groups and determine whether the content was manipulated for nefarious purposes. Using that information, the tech would flag posts for human review. "A comprehensive suite of semantic inconsistency detectors would dramatically increase the burden on media falsifiers, requiring the creators of falsified media to get every semantic detail correct, while defenders only need to find one, or a very few, inconsistencies," DARPA officials said.
But that's easier said than done. Today, even the most advanced machine intelligence platforms have a tough time understanding the world beyond their training data.
It will help them develop better fakes (Score:2)
Tools like this will allow DARPA to tweak deepfake algorithms to produce fakes that are even harder to detect. It looks like another arms race, developing both defensive tools to detect fakes, and offensive tools that make fakes that the best defensive tools can't detect. I think we're kind of screwed.
Re: (Score:2)
Tools like this will allow DARPA to tweak deepfake algorithms to produce fakes that are even harder to detect. It looks like another arms race, developing both defensive tools to detect fakes, and offensive tools that make fakes that the best defensive tools can't detect.
What you have described is exactly how deep-fakes are created in the first place: Generative Adversarial Networks [wikipedia.org].
Re: (Score:2)
Yeah GAN was the first thing that came to mind for me. The easiest way to build a bot that can fool a human is to also build a bot that can detect the fake. Then sic em on each other and let evolution do the rest.
Humans are in serious trouble if we ever manage to find a use more malicious than "Make fake porn" for these things..
Re: (Score:2)
Indeed. That was my first thought.
In principle this whole "Deep Fake" business would not even be much of a problem, were it not for countless people that cannot do plausibility checks and will believe even the most obvious lies.
Re: (Score:2)
I agree. People believe "Shallow Fakes," as it is.
Plausibility: not even new (Score:2)
and people not checking sources, etc. is already a problem long before deep-fakes:
be it mis attributed video-snippets or photos, or studio produced content, etc.
Re: (Score:1)
It's completely worthless anyway, because people who want to believe that the deepfake is won't believe a black-box AI telling them that it isn't.
Re: (Score:2)
Re: (Score:2)
It depends on how the fake detection algorithm works. But I don't really think it will help that much. Except if the goal is to make deepfakes that are undetectable by their own algorithms.
Deepfakes are already designed to fool humans. And I expect the DARPA algorithm to use other methods. For example it may notice that deepfakes tend to have more blue pixels on the lower left or any detail that humans are likely to overlook.
Making the deepfake "DARPA-proof" is possible (as other people mentioned, GAN could
Re: (Score:1)
Why? (Score:2)
We want the peepee-tape, even if it's deep(fake).
No false positives (Score:2)
I think things are going to be hectic enough with enterprising trolls getting ahold of footage before it's published (or before it goes viral), using deepfake software on it without changing anything (i.e. not changing the content, just making sure it triggers the alarm), then
Re: (Score:3, Interesting)
For instance, according to Youtube, someone saying that Qur'an-inspired anti-Jewish violence is a problem both in Europe and around the globe constitutes hate speech, e.g. the Pat Condell video "O Come All Ye Jew Haters" [youtube.com], which I can confirm was censured by Youtube on Pat's official channel before it was ultimately removed .
Re: (Score:2)
and the Jewish scriptures are anti-Palestinian, the Israelites told to wipe out the inhabitants and that all the land of Palestine is there
And,the Israelis are acting on that, committing atrocities and violating human rights of Palestinians, it's what "Zionism" is.
So, we should ban all Zionist media, correct? That would be all mainstream U.S. media companies, etc.
Re: (Score:2, Flamebait)
There are million of Muslim citizens o
Re: (Score:2)
Anyway, this is beside the point. Outside of Israel/Palestine, in places like Europe, Jews by and large aren't targeting random Muslims the way they're being targeted.
Re:No false positives (Score:5, Interesting)
You are making hate speech with lies. Showing your true evil colors now.
The Israeli have been displacing people for decades, including systematically committing murder and maiming, then retaliating with grabbing more land with invasions and mass murder. They drew "first blood". The Israeli's have the huge body count, besides thousands upon thousands of Palestinians they have displaced. The Israelis have made huge areas of land that isn't theirs into concentration camps with their barriers and checkpoints.
The Israelis are the ones doing the "ethnic cleansing" of regions.
Last July Israel passed their racist apartheid bill making arab muslims officially second class citizens. Of course, a non-Jewish person can't even get married in Israel, they have to go to another country because it's a theocracy.
Re: (Score:2)
Of course, a non-Jewish person can't even get married in Israel.
A total lie. There is no secular marriage, but Christian and Jewish marriages are recognized.
You're free to lie all you want whilst calling the truth lies... that's just how it is in 2019! Definitely don't bother to educate yourself or anything. The fact is, there are indeed millions of Arab Muslim Iraeli citizens who enjoy the same rights and privileges (there are a few small differences--Israel subsidizes some aspects of Judai
Re: (Score:2)
Re: (Score:2)
Re: (Score:3)
for 45+ years of my life I believed as you did, but I changed 180 degrees about ten years ago. I'm only stating my conclusion from studying historic facts from 19th century onward, it is not "propaganda".
There are human rights violations and atrocities being committed in the name of a political and religious movement, calling attention to that is hardly propaganda.
You are the one that called an ethnic group a "cancer".
I'm not saying anyone is a cancer. I want Israelis to have a happy safe life too, realis
Re: (Score:2)
Re: (Score:2)
No lie, people have to go outside israel to get married if not Jewish. That is a fact. You have not offered anything that contradicts that fact. Instead you are trying to deflect about what is "recognized".
No lie about drawing first blood, they marched people off their land and also murdered. Claiming they wanted to sign a piece of paper yet committing those atrocities and that somehow proves they are in the right is silly.
Gaza has been barricaded, turned into concentration camp, bombed and attacked, an
Re: (Score:2)
No lie, people have to go outside israel to get married if not Jewish. That is a fact. You have not offered anything that contradicts that fact.
This is a total lie. People go outside the country to have secular/new age (reform/interfaith/etc.) marriages. Muslims do not need to leave the country to marry in traditional Muslim ceremonies. You are lying. You are repeating lies that portray Israel as a represive, Jew-centric theocracy when in fact Jews and Muslims are living under the same exact restric
Re: (Score:2)
No lie, if a Jew wishes to marry a Christian (or anything else), they have to leave Israel. This is fact. Because it is a theocracy.
They are repressive. None of your brainwashed shilling changes that fact.
Israel isolated Gaza, choking it off and then bombing and assassinating inside of it. This is fact.
The do not seek peace, they murder and take land. You shill for murderers and thieves.
You are the one spewing propaganda.
Re: (Score:2)
what israel does have allow "civil union" as opposed to marriage for many cases, and registering of marriages (for example gay marriages) done abroad
this is different from allowing marriages
Re: (Score:2)
If you don't want to admit that you were (perhaps unknowingly) spewing propoganda designed to make Israel look like it was discr
Re: (Score:2)
Of course, a non-Jewish person can't even get married in Israel
You repeated this lie in at least two different posts, possibly three. You said this as part of a larger unhinged spiel regarding how Israel oppresses Muslims. when in fact they have equal rights (and equal restrictions) under this asinine system that caters to religious conservatives (of all sorts, not just Jewish.)
Re: (Score:2)
There's scarcely a country on earth that hasn't displaced some other people.
Of course, a non-Jewish person can't even get married in Israel.
A total lie. There is no secular marriage, but Christian and Jewish marriages are recognized.
You're free to lie all you want whilst calling the truth lies... that's just how it is in 2019! Definitely don't bother to educate yourself or anything. The fact is, there are indeed millions of Arab Muslim Iraeli citizens who enjoy the same rights and privileges (there are a few small differences--Israel subsidizes some aspects of Judaism, but on the other hand Muslim Israelis are not required to serve the mandatory 2 years or whatever it is in the armed forces.) Yeah they do some symbolic stuff emphasizing that it's a Jewish nation. Seeing as how dozens of countries openly call themselves Islamic nations and blatantly discriminate against non-Muslims in far, far worse ways, I don't have much of a problem with this except to the extent that as an atheist and antitheist I wish they would all be secular. But they're not, and the Jewish people have been incredibly tolerant of other *citizen* minorities, including millions of Arab Muslim Israelis. (They're not required to be tolerant of non-citizen minorities who've declared war on them, obviously.)
They drew "first blood".
Another patent lie. Israel was willing to sign a peace treaty right away, immediately after its founding, but the combined pan-Arab armies of Egypt, Jordan and Syria did. Palestine was "occupied" by Egypt and Jordan for a long time before they suddenly decided that Israel was a tough nut to crack and it was better to make themselves look separate and weak, for sympathy's sake.
Go ahead and make up more lies. It's interesting to behold.
I do not agree with all of Israel's land confiscations and we can argue all day about the conditions surrounding its founding. But this is not in the same category as the murderous hellhole theocracy that is Gaza. Land theft has happened re: every country on Earth. After WWII, Germany had its borders permanently shrunk, and some sources estimate that hundreds of thousands ethnic Germans perished in the forced relocations after the war ended. And you know what? Thems the breaks. We'll try better next time (if there is a next time), but the Germans don't get to launch rockets at France for the next 50 years whilst whining about it. They were the bad guys. They lost. They lost some land and some people. They got the fuck over it and worked on improving themselves and their society. As should the Palestinians. (Who were originally not a small, weak, separate group but part of a combined multinational pan-Arab force--legally part of Jordan and Egypt--that for decades sought to annihilate Israel.)
So what your saying is essentially that DARPA doesn't hope to create an AI tool that can help detect deep fakes? I'm having trouble following your argument.
Re: (Score:1)
The fact th
Re: (Score:2)
Re: (Score:1)
Re: (Score:2)
Re: (Score:2)
YouTube doesn't apply its rules equally. Your own link to a YouTube search has as the second result "Born to Hate Jews" by bullshit factory PragerU, which just happens to be a popular channel with billionaires funding it (don't fall for the donate buttons on their site).
The second part of your post is bollocks. Most anti-Semitic attacks in Europe are by white people, often but not always Christians, and usually radicalized by people like PragerU and Lauren Southern. Jews are the centre of many far right con
Re: (Score:2)
Also, as I recall the Paris gun
Re: (Score:2)
I can't find any statistics. It's not clear if anyone actually compiles them. Wikipedia doesn't seem to have any either (https://en.wikipedia.org/wiki/Antisemitism_in_Europe#21st_century).
Re: (Score:2)
Offhand, I know the Charlie Hebdo shooters killed four Jews at a kosher grocery store [wikipedia.org]. The guy who shot up the pro free speech (sorry, I forget you're one of those "freeze peach" folks, right?) "draw the prophet" thing in Copenhagen later went on to kill a Jew [wikipedia.org] and wound others as a synagogue. I know that three people at the Jewish Belgium [wikipedia.org] were shot and killed by a jihadi. I know a while back there was a guy named Mohammed who shot up a Jewish [wikipedia.org]
Re: (Score:2)
And how do you achieve that? I think 100% accuracy with no mistakes is a pipe dream.
In theory you could digitally sign videos, so you could see who authored the thing. You could tell whether it was from what you think to be a reputable source or just some random dude on the internet. But you again get into the thorny issue of whether you trust the certificate authorities, and if a CA gets hacked all bets are off. And how would you create an excerpt of a video without invalidating the signature?
But the v
Re: (Score:2)
Signing off on something being a fake, or not being a fake, needs to be not undertaken lightly. There needs to be some sort of multilayer process with m
Re: (Score:3)
Like I said, we need an "inconclusive" zone and we need to not be afraid to use it. The point is to avoid the situation where things are labeled wrongly and then we have to backtrack, leaving ordinary folks throwing up their hands in the air and saying they choose to believe X (or to believe what their favorite pundit thinks about a video's authenticity.)
Signing off on something being a fake, or not being a fake, needs to be not undertaken lightly. There needs to be some sort of multilayer process with multiple independent entities using their own separate tools before a definitive pronouncement is made.
Re: Signing, I wonder if we could get cameras that automatically sign their output in an unbreakable way. It's a pipe dream, you don't need to lecture me about how it's a pipe dream, I know I know, but just let me dream, alright? (The issue is there's just way too much incentive to break any system. There's going to be nation-state level funding going towards cracking these things.)
I have deepfake AI technology baked in. I'll prove it.
I see trump saying something stupid, using the vocabulary of a chimp, saying something racist, or saying something like "it's gonna be great you're gonna love it, it's a beautiful thing, I know a lot a really great people, I'm a stable genius, there were good people on both side, or i like people that weren't captured "
I know it's real!
If I see people that have actually been working hard their entire lives for the betterment of humanity - Bernie Sander,
Re: (Score:2)
It doesn't even have to be that obvious. Whenever you year anyone saying anything that seems exceptionally dumb or out of character, it's worth checking their prior history and the context of what was said.
Re: No false positives (Score:2)
Re: (Score:2)
It may be harder, but they will eventually master that as well.
Re: No false positives (Score:2)
Re: (Score:2)
If I remember correctly, John Brunner in his The Jagged Orbit had a take on this, where an investigative journalist of sorts used computers to both figure out stuff he's almost certain actually happened, and then fake video of it.
Re: (Score:2)
Whomever does this, this is the really, really important bit: No false positives. Zero false positives, please. Or false negatives, for that matter. We desperately need an "inconclusive" zone to avoid making false pronouncements.
While that would be nice to have, this is unlikely to be really useful. What you will get instead is either a "inconclusive" range large enough to make the whole thing unusable. Or you will get things that should have been "inconclusive", but instead went to "positive" or "negative".
Statistical classification is a guessing game and always comes with uncertainty. The uncertainty increases when there is an intelligent adversary in the picture.
Re: (Score:2)
lolz
the goal is impossible, someone wanting to make name for themselves in the bureaucracy talking out of their ass.
"deepfaking" is old, and can be done with words as well as audio and visual media.
Re: (Score:2)
What source do *you* trust to determine what is and isn't "fake news"?
Myself personally?
Jesus said to them, "When you make the two one, and when you make the inside like the outside and the outside like the inside, and the above like the below, and when you make the male and the female one and the same, so that the male not be male nor the female female; and when you fashion eyes in the place of an eye, and a hand in place of a hand, and a foot in place of a foot, and a likeness in place of a likeness; then will you enter the kingdom."
--Thomas 22
What? Jesus will return as a deepfake troll? (Score:2)
Jesus said to them, "When you make the two one, and when you make the inside like the outside and the outside like the inside, and the above like the below, and when you make the male and the female one and the same, so that the male not be male nor the female female; and when you fashion eyes in the place of an eye, and a hand in place of a hand, and a foot in place of a foot, and a likeness in place of a likeness; then will you enter the kingdom."
Um. So... deepfakes (and/or transsexuality) are the keys to get into heaven?
Re: (Score:2)
That's the problem with trying to convey esoteric doctrine to an exoteric audience...
But no, better to understand this as "there's nothing new under the sun", and "ultra-modern" technological conceptions have existed for thousands of years in the past, because, well, God is of all time.
On deepfakes, well (exoterically speaking), yes, as "deep" as you can get. On transsexuality, no, better to look in the direction of what comes before male/female differentiation, or what category we are in (or contain) that
Re: What? Jesus will return as a deepfake troll? (Score:2)
Re: (Score:2)
Many ways. They are quite distinct worldviews, though there's a cursory parallel of aspects of nonduality.
If I needed to give a simple summary distinction, I'd say Buddhism tells you to accept your plate is empty, while Gnosis fills your plate.
Re: What? Jesus will return as a deepfake troll? (Score:2)
Re: (Score:2)
How to summarize? Filled with the calm and sense of sufficiency of knowing, as distinct from hoping or believing, that reality is designed for the eternal provisioning of one's needs. The specific forms of that is really something one has to experience for oneself. It is, fundamentally, esoteric rather than exoteric. Hidden, rather than advertised ("postable"), knowledge.
But very attainable. If I were to suggest a good overview book of "actual Gnosticism" I would start with Elaine Pagels [amazon.com].
I say "actual
Re: What? Jesus will return as a deepfake troll? (Score:2)
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
"Morally" derived from what? Evolution?
Since naturalistic notions of evolution can't make any "moral" statement other than "some DNA wins, some DNA loses, the end", I propose in fact your notion of "morality" is determined entirely by cultural assimilation, and that being assimilated from religion.
Ultimately this seems to be a "I reject your religion based on my morality, which has no argument or basis except for your religion being correct" argument.
Everything -necessary- is given unambiguously. You may
Re: (Score:2)
Would you find the equivalent in an eternal sense "sadistic and unfair"?
Only if an malicious and intelligent agent were behind it. Again; this is not something I borrowed from religion--it predates it. Intelligent mammals react very, very differently to violent
Re: (Score:2)
Instincts are not morality. You have "morality" when you can propose moral axioms that can be justified. And the issue with every secular notion of morality is that it is entirely subjective (hence useless) and unbacked by anything in material existence that specifies a given norm as objectively correct (hence useless). Not to mention that the domain of morality is the domain of -choices-, and "instinctual morality" is a contradiction in terms.
Formally, that's called the Is-ought Problem in philosophy.
Re: (Score:2)
Instincts are not morality.
But instincts plus intellectual introspection is. Morality is not synonymous with authoritarianism. Religion took preexisting authoritarianism (at its most basic, the authoritarianism of parents over children, but also chiefs / alpha males over the tribe), preexisting instincts of fairness and solidarity and other social instincts, and ran it all through a fictional narrative. It didn't bring anything new to the table other than the fiction... which could sometimes be a source of solace, sure, although if y
Re: (Score:2)
Strange goalpost shift to "authoritarianism", I have not suggested that, and the only way to make that segue is to suggest that equivalently if I say "gravity is objective because gravity is based in something actually existing in reality" I'm making an "authoritarian" claim. No, my claim is that it is based on something objective that exists. You do not accept the existence of what I am basing my ethics on, but it is essential that -something- be proposed, regardless of the ethics. For the reasons alrea
Re: (Score:2)
Why is something "moral" per theism? Because God says so.
I think this is a pretty straightforward example of authoritarianism, no?
Because I reject authoritarianism as the basis for objective truth, that allows me to reject a lot more than the Judeo-Christian God. I can also reject the silly consensus of self-appointed philosophers, who proved their own incompetence (and you just might agree with me here) in the way they allowed Derrida to garner and retain so much acclaim.
Hume's is-ought is an important concept but overblown and misunderstood. The
Re: (Score:2)
Conditional, non-categorical imperatives are widely recognized to be comprehensive and logically true ("*if* you don't want to burn yourself, you *ought* to not stic
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Hmm... this thread is becoming so long, both you and Harris will probably be taken out by evolution before it finishes.
So, in the interests of time, here's my final rebuttal:
*Waits*
Re: (Score:2)
Hmm... this thread is becoming so long, both you and Harris will probably be taken out by evolution before it finishes.
Unfortunately, this new tact of yours has already failed. I've already reproduced (as has Harris). There is also the "reproduction" or natural selection of ideas, more important than genes, though I wouldn't want any of these ideas/memes to spread in a dogmatic fashion. You either think about and engage with these things, or you don't.
It really takes no time at all to admit one's abuse or at least sloppy use of logical concepts and English words. If the meaning I have on offer isn't deep or profound enoug
Getting rather complex, isn't it? (Score:2)
Two separate probelms (Score:2)
One is detecting deepfakes. The other is convincing a non-technical audience that you have correctly identified a deep fake.
Imagine a video emerged claiming to show President Trump personally committing a horrible crime 20 years ago. What organization would people who support both parties believe to identify the video as real or fake, since inevitably there would be claims of both.
(this is not intended to start an argument over whether the president did or did not commit any crimes, just using it as an ex
The solution is the problem. (Score:2)
Any solution to detecting neural network manipulated video will be used to create a better neural network which your solution can no longer detect. Digital forgery is an arms race that has no winner. (Time to go analog? o_O?)
Re: The solution is the problem. (Score:2)
Why? (Score:2)
deepfakes and other advanced disinformation tactics
IMO, these tools are aimed at protecting the credibility of the paid pundits and leadership figures. Our media talking heads can be paid or tricked to generate disinformation. No fakery needed. But the only way to combat this is to cultivate some critical thinking skills in the population. But then this might come back to bite the so-called leaders in the ass when they need to go on screen and issue edicts.
The basic problem ... (Score:2)
... is that for every asshat out there with a computer, there's another asshat out there with a computer.
People creating videos, manipulating videos, creating mashup Deep Fakes, crafting Deep Fake detectors, building workarounds for anti-Deep Fake detectors ... are all working with 100% the same technology as a 5 year old kid.
Because the DNA is the same, twins are indistinguishable.
Ultimately, the solution will be what it has always been: Human consumers believing what they want to believe.
Sadly, that popul
Don't need digital deep fakes (Score:2)
Control the media, you control the populace's beliefs and opinions
Deep Faking events is ancient technique
DARPA doing the research for Deepfakers (Score:2)
As soon as DARPA creates this AI, the deepfake makers will use it to train thier Deepfake AI to do better Deepfakes.
I personally hope the deep fakes increase because (Score:2)
the deeper and more obvious the bottomless pit gets the sooner people will start thinking for themselves.
Re: (Score:1)