Political deepfakes
« previous post | next post »
Daysia Tolentino, "Trump shares fake photo of Harris with Diddy in now-deleted Truth Social post", NBC News 9/20/2024:
Amid the recent news of Sean “Diddy” Combs’ arrest, former President Donald Trump reposted a doctored image falsely showing Vice President Kamala Harris with Combs with text questioning if she was involved in his alleged “freak offs.”
The image, which Trump reposted to his Truth Social profile, is an edited version of a 2001 photo of Harris with former talk show host Montel Williams, whom she briefly dated, and his daughter Ashley. The edit replaced Montel Williams’ face with a photo of Combs.
This is not the first time the Republican presidential nominee has posted a fake image in an effort to bolster his campaign. Trump has posted several AI-generated images, including some falsely depicting Taylor Swift and her fans endorsing him, and one of Harris speaking to a crowd of communists in Chicago during the Democratic National Convention.
Thanks to the Trump's Truth archive, here's the now-deleted faked picture of Harris with Combs:
…and the post with the faked "Swifties for Trump" pictures:
Here's an image of the (more obviously faked) picture showing Harris addressing communists, from X-formerly-Twitter:
The potential for this sort of thing is further (and even more strongly) exemplified by the "exploding goats" video. Jordan Liles, "Exploding Goats? Fake CNN Video Claimed Israel Targeted Hezbollah with Bizarre New Strategy", Snopes 9/20/2024:
A rumor circulating online in September 2024 claimed a video showed CNN anchor Jake Tapper reporting that Israel targeted Hezbollah — a Shiite Muslim political party and militant group based in Lebanon — with exploding goats. […]
However, the video clip was fake. CNN never ran any such report. Stand-up comedian, actor and writer Danny Polishchuk created the video as a gag. He also played the role of the field reporter opposite Tapper. Polishchuk engineered Tapper's mouth movements and vocal sounds with deepfake video technology and artificial-intelligence tools.
The video is quite well done — a step forward from last year's "Failure to Launch" — though the silly content means that most people would be skeptical:
BREAKING NEWS:
Mass Hezbollah Casualties Reported After Goats Rectums Explode.
— Wall Street Silver (@WallStreetSilv) September 20, 2024
A less absurd deepfake of similar (or better!) quality could go very far before being debunked.
Update — see Shannon Bond, "U.S. officials say Russia is embracing AI for its election influence efforts", 9/23/2024 NPR.
Philip Taylor said,
September 23, 2024 @ 5:41 am
I am obviously missing something, but could Mark possibly explain the relevance of this story to a web site ("Language Log") intended for the discussion of language and linguistics ?
Stephen Goranson said,
September 23, 2024 @ 6:09 am
It's about communication, with words and otherwise. Why complain, PT?
Mark Liberman said,
September 23, 2024 @ 6:58 am
@Phliip Taylor:
The fake voices and mouth movements in the exploding goats video are applied straightforward applied phonetics; the fake (I assume) t-shirt lettering has a linguistic dimension; but as Stephen Goranson says, the whole thing is about politic0-cultural communication…
J.W. Brewer said,
September 23, 2024 @ 7:31 am
A genuine file photo of A standing next to B doctored to make it appear as if A is instead standing next to C is the sort of thing that could be and was generated by old-school analog (or "artisanal") hoax techniques back into at least the mid-Twentieth Century. Muddling up that sort of old-school thing with trendy news hooks about "AI" or "deepfakes" seems likely to obscure rather than illuminate what is genuinely novel, which the pseudo-Tepper in the exploding-goats clip might be.
Benjamin E. Orsatti said,
September 23, 2024 @ 7:37 am
Stephen,
Why complain? Because we come here out of intellectual curiosity, seeking fun, interesting things about human language posted by people who know more about this sort of thing than we do. But, oftentimes, what we find is not really about "language" at all; it's about… something else.
Stephen Goranson said,
September 23, 2024 @ 7:55 am
BEO:
"we"?
Philip Taylor said,
September 23, 2024 @ 7:57 am
Stephen — I rarely, if ever, complain (except when a comment is deleted without apparent reason or explanation) and my question (above) to Mark was just that — a simple question, with no ulterior motive whatsoever. As my maths master told me 60+ years ago : "There are no stupid questions, Taylor, only stupid answers".
Don said,
September 23, 2024 @ 7:59 am
I think Professor Liberman just sometimes wants to post about American politics as well as language. I don't really begrudge him that: it's his blog and it's free. (That said, I personally usually skip the Trump posts, and do kind of feel a pang of exhaustion at hearing about Trump everywhere.)
By the same token, I don't know that the author necessarily has to feel like he's on the hook to figure out a language tie-in every time (which often ends up seeming like a stretch to me – everything is "about communication" if you squint hard enough).
Stephen Goranson said,
September 23, 2024 @ 9:03 am
Like it or not, Mr. Trump uses language.
Benjamin E. Orsatti said,
September 23, 2024 @ 9:14 am
Don is right — we're marching into Prof. Liberman's stadium, playing with his ball, and complaining to a non-existent "ref" about unsportsmanlike conduct. I guess it's helpful to bring the complaint out into the open a few times, but, once it's clear that the caution has either been or not been heard, there's no sense in routing the Complain Train on a repeating loop.
Instead, I guess we just expect a certain corner of LL to be "Inimicos Libermani delendi sunt" for awhile.
Mark Liberman said,
September 23, 2024 @ 9:51 am
@J.W. Brewter: "A genuine file photo of A standing next to B doctored to make it appear as if A is instead standing next to C is the sort of thing that could be and was generated by old-school analog (or "artisanal") hoax techniques back into at least the mid-Twentieth Century. Muddling up that sort of old-school thing with trendy news hooks about "AI" or "deepfakes" seems likely to obscure rather than illuminate what is genuinely novel, which the pseudo-Tepper in the exploding-goats clip might be."
The innovation in photo-hacking is less than in deepfake video creation, but it exists — because it's getting easier, quicker, and cheaper to do what used to take time, expertise, equipment, and money.
Haamu said,
September 23, 2024 @ 12:26 pm
Delendi? Seriously?
Haamu said,
September 23, 2024 @ 1:22 pm
There are several legit linguistic angles here beyond (or simply extending) the ones already mentioned:
"Fake" vs. "Deepfake" — I'm with J.W. Brewer, I think: the first example is what we'd normally call a fake, not a deepfake. It's basic image editing, whether or not the latest tools use AI to simplify that.
I would have expected "fake" to take over for "deepfake" eventually as people got more used to the new capabilities and techniques. For the moment, though, I sense that "deepfake" is encroaching on "fake." Perhaps "deep…" is beginning to signify not "AI-assisted" or "new capability" but "good enough to fool me, at least initially."
Text rendering within AI-generated images — Getting accurate text to appear in a generated image simply by specifying it in the prompt has been notoriously hard. My own experiences with DALL-E, last attempted a couple months ago, bear this out. Images like some of those in the "Swifties" example would likely be a combination of AI generation followed by photoshopping the actual text.
Text-rendering deficiency is unsurprising, given my guesses about what the models are doing. But I also wouldn't be surprised if the issue has already been resolved by some models, given that rapid pace of advancement. If true, then the interesting question is: How are they doing it?
(Incidentally, the only complaint I have about this post is that the first example was reproduced in full, complete with the scurrilous caption. (If it had to be shown at all — and there's plenty of psychological evidence that debunking a false image doesn't undo the harm of showing it — then there are plenty of available versions that show just the altered photo with no words.)
Bybo said,
September 23, 2024 @ 1:32 pm
Every time someone says this to me, I'm tempted (and more often than is good for me, which is probably never, give in to the temptation) to reply: 'Challenge accepted!' before asking the question I was going to ask …
Yves Rehbein said,
September 23, 2024 @ 2:48 pm
@ Bybo, that's a slippery slope at the end of which a question is "out of the question" or simply not a question. It can be approached from logic, theorizing about syntax and semantics, or with rhetorics, disquallifying the question as illegitimate under some rule system. The quintessence of the motto is that a "master" (thus Phil Taylor) should be able to point out the mistake, and that a student ought to learn from it. Your "Everytime someone says this to me …" works out to your benefit only if your questions are not actually stupid enough. Corollarily, this blog post's relevance to linguistics does not compute, logically, while the other side of the fence claims house rules, the question being whether this works out to your benefit or not.
David Marjanović said,
September 23, 2024 @ 4:07 pm
Ha! Good luck finding that many communists in all of the US – and putting them in a uniform no less!
Should be inimici, not the accusative inimicos.
Jonathan Smith said,
September 23, 2024 @ 5:53 pm
the thread is pretty obviously about PUFF DADDIES FREAK OFFS which should clearly be PUFF DADDIES'S FREAK-OFF'S
Andrew Usher said,
September 23, 2024 @ 7:09 pm
Puff Daddy's Freak-offs, of course, but you probably knew that (and I don't even know what a freak-off is in this context, but I can get the grammar right).
I agree that it would be useful to keep the distinction between 'fake' and 'deepfake", if the latter refers to something truly new, but it's going to be too subtle for most. This is just a 'fake' – and photoshopping has been around a long time – it's decades since anyone needed to do photo manipulation the analog way for normal purposes. Putting one person's face with another's body is not new.
The confusion here origirnates with the posted news article, it seems, which doesn't use the term 'deepfake' but call this 'AI-generated' to mean the same thing as 'computer-generated' which is what would traditionally have been said. I'm not convinced, and the author shouldn
't have been, that any deep AI is involved or necessary.
k_over_hbarc at yahoo.com
Benjamin E. Orsatti said,
September 24, 2024 @ 7:45 am
David Marjanović said,
Right; "they", not "them". Maledictus sit school system quod numquam taught mihi proper Latinæ!
KeithB said,
September 24, 2024 @ 9:00 am
@Mark Liberman:
"because it's getting easier, quicker, and cheaper to do what used to take time, expertise, equipment, and money."
Serendipitously the Online Photographer just had a post about exactly this:
https://theonlinephotographer.typepad.com/the_online_photographer/2024/09/ttk.html
with examples
Bybo said,
September 24, 2024 @ 1:11 pm
@Yves Rehbein, and regarding 'there are no stupid questions'
So, be that as it may, my motive in the relevant situations is of course to give a silly answer to what I consider to be an overused cliché. Or rather, because the phrase is used patronisingly, I am tempted to answer in a falsely humble way.
(For what it's worth, I could do without the reproduction of right-wing memes, even if it only happens in the context of otherwise interesting linguistic arguments. However, I'm pretty biased at the moment. I'm freshly frustrated by the fact that the local post-Nazi party is once again bringing American Republican slogans into German public discourse. I wish I could just say: nicht mein Affe, nicht mein Zirkus. Unfortunately, everyone here is once again jumping over every stick that is held out to them. I hate this timeline.)
Matt Juge said,
September 25, 2024 @ 9:09 am
Perhaps some readers missed the tag "Artificial intelligence", a topic discussed in the post and in numerous prior posts.
BTW, BEO may have been thinking of doceo, which takes a double accusative: me Latinam.