2650: Deepfakes

|< < Prev Comic #2650 (July 25, 2022) Next > >|

Deepfakes

Title text: If so great a deductive mind as Arthur Conan Doyle can be fooled by the Cottingley Deepfakes, what chance do we mortals have? Soon our very reality will be dictated by the whims of Frances (9) and Elsie (16).

Explanation

A deepfake is an altered video, designed to deceive, by replacing a person in a video with someone else. White Hat believes that this technology will make it difficult to trust video evidence from now on. However, Cueball responds by saying that "fakes" have always existed, in photos (either through alterations by software such as Adobe Photoshop, or deliberately staging faked images, for example of the Loch Ness Monster) and even more so by people simply lying. White Hat comes around to Cueball's position and says that even the written word is prone to deception and lying.

Scientific studies of deepfakes have produced surprising results, suggesting that they are more likely to increase uncertainty than persuade,[1] that their increased prevalence could inoculate the public against disinformation,[2] and that they are more likely to be shared because of their humorousness than persuasiveness.[3] Other studies have found that deepfakes are persuasive, especially among those who are unfamiliar with them.[4][5]

The complaint tablet to Ea-nasir is a 3,800 year-old clay tablet containing the oldest known written complaint, in which a customer complains to a merchant, Ea-nasir, about the quality of his copper ingots. Cueball's last statement says that perhaps this complaint could have been a lie to begin with, and there was nothing wrong with Ea-nasir's wares. This supposition is arguably the humor of the comic, apart from the hyperbole of the title text.

The title text references the Cottingley Fairies, a series of five photographs produced in 1917 by two children, Elsie Wright and Frances Griffiths, who were 16 and 9, respectively. The photographs appear to show the children playing with fairies in their garden. The photographs received widespread attention when Sir Arthur Conan Doyle, the author of the Sherlock Holmes stories, used the photos as proof of paranormal phenomena in a 1920 magazine article. Conan Doyle was noteworthy for being a strong proponent of reaching conclusions based on evidence and reason, and also held a deep belief in paranormal and supernatural phenomena. In 1983, Elsie and Frances finally confessed that the photos had been faked, by the simple process of making and posing cardboard cutouts traced from figures in a children's book with wings added. Due to technical advances, young children now can more easily create convincingly realistic fakes,^{[actual citation needed]} but similarly there are many more self-styled 'experts' willing and able to dedicate themselves to 'proving' one or other side of any argument about authenticity.

A similar dilemma was alluded to in the 1958: Self-Driving Issues comic, where technology does not create a new way to lie, but may make certain lies more convincing to some parties, such as self-driving cars in that comic.

Transcript

[White Hat and Cueball are walking to the right.]

White Hat: Thanks to deepfakes, soon we won't know what's real anymore. Video will become meaningless.

[White Hat and Cueball walk on, Cueball lifts one hand with the palm up.]

Cueball: I dunno.

Cueball: We've had Photoshop for decades and staged photos for centuries.

Cueball: It hasn't made photos meaningless.

[Zoom in on Cueball, who is turned left towards off-panel White Hat.]

Cueball: The bottleneck for fake stuff isn't technical. The bottleneck is willingness to lie.

Cueball: "People lying" is a very old problem.

Cueball: It's a known exploit.

[Zoom back out on White Hat and Cueball who have stopped. White Hat has a hand on his jaw. Cueball holds his hands out to the sides.]

White Hat: I guess technically we've been able to make text deepfakes for 5,000 years.

Cueball: Maybe Ea-nasir's copper ingots were actually fine!

 add a comment! ⋅  add a topic (use sparingly)! ⋅  refresh comments!

Discussion

Is it worth mentioning that this comic is merely sincere discussion, without (please correct me if I'm wrong) any sort of a joke or irony? The closest it gets is hyperbole in the title text. I know it's not unique in this respect, but it does seem to be different than other such comics because it seems like it might have a joke, given the obscurity of the Ea-nasir reference. If our job is truly to explain, should we let people coming here to figure out the humor know there isn't any? 172.70.214.95 06:48, 26 July 2022 (UTC)

Did but reverted, other opinions? 172.70.210.145 08:39, 26 July 2022 (UTC)

I have reverted and added more. I just used a long time on wiki because of those two tidbits of info that has nothing to do with Deepfakes...  :-) --Kynde (talk) 09:49, 26 July 2022 (UTC)

Snap... I (not the above IP) was also on a long wikiwalk. (Did you know that the map of the copper-fraudster's house is one of the top 200 diagrams that is considered important to resubmit in vectorised format? Amongst many colour-model diagrams and how much money goes to which US surveillance and intelligence agencies. :P ) I really ought to do something important, instead. Like vectorise some diagrams. Hand me my spline-wrench and my gradient-planer! 172.70.86.34 10:17, 26 July 2022 (UTC)

Not sure about there being no joke. White Hat realizing that you can write untrue things (most common types known as lies and fiction), that people have done it for a long time and calling it the new buzzword ("text deepfakes") certainly was funny to me. Cueball's somewhat obscure reference (which you don't really need to know to understand) drives home the point.627235 (talk) 10:52, 26 July 2022 (UTC)

The Ea-nasir punchline made me laugh, I think its a bit of a stretch to say there's no joke here 172.70.86.34 11:00, 26 July 2022 (UTC)

The text about no joke is utterly wrong. There's no single punchline, but this is a very funny strip. Nitpicking (talk) 11:02, 26 July 2022 (UTC)

It's though-provoking, interesting, and insightful — maybe even profound — but I wouldn't call it humorous. It's probably a good idea to put something in for people like me who come here looking for the joke. 172.70.211.90

The Ea-nasir reference is hilarious. Fake product reviews on clay tablets!--162.158.134.11 12:34, 26 July 2022 (UTC)

Or "Instead of Copper Ingots, package contained Near Eastern Wildcat"... 172.70.85.5 21:23, 26 July 2022 (UTC)

Not all humor is 'laugh-out-loud' or 'clownish' - - - or, even necessarily 'funny' depending upon your definition. 17:37, 26 July 2022 (UTC)

It's all a matter of taste. For example - take the closing reference in the explanation "A similar dilemma was discussed in 1958: Self-Driving Issues, where technology does not create a new way to lie, but may make such lies more convincing to certain parties (in the other strip, self-driving cars)." --- although there is no punchline it is humorous (absurd) because there were no self-driving cars 64 years ago, and I am pretty sure XKCD was not even around in 1958. Or, maybe it is an example a text deepfake provided for elucidation. DMG (talk) 17:56, 26 July 2022 (UTC)

For the record, that last reference is referring to xkcd comic #1958, not the year 1958... Mathmannix (talk) 02:00, 27 July 2022 (UTC)

Thanks for that Mmx - but I intended my comment as an ironically absurd 'me-being-clueless' joke about ambiguity in humor. I will be more explicit next time. LOL. DMG (talk) 04:00, 31 July 2022 (UTC)

Poe's law in action, almost exactly as he described in The Pit And The Pendulum... 162.158.34.61 09:22, 31 July 2022 (UTC) ;)

Thanks for reference. I will attempt to commit that mistake 'nevermore.' (Yes, I NOW know which Poe - lol). DMG (talk) 20:17, 31 July 2022 (UTC)

I was thinking of the fact that White Hat is actually convinced and that Cueball is not looking down upon him or thinking hi is stupid. Many of those conversations ends pretty bad for White Hat, which is not the case here. So after reading the above I agree that saying there is no joke is wrong. But I still think the play out of this comic is far from the regular style of White Hat and Cueball conversations. Another where they discuss photographs, one of my favorites, 1314: Photos, pans out a more normal way for this type of conversations. Not sure it needs explanation, but that was why I felt that there was some true part in the "no standard joke" idea. Because there was no joke on White Hat. --Kynde (talk) 06:54, 27 July 2022 (UTC)

Was looking for other examples of what I feel is more standard and found these: 2557: Immunity, 2555: Notifications, 2475: Health Drink, 2368: Bigger Problem and 2165: Millennials. Needed to go back three years to find five, so they are not all over the place. But I only looked through about a third of the 156 comics with White Hat. ;-) But there is probably also some similar to this one. I just noticed these where White Hat is frowned at and remembered them. --Kynde (talk) 07:18, 27 July 2022 (UTC)

I would have said that at least part of the humour is that White Hat can only conceptualise things in a technological framework - hence why Cueball has to use a technologised expression ("It's a known exploit") to explain the point to him (lying is part of being human). 172.70.91.80 09:53, 27 July 2022 (UTC)

It's worth noting that xkcd is "a webcomic of romance, sarcasm, math, and language". Humor is not guaranteed. 172.69.69.207 17:58, 27 July 2022 (UTC)

Sure, but what percentage of Explainxkcd readers are here to figure out the joke, if any? 172.70.211.90 06:21, 28 July 2022 (UTC)

To help the editors who put in/dealt with the non-existent {{Citation actually needed}}. It's {{Actual citation needed}} that must have been intended... Although I don't think it's so much in doubt that it it is verifiable fact, anyway.^{[actual citation needed]} 172.69.79.171 22:23, 27 July 2022 (UTC)

Seems like a useful redirect or sub-transclusion. With modern camera high resolution, color fidelity, contrast, and depth of field, it's reasonable to ask whether making convincing fake photos is harder today than when cameras weren't as good and people weren't familiar with giveaway artifacts. 172.70.211.90 06:20, 28 July 2022 (UTC)

Everyone^{[citation needed]} is familiar with giveaway artefacts (such as "not-off-by-8" JPG copypasta, shadow hue discrepancies, general stippling mismatches between sources) and so they do very simple things to just not exhibit such things. Such as take/use the highest resolution images, edit them accordingly (with carefully chosen off-source imagery, chosen for similar/adjustable shadows/colour-spread/etc) and then save them ('new', with any EXIF intact/changed as necessary) at an acceptibly lower resolution so that any edge-case pixels they didn't quite get right get smoothed out amongst all the others.

Heck, I've even suppressed the saturation of a source image slightly to make up for the problems of a pasted-in element being too dull (not being actually artistic myself, I know I can't always as easily up-saturate a new bit to match the colour-warmth properly) in occasional paradoic pieces I've concocted via GIMP (I presume it matches Photoshop for natively available tools, and I rarely need to write my own scripted ones to achieve a blending filter of one sort or another).

Like I just hinted at, I consider myself to have zero artistic skill, and nor do I aim to actively deceive, but I have a decent eye for what looks wrong and I have more or less equal access to the tools that can find (e.g.) misaligned .jpg artefact-blocks so that I can avoid/squish those discrepancies. - Someone with actual profficiency in digital art and good mouse(/stylus) coordination can probably dodge/burn/smear or even pixel-by-pixel repaint segments of an image into a far more reasonable facsimile of the intended New Reality, and check that they've not left any clear digital fingerprints.

The Cottingly duo just tried to set up a real (but misleading) scene, pointed and shot and had no further recourse to go back into the developed image/negative to suppress any accidentally showing bits of the hairpins. And any actual attempts to do so were as likely to ruin more of the photographic plates than would have been improved, before they got their hand in (at great cost to unreusable materials), whilst I can try a small smudge and Undo it, with one of more Ctrl-Zs, if it actually does not improve any visual discrepancy...

Depending upon the actual intent, I can think of many ways that a committed hoaxer (who is familiar with hoax-busting details) can at the very least maintain plausible doubt over any idea of image-manipulation, at least enough to keep those who 'want to believe' unconvinced by any suggestions that something (other than the scene attempted to be depicted) is clearly technically wrong. 141.101.99.242 09:55, 28 July 2022 (UTC)

Still, presumably that is more effort and requires considerably more expertise than clipping cardboard cutouts to some scenery. Whereas with today's cameras the clips may be more visible, and the visual characteristics of the cutouts may be more prominent. 172.70.211.88 11:58, 28 July 2022 (UTC)

If you only want to clip cardboard cutouts to scenery, it doesn't matter whether your camera is digital or analogue (but the grain of any decent analogue film will hold more detail than any CCD/whatever, if you want to let someone to check the negative or memory card directly). When faking it from that stage on, the electronic revision of an image is trivial compared to convincingly messing about with plates/film/etc. 172.70.85.221 14:46, 28 July 2022 (UTC)

That depends on whether the cutouts are matte or gloss, and other things. 172.70.211.88 15:37, 28 July 2022 (UTC)

??? Shouldn't be different between analogue and digital 'raw' capture. Equally good/bad effect (or maybe harder to fake on film because of the intrinsically higher resolution that could show up giveaway defects in your props). That's not the medium, it's just the materials used.

Unless it's something like the old theory why vampires cannot be photographed (silver solution is mystically unable to record their image, like a silver mirror cannot reflect it (...with some hand-waviness about how it now shows whatever view ought to be obscured except for the vampiric nature of the person/clothing present...), but if you have a silver-free digital light sensor then there's no problem) matte or gloss or silk-finish have no issues. ...nah, that's too horror-geeky a caveat. Scratch that.

Oh, except that if there's an unwanted (or desired) glare driectly into the camera lense, with a non-reflex film camera it would be hard to determine exactly what the final image looks like (and always without definite exposure levels, etc), but a 'live' electronic viewfinder helps immensely (as with a reflex viewfinder, in some ways). Or at least you could do a playback of the last shot onto the almost ubiquitous LCD screen, showing you where you went right or wrong in all your chosen f-stop/shutter-speed/focus/etc menu-options. 172.70.91.128 16:26, 28 July 2022 (UTC)

Are they comparing the copper ingot complaint to a forgery or a fake review? If the actual customer was lying about his copper ingots, I don't see what that has to do with deepfakes and photoshop forgeries. It seems to me that they're suggesting that someone other than the customer forged the message to Ea-Nasir, posing as the customer. Elizium23 (talk) 02:21, 28 July 2022 (UTC)

The idea is that lies in text are analogous to fake photos and videos. It doesn't really work for "deep" fakes, because it doesn't take sophisticated neural network methods to write e.g. "Elizium23 said they eat worms." 172.70.211.90 06:20, 28 July 2022 (UTC)

Yes it does. Elizium23 (talk) 01:35, 30 July 2022 (UTC)

When I got to the part where it read "...we've been able to make text deepfakes for 5000 years.", I needed to stop and look up when the Bible was written. Darn, I thought I had them this time. These Are Not The Comments You Are Looking For (talk) 01:53, 31 July 2022 (UTC)