From a skim: while I agree with surface level details about deepfakes (i.e. they're mostly used for funny memes, humans are mostly capable of detecting them), I am still slightly perturbed by people using it to generate NSFW porn of people (particularly underaged ones) without their consent [0].
Very much a "four horsemen of infopocalypse" statement, that'll I'll concede.
What really bothers me is when people here reply to stories like this and say things like "well those girls shouldn't feel bad at all, it's just nudity, our society is so prude and _that's_ the _real_ problem." I don't know what the solution is, I'm not suggesting anything draconian - I just really don't like the idea that we should tell these girls to just get over it.
>"There was huge concern right away that you’d have videos appear in a political context that could change the course of an election or lead to political violence. But none of that has transpired."
This kind of "it hasn't happened yet so it won't" dismissal of the ever-closer ability to create a perfect fake video of X person saying this or that is so myopic I feel like it doesn't even deserve elaboration as to why it is wrong, we should skip straight to mockery and derision. We've already had multiple cases of fake Biden and fake Trump videos posted to the internet with hundreds of thousands of views before any kind of clarification came to indicate they were fake.[1] 2024 will put this to the test, and I have little confidence that this will remain the case.
Well, yes. I saw Biden recite a copipasta from beginning to end. I still marvel at how well it's done. The skill of the AI or whatever, or the person who did it, approaches art. I still don't believe for a second that it's real, because it's a copypasta, you know? And especially because I am a human being with normal intelligence.
Ok that's wonderful but a bit of a red herring. I'm not worried about the AI presidents gaming channels that are out there with Trump Obama and Biden playing Minecraft together, I'm worried about "new leaked audio from our undercover reporter of Biden telling the head of the FBI to sink Trump's campaign" or "Video leak of Trump calling for the intimidation of voters in blue states" exploding on the internet with no real way to confirm or deny it. You may say you're a human being with normal intelligence, but I don't have a ton of faith that people would be able to tell one way or the other - especially younger people who seem much more likely to believe things on the internet. We already have cybercriminals using AI voicechangers to impersonate people's coworkers to game helpdesks for breaking into systems.[1] This is a tool that can be used for fun but it's as game changing as the printing press, possible more.
You are right not to rely on people's intelligence, but we are obliged to do so because we are not monarchists and we believe in democracy, or at least we try to.
I don't know for sure. And certainly we are not what you call "leftists" in HN. But let me say this on a personal level. I know that people sometimes, all the time, often, certainly, behave stupidly. But I want to believe, ergo I believe, that people can make good decisions, that there is such a thing as "common sense". Looking at the facts sometimes made me sad. Maybe people can't think for themselves, and then they can't choose. But I prefer not to think in terms of peasants and elites, even if history tells me so and Ortega y Gasset tells me how it works. I want to believe that we are capable of governing our lives. I want to believe. Forgive me for expressing not a fact, but the way I think about democracy.
At the same time, you have to consider that if the same crooked reporter reported these leaks without providing any audio at all, or said they were emails, or texts, a large number of people would believe them anyway. You have to check what the delta is of people who would not have believed a plain text lie but would believe a video lie, and as deepfakes become more and more sophisticated in plain sight, that delta is bound to shrink.
I feel that ultimately the delta will become negligible, because the cognitive flaws that would push us to believe deepfakes at an era when anyone can create a deepfake of themselves in 30 seconds on a website (not now, but not long after they would be used for propaganda) are the same that currently push us to believe clear textual propaganda. If you don't fall for one, you won't fall for the other.
Doesn't even need a crooked reporter, just a troll/extremist/propoganda creator and a news organisation without much interest in fact checking/research.
In fact, forget deepfakes and disinformation campaigns. What should truly terrify people is how low the standards are for sourcing information in many media publications now. Oh sure, you could use AI to fake a video of Biden or Trump saying something controversial and send that in... or you could print out a document titled 'Trump's Super Secret Plan for World Domination!', take a photo of it on a mobile phone and send that into many outlets for much the same effect.
Just like the first time I saw the old lady on the surf table standing up in the sky to welcome me to the Internet and offer herself as my guide, I knew it couldn't be real and suspected that there was some irony in the whole wonderful scene.
I think that will be the issue, where even though an image or video will be revealed as fake, how many people saw it before that was revealed (and of the people that saw it and believed it, did they hear of the fakery being revealed)?
> People believe that stuff far more easily than they believe something they read.
I'm not convinced it's "far" more. People believe things they read extremely easily when they say what they want to hear. Propaganda is already efficient enough that there is little room for improvement. As for people who question what they read, they will also question what the see once they become aware of the power of deepfakes.
That's if they even watch the videos. I generally don't, because it's a waste of time when trustworthy people can watch them for me and tell me what's in them.
Ideology is believing things a priori, sometimes despite any available evidence. There will always be ideology and ideologues, but it is sort of an orthogonal discussion to TFA. Ideologues simply don't care about the evidence, deepfaked or not, otherwise they wouldn't be ideologues.
This is a very binary view. Every person is to a degree an ideologue, and to a degree believes evidence. Shifting the scale from "videographic evidence is pretty trustworthy" to "any video can be a fake and you have no chance to detect a difference" will make a big dent.
Stop worrying about deepfakes, as in "we can't do anything about them and have to resign to have unrealiable media evidence from now and till forever.". Don't worry about that, right?
Also this professor's methodology is laughable. He is surveying popular deepfakes on the popular sites. OF COURSE they will be jokes. Who will share and view a million times a deepfake of some CCTV with a face of some robber substituted for some other equally unknown guy. Or a deepfake of a car crash with car plates swapped for different ones. Etc.
"We were worried that they will affect elections or politicians, but none of that has transpired." Says who? Who did this research? How was this determined? By a gut feeling while watching deepfake memes? Thought so.
Read history. This has come up countless times. The world did not collapse. Some deepfakery has even been addressed by US courts (Hustler v. Falwell, a literary deepfake). There was the war of the world fiasco, a radio deepfake.
Before that people panicked over the ability to edit photographic plates, combining images to make people appear/disappear. Society did not crumble. New technology comes along. People fear that it can be used to create fake news. Then, a few years later, everyone chuckles about it. A few years from now we will have adapted to the current latest deepfake tech and will also chuckle about the panic.
Current technology is so much better in its quality and ease of use as to be barely comparable to what came before. A lot of people in this thread seem to pride themselves on being able to spot deepfakes (how do they know the ones they don't spot?), but no person or machine will be able to tell real from deepfake in 5 years. I'm not sure what the effects will be of not being able to trust any photo or video (e.g. surveillance video, drone footage, etc.)
Information has been faked, to varying degrees of success, for as long as humanity has engaged in intrigue. "AI" does not change this, it's still ill-willed humans and trolls at the wheel faking information.
I always roll my eyes at fears of "AI" faking this, creating that, etc.
Whatever "AI" has done or can do, we've already seen much worse from humanity. Been there, done that.
True, AI is just the tool and not the perpetrator. Murder has also exist since the dawn of humanity, but a murderer with a stick is very different from one with a fully automatic rifle.
While I agree with you, I think the issue is that the tooling is starting to become accessible to anyone, which is going to create a flood of deepfake content. In the past, you usually needed to have a specific set of skills to fake information, and even then experts were able to analyze things like images for distortions or slight alterations that couldn't happen in the natural world. As the technology gets better, and into the hands of people that wouldn't have had access to it previously, only then will we really see if this is going to have any kind of effect.
I don’t know what will happen. Possibly deepfakes won’t be a big issue. But I’d still put myself strongly in the “this time is different” camp.
“Stalin could remove a guy from
a photo”. A specialized state intelligence group. “Any photoshop user can do that”. A tiny percent of adults who use photoshop. Now: billions of people can make perfect (soon) fake video, audio, etc and distribute to the world.
I can’t predict the future, but to be confident that this won’t sow chaos seems silly.
> There was huge concern right away that you’d have videos appear in a political context that could change the course of an election or lead to political violence. But none of that has transpired.
Citation needed for this. Margin for victories has been so small in recent victories that it is very plausible that enough voters have been influenced by fake news accompanied by fabricated/deepfaked images to change outcomes.
Exactly. And it's not a single fake video of Biden promising to jail Trump supporters, which could be shown to be fake. It's dozens of little videos shared on Facebook where politicians are made to look stupider, where their speech is slowed and slurred (they've done this to Biden, Harris and Trump), where false audio is added.
We can't accurately track the effects of such videos.
Yup, and I'd expect that for an election year there will be a lot more. Even if none of it is directly sponsored by a candidate, supporters will likely create deepfake content, and it will be believable to some percentage of people.
Plenty of dimwits have been scammed by deepfakes on places like facebook.
I work for a business man that is suing meta for this reason. Obviously meta dont produce the deepfakes but they are happy to take advertising money to allow the deepfake adverts to play on their portal.
> There was huge concern right away that you’d have videos appear in a political context that could change the course of an election or lead to political violence. But none of that has transpired.
On the other hand, there's no evidence this stuff isn't influencing elections either, is there? It's happening, and people are believing it, so I'd say it's almost self-evident that it's influencing elections.
People usually expect politics to have decisive, iconic moments like Napoleonic battles. And possibly that was true, back in Napoleonic times. But modern politics is much more like modern warfare, which is nuanced and messy and complex. We don't and can't currently track the factors that lead to someone casting their single vote in a particular way (or at all), but we still need to recognize that they are myriad and complex reasons (in the case of someone who can be swayed... many of us are going to vote one way or another virtually regardless of what happens).
The original assertion was about "none of [videos in a political context that could change the course of an election or lead to political violence] having transpired". Of course a single example of a deepfake appearing in a political context is evidence of the contrary.
As formulated, it isn't necessary for either election outcomes or political violence to be attributed to said fake videos, and I would say that demanding that kind of evidence is unreasonable (it's a common denialist tactic to require an absurd level of evidence). So why exactly are those not evidence of the contrary according to you?
The true reason to not worry about deepfakes is that both sides of any conflict will learn to use them equally viciously. Like always, the only way out is to take a step back, study things objectively through the lens of economics, and figure out the tradeoffs inherent to the choice.
Others have pointed this out but it’s possible (or likely) that the biggest harm is just the diluting of the truth so no one believes anything.
When someone sees or hears something they don’t want to see or hear (against their political outlook), they will now have plausible deniability that it’s just bogus evidence and will dismiss it.
1. Don't believe anything you see or read on the internet.
2. Don't post any personally identifiable information about yourself on the internet.
3. Every woman is a man, every teenager is an FBI agent, and every man is a dog.
Every single problem or controversy surrounding the internet derives from breaking one of these three rules. If we just started teaching kids these again we wouldn't have the problems that we do.
It's also worth noting that there far stupider methods that can (right now) generate much more convincing images than deepfakes. The simplest is taking an existing, real picture and taking it out of context, lying about what the image shows, etc. This is actively being done in current conflicts, e.g. taking real footage from the Syrian civil war or from previous wars in Gaza and presenting them as photos from the current war. It works so well because there are enough absolutely horrible photos from the current war, so the fake photo appears plausible.
If AI-generated imagery should ever reach the level of being completely artefact-free and absolutely indistinguishable from real photos, then deepfakes will be about on par with the propaganda effect of today's real photos.
Agreed. Most of the propaganda videos I see around are more like things taken out of context. When you see a battle or atrocity video, it's hard to be sure where it was actually filmed, when it was actually filmed, who each side actually is, what happened before and after the camera rolled, what is happening outside the camera frame, etc. It seems everyone is doing this on both sides of the Russia-Ukraine war and Israel-Gaza war.
Do we even need deepfakes when there's a practically unlimited amount of such video that can easily be cut deceptively, taken out of context, and misattributed to show anything you want?
48 comments
[ 3.0 ms ] story [ 63.4 ms ] threadVery much a "four horsemen of infopocalypse" statement, that'll I'll concede.
[0]: https://english.elpais.com/international/2023-09-18/in-spain...
This kind of "it hasn't happened yet so it won't" dismissal of the ever-closer ability to create a perfect fake video of X person saying this or that is so myopic I feel like it doesn't even deserve elaboration as to why it is wrong, we should skip straight to mockery and derision. We've already had multiple cases of fake Biden and fake Trump videos posted to the internet with hundreds of thousands of views before any kind of clarification came to indicate they were fake.[1] 2024 will put this to the test, and I have little confidence that this will remain the case.
[1] https://www.bbc.com/news/62338593
https://www.pcmag.com/news/hacker-deepfakes-employees-voice-...
I feel that ultimately the delta will become negligible, because the cognitive flaws that would push us to believe deepfakes at an era when anyone can create a deepfake of themselves in 30 seconds on a website (not now, but not long after they would be used for propaganda) are the same that currently push us to believe clear textual propaganda. If you don't fall for one, you won't fall for the other.
In fact, forget deepfakes and disinformation campaigns. What should truly terrify people is how low the standards are for sourcing information in many media publications now. Oh sure, you could use AI to fake a video of Biden or Trump saying something controversial and send that in... or you could print out a document titled 'Trump's Super Secret Plan for World Domination!', take a photo of it on a mobile phone and send that into many outlets for much the same effect.
Videos are three-sense. People believe that stuff far more easily than they believe something they read.
And couple it with the fact that people want to believe the propaganda and you have a massive political lever.
So I'm only not worried insofar as I'm not worried that politicians and superPACs will not utilize a massive political lever.
I'm not convinced it's "far" more. People believe things they read extremely easily when they say what they want to hear. Propaganda is already efficient enough that there is little room for improvement. As for people who question what they read, they will also question what the see once they become aware of the power of deepfakes.
That's if they even watch the videos. I generally don't, because it's a waste of time when trustworthy people can watch them for me and tell me what's in them.
There's already this fronge group claiming that war in Ukraine is fake, but I expect such things to move more mainstream with generative AI.
Also this professor's methodology is laughable. He is surveying popular deepfakes on the popular sites. OF COURSE they will be jokes. Who will share and view a million times a deepfake of some CCTV with a face of some robber substituted for some other equally unknown guy. Or a deepfake of a car crash with car plates swapped for different ones. Etc.
"We were worried that they will affect elections or politicians, but none of that has transpired." Says who? Who did this research? How was this determined? By a gut feeling while watching deepfake memes? Thought so.
I always roll my eyes at fears of "AI" faking this, creating that, etc.
Whatever "AI" has done or can do, we've already seen much worse from humanity. Been there, done that.
“Stalin could remove a guy from a photo”. A specialized state intelligence group. “Any photoshop user can do that”. A tiny percent of adults who use photoshop. Now: billions of people can make perfect (soon) fake video, audio, etc and distribute to the world.
I can’t predict the future, but to be confident that this won’t sow chaos seems silly.
This is an extremely poor metric. (The worst possible metric?)
Did bad things happen due to deception? Yes, actually, they did. Read history!
Citation needed for this. Margin for victories has been so small in recent victories that it is very plausible that enough voters have been influenced by fake news accompanied by fabricated/deepfaked images to change outcomes.
We can't accurately track the effects of such videos.
I thought nautil.us was a somewhat reputable publication.
I work for a business man that is suing meta for this reason. Obviously meta dont produce the deepfakes but they are happy to take advertising money to allow the deepfake adverts to play on their portal.
This is wildly wrong.
https://web.archive.org/web/20190415011409/https://www.washi...
https://www.foxnews.com/politics/twitter-erupts-trump-trolls...
https://www.politifact.com/factchecks/2023/may/05/facebook-p...
https://theconversation.com/faked-videos-shore-up-false-beli...
People usually expect politics to have decisive, iconic moments like Napoleonic battles. And possibly that was true, back in Napoleonic times. But modern politics is much more like modern warfare, which is nuanced and messy and complex. We don't and can't currently track the factors that lead to someone casting their single vote in a particular way (or at all), but we still need to recognize that they are myriad and complex reasons (in the case of someone who can be swayed... many of us are going to vote one way or another virtually regardless of what happens).
As formulated, it isn't necessary for either election outcomes or political violence to be attributed to said fake videos, and I would say that demanding that kind of evidence is unreasonable (it's a common denialist tactic to require an absurd level of evidence). So why exactly are those not evidence of the contrary according to you?
See also: Michael Huemer's "In Praise of Passivity", https://fakenous.substack.com/p/in-praise-of-passivity .
When someone sees or hears something they don’t want to see or hear (against their political outlook), they will now have plausible deniability that it’s just bogus evidence and will dismiss it.
1. Don't believe anything you see or read on the internet.
2. Don't post any personally identifiable information about yourself on the internet.
3. Every woman is a man, every teenager is an FBI agent, and every man is a dog.
Every single problem or controversy surrounding the internet derives from breaking one of these three rules. If we just started teaching kids these again we wouldn't have the problems that we do.
If AI-generated imagery should ever reach the level of being completely artefact-free and absolutely indistinguishable from real photos, then deepfakes will be about on par with the propaganda effect of today's real photos.
Do we even need deepfakes when there's a practically unlimited amount of such video that can easily be cut deceptively, taken out of context, and misattributed to show anything you want?
https://medium.com/readyai-org/the-dawn-of-a-new-era-artific...