I might be far fetching here, but wouldn't this lead to people being more mindful of what they watch and interact with? I think all that it will take is a few "state of the art" deepfakes to cause a ruckus and the domino effect should do the rest.
Anyone in the field spent time thinking on this or has had similar notions?
I imagine that deepfakes will follow a similar path to edited photos -- lots of deception, followed by trustworthy sources gaining a little more cachet, but with many people still getting fleeced. Skepticism will ramp up in direct relation to youth, wealth, and tech-savvy.
Text, photoshop, special effects, deepfakes they're all just tools for spreading ideas, but we've been dealing (to some degree of success) with folks telling lies for as long as we've had language. I just can't see this fundamentally changing anything except the level of skepticism we give to video which (considering what hollywood has been capable of for some time) we should have been developing already.
Seems like there'd have to be some big caveats for that to be at all true, but it is interesting. I've read a lot of ridiculous crap and haven't been persuaded by much of it.
> (Do you believe me?)
No.
So when a global pandemic occurs and we're trying everything we can to isolate and socially distance, that priority changes real quick. People get talking and problems get solved.
Of course, sore losers will complain about anything to justify their loss, and this "new thing" was a prime scapegoat. It was also well known ahead of time that the mail in votes would be largely Democratic (because COVID was VERY politicized and democrats were more likely to follow quarantine guidance and therefore vote by mail). So when the votes came in, they pointed to that imbalance and called it "fraud".
Besides all that, there's no reason to be more suspicious of mail-in ballots than in-person ones. In-person, you mark a paper ballot and then put it in a stack... which then gets mailed somewhere else. If someone is going to be changing mail-in ballots, then they're already in a position to be changing regular ones as well (and every election security professional will tell you that paper ballots are more secure than electronic ones).
The one advantage of physical voting I can think of is the ability to just be close to voting station on voting day, counting people who go in there, asking people (who are willing to share) for whom did they vote. This allows to independently check if fraud exists.
Why? Mail-in voting is hardly unique to the US; what made you suspicious?
(A few people have been charged with voting their spouse's ballots.)
Unfortunately in 2020 the fringe became the GOP mainstream, treating equally soft claims as fact.
"Seventy-two percent (72%) of Democrats believe it’s likely the 2016 election outcome was changed by Russian interference, but that opinion is shared by only 30% of Republicans and 39% of voters not affiliated with either major party."
https://www.rasmussenreports.com/public_content/politics/gen...
Russia absolutely did and continues to push propaganda into elections in the USA and elsewhere. That's not really in dispute at this point so I'm not surprised it polls that high.
Got a poll that shows similar numbers for fraud? I would be genuinely surprised to see that.
I should actually note here that I didn't vote for Trump, either time, nor did I vote for Clinton or Biden.
I just hate hypocrisy.
That sounds like a perfectly reasonable claim with evidence that supports it, paralleled by other elections in other countries as well; quite obviously very different to what was discussed above.
https://www.washingtonpost.com/politics/hillary-clinton-trum...
https://news.yahoo.com/hillary-clinton-maintains-2016-electi...
In 2020, Hillary Clinton was still casting aspersions regarding the outcome of the 2016 election, sowing discontent about the electoral college, preparing Democrat voters to ignore the results until Joe Biden was declared the winner.
Portraying this game as if it's only being played by one team does not help restore any trust in the federal election process.
The bigger problem in that election was Russian-intelligence-stolen (and possibly tampered with) documents being released to the press in the lead up to the election in coordination with the Trump campaign (with the FBI keeping its investigation of that secret), and then the FBI director making an unprecedented and (we found out only afterward) unsupportable statement attacking Clinton immediately before the election, after being pressured into it by a handful rogue FBI agents who were friends of Trump’s campaign threatening insubordination.
And perhaps the biggest problem of all, an entirely too credulous mainstream media who didn’t put those developments in context, leaving voters to draw mistaken inferences, and giving oodles of free airtime to Trump’s rallies without making any effort to dispute outright lying in real time.
But she doesn't count as "anyone", I guess?
In other words: not saying that there was actual fraud sufficient to change the election, not saying the election was "stolen" in the sense people seem to be saying here.
Stolen by foreign influence is very different than what the 2020 nuts have claimed. How many court cases did they lose? How many times were they asked to produce evidence and came up with ... nothing?
But we do.
There's research. Even if you read something that you know is wrong you still believe it. Especially when distracted or not taking the time to analyze. As we rarely do.
https://techcrunch.com/2013/01/24/study-finds-that-we-still-...
https://www.businessinsider.com/why-you-believe-everything-y...
That seems like bullshit to me. I read your words (you even posted links!) so how come I don't just instantly believe you? If it were true, wouldn't it make all fiction inherently dangerous?
Let's see how it holds up in real life... here's a lie: "My uncle works at Nintendo and he told me that Mario (Jumpman at the time) was originally intended to only have one testicle, but the NES (famicom) didn't have powerful enough graphics to show that so they scraped that part of his official character design and have left the number of testicles unspecified ever since."
Somewhere, secretly deep inside you, do you believe that now?
Nah. I think we don't have to worry about people believing everything just because they read it. Reading things can put ideas into your head (have you ever even considered Mario's testicles before today?) but at this point we're straining the hell out of "belief" and going into philosophical arguments. In real life though, we are capable as a species of separating fact from fiction some of the time.
It works because people don't judge truth based on sober, rational, purely logical analysis but emotion and bias and most importantly comfort. If you're in an environment in which everyone believes X, you will inevitably begin to conform if the social pressures to do so are strong enough, and counter-signals weak enough. This is how radicalization works, through the gradual osmosis of a worldview, and the acceptance of smaller lies that lead to accepting bigger lies. It's how Nazi propaganda worked, and it's how modern advertising and politics work. It's why witness testimony is unreliable and how police can convince suspects that they committed a crime they went in knowing they were innocent of.
The effect isn't universal - nothing about human psychology is. But it is real.
Repetition isn't enough, but I'll accept that with enough effort from enough people you might eventually be gaslit or brainwashed into believing just about anything so long as it didn't violate some fundamental aspect of your identity in which case it's more likely you'll just lie about believing it to make your life easier.
By '06 I had an MBA with a Masters Thesis on the creation of a new Advertising format where the viewer, their family and friends are inserted into brand advertising for online advertising. By '08 I had global patents and an operating demonstration VFX pipeline specific for actor replacements at scale. However, it was the financial crisis of '08 and nobody in the general public had ever conceived of automated actor replacements. This was 5-7 years before the term deep fake became known. VCs simply disbelieved the technology was possible, even when demonstrated before their eyes.
Going the angel investor route, 3 different times I formed an investor pool only to have at some point them realize what the technology could do with pornography, and then the investors insist the company pursue porn. However, we had Academy Award winning people in the company, why would they do porn? We refused and that was the end of those investors. With an agency for full motion video actor replacement advertising not getting financing, the award winning VFX people left and the company pivoted to the games industry - making realistic 3D avatars of game players. That effort was fully built out by '10, but the global patents were expensive to maintain and the games industry producers and studios I met simply wanted the service for free. Struggled for a few years. We closed, sold the patents, and I went into facial recognition.
https://patents.justia.com/inventor/blake-senftner https://www.youtube.com/watch?v=lELORWgaudU
I was bitter about this all for a long time.
I'm not sure there's a huge market for people wanting to insert their friends and family into porn, but if there was why not just try it? Seems like it'd have demonstrated the tech worked commercially which could have attracted investment in non-pron uses and it could have ended up just one more technology in a long list of tech made successful as a result of porn.
Meanwhile, simply the process of inserting consumers in the trailer of any Marvel superhero film and charging $1 for you own copy would make tens of millions. Repeat with any highly desirable fantasy franchise. That is the most obvious application. Actor replacement with ordinary people has a huge number of positive to society applications. Porn is not one of them.
I think for claims that you think are important to determine an objective truth value for (like who the President of the U.S. is), your determinism mechanism is based on trusting sources you deem reliable and looking for broad agreement among many sources you deem to be independent. You're probably not just looking at a single sourceless video of Ronald Reagan behaving as if he's the president and believing that claim because the video couldn't possibly have been faked.
And for other claim that you don't think are important to determine an objective truth value for, I don't think you need very high-fidelity evidence anyway. For example, people have no trouble believing claims that corroborate their closely-held ideologies even with very low-fidelity fraudulent evidence, or even claims made with no attempt whatsoever to provide even fraudulent evidence!
A lot of people point to Photoshop not breaking truth, but in my experience, we simply rely on video instead. When I served on a grand jury, essentially all of the non-testimony evidence I saw was video, and it was incredibly compelling. Neutering the reliability of that evidence will hurt.
I don't think people will be more mindful of what they watch and believe, I think the opposite will happen: an attraction to fake content. People will embrace the fantasy and share deepfakes at a scale so large governments will be running campaigns to alert the public that such-and-such video is fake, possibly attempting to regulate how content shared online must be labeled.
That said I still believe when these lies are closer to us, enough for us to care either as professionals or friends and family, that we will be more discerning about reality.
There was a demo earlier this year (Jan) showcasing the proposed 1.0 spec working in Microsoft Edge: https://c2pa.org/jan-2022_event/
Just like it was for millions of years before now.
Face-to-face is only applicable with you small social network.
That is my point precisely. And I will point out here that small social networks are our historical environment. I believe not only can we return to them and flourish, but also that it would be a great boon to human flourishing, the only outcome that really matters.
I've worked at home for nearly 20 years so I've had to learn to create a strong in-person social network. The pandemic and some medical issues have interfered, but I still much more enjoy having fun with my friends than doing things online.
No. We have already run the case study where people on Reddit, Twitter, and other social media will seethe at mere screenshots of headlines and captions under a picture with zero need for verification.
Here on HN we will pile into the comments to react to the title without even clicking the link to read it ourselves.
Deepfakes feel like a drop in the bucket. What does it matter that you can deepfake a president when people will simply believe a claim about the president than spreads around social media? I don’t see it.
It’s a surveillance thriller.
Let me quote myself from a discussion I was having this morning with a friend who is a tenured professor of philosophy working on AI (as an ethics specialist his work is in oversight),
we were discussing the work shared on HN this week showing a proof-of-concept of Stable Diffusion as better at image "compression" than existing webstandards.
I was very provoked by commentary here about the high "quality" images produced, it was clear that they could in theory contain arbitrary levels of detail—but detail that was confabulated, not encoded in any sense except diffusely in the model training set.
"I'm definitely inclined to push hard on the confabulation vs compression distinction, and by extension the ramifications.
I see there a very meaningful qualitative distinction [between state of the art "compression" techniques, and confabulation by ML] and, an instrumental consequence which has a long shadow.
The thing I am focused on being, whether or not the fact that a media object is lossy or not can be determined, even under even forensic scrutiny.
There was a story I saw this week about the arms race in detection of 'deep fake' reproduction of voice... which now requires some pretty sophisticated models itself. Naturally I think this is an arms race in which the cost of detection is going to rapidly become infeasible except to the NSA. And maybe ultimately, infeasible full stop.
So yeah, I think we're at a phase change already, which absolutely has been approaching, back to Soviet photo retouching and before, forgery and spycraft since forever... so many examples e.g. the story that went around a couple years ago about historians being up in arms about the fad for "restoring" and upscaling antique film and photographs, the issue of concern being that so much of that kind of restoration is confabulation and the presumptive dangers of mistaking compelling restoration for truth in some critical detail. Which at the time mostly seemed a concern for people who use the word hermeneutics unironically...
...but we now reach a critical inflection point where society as a whole integrates the notion that no media object no matter "convincing" can be trusted,
and the consequent really hard problems about how we find consensus, and how we defend ourselves against bad actors who actively seek their Orbis Tertius Christofascist kingdom of rewritten history and alternative facts.
The derisive "fake news" married to indetectably confabulated media is a really potent admixture!"
If I want to compress the movie Thunderball -- a sufficiently clever "compression" algorithm could start with the synopsis at https://en.wikipedia.org/wiki/Thunderball_(film) add in some images of Sean Connery, and generate the film. That's...maybe a 100K to 1 compression ratio?
If the algorithm itself understands "Sean Connery" then you could (theoretically) literally feed in the text description and achieve a reasonable result. I've seen Thunderball, but it was years ago and I don't remember the plot (boats?). I'd know the result was different, but I likely wouldn't be able to point to anything specific.
put it this way:
written text has always been extremely mutable and therefore falsifiable. video is simply going to become more like text. people will trust it based on the context and their own judgment rather than the content. I suspect most people already do this anyway
I assume that full-on impersonation will still be illegal, but certain looks that are sometimes quite similar to a real celebrity will trend now and then.
The context for this is the continual improvement in the capabilities and comfort of VR/AR devices. The biggest one I think is going to be lightweight goggles and eventually glasses. But also the ability to stream realistic 3d scenes and people using AI compression (including erasing the goggles or glasses if desired) could make the concept of going to a physical place for an event or even looking exactly like yourself feel somewhat quaint.
I formally built an ad agency with a feature film VFX pipeline for actor replacement advertising - but I built it back in '08, years before "deep fakes" and nobody believed what I had working was possible.
It's less about using fakes to push your agenda, and more about being able to (plausibly or implausibly it doesn't matter) claim that whatever video is a deepfake.
The truth is meaningless, and as tools like deepfakes become more and more sophisticated, it's harder and harder to establish baseline realities.
And someone is benefiting from that shift away from reality, I just don't know who.
People made the same arguments about photoshop, but it's really not a problem. Almost never is a single video the only evidence of anything and in the cases where it is and that video can't be verified it's probably best not to ruin someone's life over it.
Skepticism in general will only be applied to people we don't like and ignored for people we do.
The continued lapping up of blatant Ukrainian propaganda in mains stream media for example doesn't even need photoshop to be believed, just the vague 'sources said'.
Have you been paying any attention to what's going on the last several years?