Audio and video evidence isn't admissible because it's audio or video (and may be inadmissible nonetheless). It's admissible because someone testifies under oath that they have personal knowledge of its provenance. The burden is on the party introducing the evidence to show that it's reliable, and the question of whether or not it is indeed reliable is a factual one for a judge or jury to answer. It's not assumed to be "true" or to accurately reflect reality just because it's a purported photo, video, or audio recording.
Does it though?
Suppose there is a theft at a company. The police go to the company's security team and get the surveillance footage. Presumably admissible.
If it was an inside job, the surveillance footage could be a deepfake showing someone else committing the crime. Or maybe it's real surveillance footage. Without some way to distinguish the two, how do you know?
To rip from current headlines: https://www.justice.gov/usao-wdpa/pr/erie-man-charged-arson-...
(Leaving aside the merits of the case / opinions, and just using it as an example)
".. the [Facebook Live & coffee shop] videos depict a male – with distinctive hair to the middle of his back wearing a white mask, white shirt, light blue jean jacket, black pants with a red and white striped pattern down the side and red shoes - setting a fire inside of Ember + Forge."
"A review of additional Facebook public video footage from the area of State Street near City Hall in Erie on the evening of May 30, 2020, shows the same individual without the mask but wearing identical clothing and shoes. The subject’s face is fully visible in this video footage."
And that's a federal arson charge (min 5 years, max 20 years prison).
Ultimately you end up back at the beginning: holding people, and not pieces of information, accountable for their actions. That's an issue with civil society, not social media.
Plus, deepfakes have the potential to create much more noxious videos than real life. Real life videos depict real life people, flaws and complexities all included; deepfakes will be constructed to depict representations of the target that're most likely to generate inchoate rage of the mob. It's rare that the former happens to be exactly the latter.
Why not blame the person who confirmed your husband's death at this point if accepting comfortable illusions of someone easy to punish is what we are doing? Without them you would still have some remote hope of survival!
The first step of solving a problem is recognizing the actual problem - what we find comforting is only a distraction for rationalization purposes. Better to recognize that microscopic organisms caused the crop failure instead of burning an old woman as a witch.
Politicians don't have a particular right to remain in office despite inciting violence, even if it gives them the sads if they face repercussions.
If the video weren't fake, would that somehow make the murder of Rohingya acceptable?
Just one example: A radio station (RTLM) in Rwanda laid the groundwork of genocide by dehumanizing Tutsis, among other things by referring to them as "cockroaches" repeatedly. The station was deemed instrumental in the resulting murders of at least half a million people.
It's not fear mongering to look at the facts of history and see that propaganda is a primary way that violence scales. Deep fakes are another tool which will certainly be used towards these same ends. What propagandist would not want to use such a tool?
Propaganda is a primary way that human action and organization, whether violent or not, for good or ill, scales.
By being so open about it Stone and Parker will just increase awareness of how easy it is to do, so people will know how realistic these things can look and not blindly believe what they see.
So, again, what these guys are doing is a true public service.
I think the first question asked of a video is going to be "is this a deep fake?" just like we currently ask "is this picture photoshopped?".
It doesn't mean that we no longer use pictures, or that we don't alter pictures, just that we are more critical of them.
That said, there is just more realism in a series of moving pictures, but I don't see why the situation for a series of fake pictures has to be wildly different from the situation of a single fake picture.
One thing that worries me is that everything is becoming immutable, saved forever, and digitally signed. Hard to claim you didn't post something if your account is super secure. You can't go anywhere without cameras taking your picture. And if someone steals your bitcoin or your smart contract has a mistake, your money is gone. You can't argue with an algorithm. And the internet never forgets. (Well, unless it is inconvenient for you, then Murphy's law applies ;-))
There are many such phenomena but I'd say they are all related: They mean you have less wiggle room for mistakes, or social deviance. If you are in a situation where you have to break a law, or if you are just having an affair, chances are someone or something is going to see you. There is no anonymity in the crowd anymore, on the contrary.
But this deep fake technology, if it really evolves to be undetectable, can be liberating, if it erodes the trust in pictures. At least the social control mechanisms based on cameras are going to stop working.
We were living in a very specific phase of human history, where we learned how to produce pictures from real scenes, but haven't learned yet to easily fake them. I just think this is going to come to an end, and we'll have to adapt. (I hope we'll adapt socially, and not just by cramming DRM into our technology like we tend do to; but that is a different topic.)
You just described a GAN (generative adversarial network)!
>The model for the discriminator is usually more complex than the generator (more filters and more layers) and a good discriminator gives quality information. In many GAN applications, we may run into bottlenecks where increasing generator capacity shows no quality improvement. Until we identify the bottlenecks and resolve them, increasing generator capacity does not seem to be a priority for many partitioners. [1]
Put another way: GAN training ends when the discriminator can no longer meaningfully distinguish real from fake. By definition then, the best generator will have no useful discriminator that can distinguish its output from real data. (conversely, if you did have such a discriminator, you could use it to train a better generator)
[0] https://developers.google.com/machine-learning/gan/generator
[1] https://towardsdatascience.com/gan-ways-to-improve-gan-perfo...
How far ahead do they need to be?
Suppose that it's cat and mouse, at least initially. Every six months someone comes up with a new way to detect the best known deepfakes, then six months after that there is a new way to evade that means of detection as well.
Someone drops a deepfake five weeks before an election.
You cannot underestimate the global political ramifications of this. If you think the amount of video manipulation was bad enough this cycle in the 2020 presidential election? Imagine a few years from now when video's are being put out by political operatives and people with nefarious agenda's with impunity all over social media.
This sort of thing absolutely terrifies me since you can start to twist reality to whatever you want and influence people in ways we never thought possible. I really feel like the genie is out of the bottle and this has the potential to become a very dangerous tool for people with bad intentions.
You are assuming people will have never heard of these deepfakes. Sure there are still a few grannies today who have never heard of photoshop. But your average twitter user?
Every piece of controversial video content will have a group loudly screaming "TRUTH!" at the same time as another group screaming "FAKE!", and it will just amplify the social polarization we're already experiencing.
A weird thing about this thread, and others like it, is that there seems to be this broadly shared implicit premise within the thread context that the feared ill effects of new technologies like this have not already been prevalent in our societies and information ecosystems for decades. In a thread about bias, fake news, propaganda, etc, people seem to have no problem realizing and acknowledging that we already have a very serious problem (often only visible in one's personal outgroup, but that's better than nothing) - but when the specific topic of conversation is a new technology, the majority of the comments seem to be written as if we don't really have any significant issues currently. It seems as if there's some sort of a phenomenon whereby the logical methodology for evaluation of the situation changes according to the topic, as opposed to there being a consistent methodology that at all times has an explicit awareness of the ever-present bigger picture.
Here's [1] an 8 minute video on Presidential debates. This fairly well demonstrates how this aspect of our political system is largely pure theatre...and yet, intelligent people often speak (again, depending on the specific(!) topic of discussion) as if this charade is highly legitimate process, within a larger political (and journalistic) process that is also highly legitimate.
The way I view the ecosystem is that the vast majority of things are to a very large extent ~fake (in whole or in part). Cranking up the absurdity to 11 in classic South Park style, making a complete mockery of both the politicians as well as those who can't consistently(!) conceptualize the true nature of our system, seems like an excellent response to a situation that has been sorely in need of some good old fashioned satirical mocking for decades. Western society & politics lost the right to be taken seriously ages ago - admitting to ourselves that there's a problem seems to me like a prerequisite first step in fixing it.
[1] Winning the Presidency: Debating
ex:
Trump access hollywood tape -> fake Soros conspiracy to save the pedophiles
I believe the optimum approach is to be concerned with all risks, and weigh the magnitude of each in a state of careful self-monitoring of one's potential biases (and ideally, have your conclusions reviewed by others, preferably from a diversity of ideologies and perspectives in an attempt to minimize the well known affects of groupthink). Noteworthy to me is that a significant number of people (if not the majority, depending on which community you are in) are easily able to see the epistemic errors in their outgroups thinking, but has more difficulty in doing the same within their ingroup.
For example, in your comment it seems that you have noticed shortcomings when it comes to politicians of one general ideology, but I wonder if you are of the belief that this phenomenon does not occur across all ideologies?
Well, something really bad is definitely ahead with fake videos (as if real videos, with significantly meaning-changing omissions made with the plain old cutting process weren't bad enough) and people with high visibility creating awareness by doing it in the open is the closest thing to a defense that we have. It's a hopelessly weak defense but better than nothing.
https://thehill.com/homenews/media/523045-cable-news-october...
Tucker Carlson spent the month of October pushing Russian disinfo on Hunter Biden, all for it to culminate in "losing" his documents, re-finding them and then suddenly, inexplicably backing down.
https://www.thedailybeast.com/tucker-carlson-suddenly-says-i...
That's good. Courts should not have high confidence in any single bit of evidence. It can all be manufactured. And long before now. It's just (too) hard to manufacture, not be detected, and get larger number of people required to "go along" with it, when you have manufacture multiple collaborating pieces.
Good. We should dispense with retributive justice, and replace it with restorative, transformative systems which are a noop on the wrongly accused.
Huh? Don't deepfakes do the exact opposite - make plausible deniability in audio and video a much stronger argument?
Before if you denied a video of you was genuine that would not be plausible. Now it would be plausible.
It’s going to get to the point where you can’t even say in a court room “well we have video of him doing this”. The fact that deepfakes will exist will erode confidence in even those things that are true. At the same time it will add additional fake situations to the conversation.
Worst part is that eventually even the AI-powered counter measures are going to fail eventually. The moment a computer knows what gives away a hint that it is a deepfakes and not real, a computer can solve to not present that give away. The “good guys” and “bad guys” will iterate with each other until it is perfected.