Hacker News Reader

Comment by SamBam | Hacker News Reader

What really worries me about the DeepFakes stuff is not so much the fakes -- I'm worried about that, but that's been written about to death -- but that now it adds plausible deniability to anyone caught on camera doing anything.

If we had video footage of, say, a politician doing something clearly illegal on camera, then it's simply the word of the politician against the word of the source -- the latter of whom may need to remain anonymous.

That said... the other way around is still the more dangerous, I guess. If, say, a US adversary creates a great-looking video of the US president doing/saying something really heinous, that has the potential to inflame the world long before the truth can get its boots on.

View on HN

It just pressures us to come up with ways to prove authenticity. Those ways definitely exist but are not common yet and we have courts and governments full of people with a poor grasp on technology. But basically it calls for chains of evidence that are cryptographically tamper proof. That's not a thing right now. But it's going to become a hard requirement when evidence can be fabricated, falsified, etc. It's also the key to countering fake news and a few other things.

People accepting everything at face value is not going to stay a thing when world+dog is going to abuse their new powers. Only fools would believe what they see after they've been fooled a few times and suffered the consequences.

Short term it's going to be a mess, but long term it's a good thing for us to figure this out and move on.

I don't see how you can cryptographically validate much more than "this was validated by this source before this time", which doesn't seem to solve the problem stated by the parent at all.

Maybe you could have some DRMish thing where the camera signs it with a "secret" key, but this would be terrible for various reasons and also likely broken very fast.

I don't think so. You can cryptographically sign anything much like how SSL works now. You'll have to rely on certificate authorities to assign these certs, but it works.

Videos should be cryptographically signed, and verified once online. You can spoof certs but you can't really fake the cert authority

That's what I meant by "validated by this source". But unlike with CAs, where they're (meant to) just base issuance on the simple objectively testable criterion of whether you control the domain in question, an external authority cannot easily know whether a video represents real events, whatever that means.

But, signing some data with a certificate only indicates that a key belongs to a particular name. It doesn't tell you whether the person or organization with that name is trustworthy.

If you authenticate the video comes from a credible unrelated source, that would be different than if it came from a mysterious unknown source. Additionally if you have the chain of trust, you can interrogate every step manually for credibility and consistency.

Which is somewhat helpful, but also just pushes the validation work off onto large entities of some kind.

Yes, of course.

The value of it is that the legwork only has to be done once, instead of requiring everyone to independently do it (which would basically turn every accusation of crime into a DDoS against the accused).

If you have cameras that sign their video feed with some id; editing software where an editor signs off on any edits, peopling handling/validating the content adding their signatures, etc. you build a chain of digitally signed content based evidence that you can follow all the way back to the original recording.

Then you can get people into court testifying whether they used a given piece of equipment to film something, edit something, etc. and you can guarantee that you are watching the exact output of that chain of recordings, edits, etc.

As I said, not a thing right now. But also not that technically hard to build. Right now we're just trusting witnesses that might be lying through their teeth without us knowing or being able to prove otherwise. Once we had such capability; anything else would be inadmissible in a court and no self respecting journalist would touch equipment without this capability. Why would they?

A deep fake would look plausible but lack this chain of evidence.

This doesn't seem significantly better than just having the organization providing a video sign it as "authentically theirs", in cases where that's possible; if you mean some sort of thing where editing software and cameras will sign things as "not tampered with", then this is effectively a DRM system and subject to the excitingly wide range of issues affecting that. This would not work for many situations, particularly the ones SamBam describes (not least due to the anonymity thing), as it is unlikely that there will conveniently be someone there with chain-of-trust-capable recording equipment and software.

Even if I don’t have the private key to sign the video nothing stops me from sending the processing element of the camera the same signals as the photo array. Even if you encrypted the connection between the processing elements and the photo array. A photosensitive array is already an exposed die so I could easily just set some needles on a few internal traces and again do the same thing.

There is no known solution to the analog hole.

It's turtles all the way down :-)

Maybe a service or a public blockchain where you send a hash of a digital artefact which is signed with a time constrained key. The signed hash is attached to the digital artefact and you can check the hash on the blockchain or on the api's service.

A blockchain is more wasteful, but a service requires a leap of faith in the provider.

This is still just a way to validate when something existed, isn't it?

Yes, and this is rather strong feature. A fake will be effective only if you have prepared it in advance instead of revising the past.

> But basically it calls for chains of evidence that are cryptographically tamper proof

JusticeCoin ICO when??

Just because some opportunists did a few ICOs does not mean all crypto is bad. Without crypto there would be no online banking, or any form of digital security, secure logins, etc. Crypto is a useful tool. Blockchains are a tool. And so are digital signatures. If you combine those tools, you can do some useful things like creating tamper proof audit logs documenting where information came from all the way from the sensor to your eyeballs. It's just a chain of digital signatures.

I tried hard to avoid using the word block chain in the original comment to avoid exactly this kind of knee jerk response. But yes, kind of an obvious tool to use to record chains of evidence in a tamper proof way. Glad you jumped to that conclusion as well.

Contrary to the popular belief, not every block chain based thing has to be an investment scam. I don't think we need a separate coin for this; just a shared repository of truth and fact with full auditing. Blockchains are kind of designed to be that. If you know a better way, please provide it.

And just to pre-empt it, obviously my preferred flavor of block chain for this would be miner free proof of stake rather than proof of work.

You jest, but distributed publicly verifiable proof that a certain piece of information existed, and was cryptographically signed at a certain date and time, and has not since been modified, is basically the only thing blockchains are actually useful for. And that sounds much like what we need here.

Even then it's not required, because it's overly distributed. You only need one copy of the Merkle tree (+ backups as necessary); to verify it hasn't been changed you don't need everyone else to copy the whole thing, just random samples of it to verify it hasn't been rewritten recently.

This is not a technical problem IMO. You can't cryptographically sign reality. There's always the analog hole[1].

It's a problem of trust within society. Look at many very mainstream conspiracy theories these days: there's ample proof that it's not true, but people want to believe so they'll believe.

You can't fix the lack of trust in society with cryptography.

I just went on a constructionist website I sometimes lurk when I'm bored, literally the first story I find is titled `The Age of "Credentialism" and "Experts" is over. Every Single Institution works against your interests'. You can't fix this mindset with maths.

Video is cryptographically signed? But what about the secret computers in the Pentagon's basement that run on Quantum CPUs using ancient alien technology found in the pyramids? They can certainly break ECC. Here, watch this Youtube video...

[1] https://en.wikipedia.org/wiki/Analog_hole

Seconding this. The fracture at the heart of our society is not rooted in unintelligence, or lack of education or access to facts. It's rooted in a lack of trust. The most rigorous science is only as good as the trust people place in those performing it (and the entire chain of reporting from there to its reception). Those links are being/have been broken. No amount of added rigor will fix that. I don't know what will.

> Those links are being/have been broken. No amount of added rigor will fix that. I don't know what will.

The institutions could start telling the truth once in a while. Statements like "27 police officers injured during largely peaceful anti-racism protests in London" (BBC) and "Fiery but mostly peaceful protests" (CNN) with the city burning down in the background account for why trust in media is rapidly approaching zero. It makes it clear that the violence and devastation is just a curiosity to the upper-class elites that control these institutions, but normal people who actually have to live with aftermath are not amused.

Fauci is another huge contributor. First don't wear masks, they can actually hurt. Then you have to wear masks. Now wear two or three! But earlier he was ridiculing people who wore multiple masks. Along with all the knowingly false statements about how long lockdowns would last. Clearly Fauci should not be allowed to speak in public, but unfortunately the blast radius of his mistakes extends beyond him to the government and media institutions that defend him and amplified his misinformation.

with the city burning down in the background

No exaggeration there ◔_◔

I get that you're complaining about 'liberal' media downplaying disorderly and often dangerous events such as riots to suit a political agenda, but the converse is also true; if one relies on 'conservative' media then you'd think many major American cities are post-apocalyptic smoking holes in the ground.

My larger complaint was about rich people who don't care about the several dozen deaths, uncounted injuries, and billions of dollars of property damage caused by the riots because it doesn't affect their class. I don't watch conservative media, so I don't know anything about the portrayal of major American cities therein.

I'd argue it's the opposite. We've relied too much on trust and promises and not enough on rigor and exactness. Most people aren't given facts, constraints and limitations until it's too late, if ever. What they're given are viewpoints with selective evidence and glaring omissions. It's remained a unsolved since the beginning of humanity. A solution to this problem is a rigorous self-proving system that wouldn't need one to convince another of the facts.

I don't want to live in a trustless society. What you describe sounds like a totalitarian dystopia to me.

I want escape hatches. I want plausible deniability. Facts in a vacuum are useless and can be used to propel all sorts of narratives. Facts without framing and contextualization aren't worth much. You can manipulate easily without technically lying, just by cherry picking facts that suit your agenda.

You need some amount of trust and solidarity if you want to live in a healthy society.

You can't sign reality, but people can sign statements.

If you see a video at a date D1 where you say Statement1, and cryptographically sign it with your key K, then at a later date someone can verify that at least you said you've watched an backed your statement.

In a way all of security relies on the physical safety of some kind of secret data. So you have the deniability of key compromise in any case.

If everything a president states publicly is signed with his key Kp, then:

1) If something controversial is published without signature, the president can say it's not standard procedure and a plausible forgery;

2) If the president publishes officially without a signature, the public can demand one so there's no later equivocation;

3) Anything that has been said can be verified in the future by checking the presidential signature.

In this case, the worst case is really a compromised key (although key scheduling should mitigate it), but most forgery cases of statements (and potentially documents, mandates, etc.) are eliminated.

In practice, it would be difficult to get your public figures to sign everything they say (and difficult to get them to accept this kind of potential auto-incrimination for the public good).

That doesn't work for adversarial recordings. No one will sign an embarrassing or damning video of themselves, and those are the cases where authenticity is the most important.

Your solution is technologically cool, but I think the current system of "was this published by a domain controlled by the office of the press secretary" is probably accomplishing this case well enough.

For adversarial recordings, I wonder if we could have a camera that instantly uploads a timestamp and hash of each video taken to a blockchain. This way, we could have videos that we know for a fact were recorded at the latest at a certain time. It would still be vulnerable for a fake video to be post dated, but never pre dated.

Doesn't stop the adversary from simply timestamping, and uploading their fraudulent video's hash to that blockchain.

I was recently watching The Brainwashing of My Dad. Rush Limbaugh made millions of dollars selling lies to the American public. At one point there is a video clip of him admitting that the truth of what he is saying on radio is irrelevant. It doesn't matter to him. He lays out the recipe for generating fear and uncertainty. Basically, if you say something loud enough and with enough confidence no one will stand in your way. Because, as fact checkers know, it takes considerable time to research a bullshit claim. By the time the research is done and published, the bullshitter has moved on and told an additional 20 lies. He also makes a statistical argument for how his business works. He doesn't need to hook every listener. But he does know that enough people will fall for his shtick.

The troubling aspect is that all of this bullshit is blending together. My dad watched Fox News. Now he's hooked on Youtube conspiracy garbage. I'd be terrified if he ever became a QAnon type. We're dealing with literal internet cults becoming a mainstream phenomenon. We're nowhere near equipped for the mess we, the technologists, have made. You have Alex Jones out there claiming that an elementary school shooting didn't happen. You think these guys are going to trust encryption? Or anything that their Youtube priest tells them is a "hoax"?

Education would be the answer. But education is at war with engagement algorithms and attention spans.

This is a technique known as the Firehose of Falsehood:

https://en.wikipedia.org/wiki/Firehose_of_falsehood

The goal is not necessarily to convince people of a particular claim, it's to levy so many claims and use the scrambling of media acting in good faith to vet those claims as an overloading mechanism to get regular people to tune out entirely.

Also known as 'flooding the zone'. Like many political tropes, this originates in team sports which in turn is an abstraction of war. It's an awful lot easier to understand the media landscape if you consider it as cultural warfare with ideas and tropes as territory, although this is hard to visualize in spatial terms.

The answer to this (and the accompanying tribalism that pervades public discourse nowadays) is often said to be education and critical thinking, but that requires years of investment and often-unwelcome external discipline to internalize and actuate; it's a statement of what we would like to have instead rather than an actionable solution to its own absence.

Friendly emotional persuasion can work better as a de-escalation-bridging tactic, as suggested here: https://dr-gleb-tsipursky.medium.com/how-to-talk-to-a-scienc...

This is also helpful for gathering information to understand the dynamics and attractiveness of false information, even if no changing of mind can occur; think of it as the difference between carefully dismantling an unexploded munition in order to figure out how it works vs. a controlled explosion to minimize future risk at the expense of continued vulnerability.

Where conflict is unavoidable or deliberately fomented (eg people arguing in bad faith rather than sincerely believing falsehoods), an overtly hostile response imposes a cost on the aggressor, and when consistently and predictably applied it effectively alters the payoff matrix in an adversarial game: https://snap.stanford.edu/conflict/

Many people are aware of Mutual Assured Destruction as a kind of nuclear diplomacy, where you are deterred from nuking me because I've made it very clear that if you do I will take you down with me, leading to a heavily armed but uneasy peace. There are also lesser-known concepts like Power Transition Theory (about how wars originate from weaker countries challenging stronger ones) and nowadays scholars of international relations tend to adhere to Hegemonic Stability Theory (one very powerful country plays Teacher/cop) or World System Theory (every dog has its day). Developing familiarity with the broad concepts of interstate conflict (without going too deep down any intellectual rabbit hole) can be helpful in modeling smaller scale political conflicts, divisions in civil society etc.

What is this analog hole of which you speak? Only nature delivers analog ;)

But nature doesn't create deep fakes - though not as the term is being used here. I would argue that nature has been making deep fakes for millions of years.

Anything that is created by CNN Deep Fake tech is delivered via computer - either using printer or with some sort of screen screen. Let's eliminate printers as nobody uses them anymore. What about screens? I smell a business opportunity.

This is not a technical problem IMO. You can't cryptographically sign reality. There's always the analog hole

Reality has a pretty strong "hash" naturally built in. Even manipulated picture carries a huge amount of information that isn't changed or isn't changed as much as one thinks.

The manipulated pictures the motorcyclist uploaded still would give someone a good idea where and even when they were taken. That doesn't matter here but if you're making a more detailed argument, it's harder.

Just consider, "computer forensics" is a thing even though any single bit on the computer can be overwritten and "faked".

Your theory is basically the theory behind Minority Report.

>People accepting everything at face value is not going to stay a thing when world+dog is going to abuse their new powers. Only fools would believe what they see after they've been fooled a few times and suffered the consequences

This seems overly optimistic, and requires people to themselves suffer unambiguously from the doctored evidence.

On an individual level, regular discovery of police and prosecutorial has not led widespread reform in those areas. And on a larger scale, even after things like the Gulf of Tonkin people largely accepted claims of WMD's in Iraq.

I think people got my point, but I somehow forgot the word "misconduct" after "prosecutorial."

There are people who still claim in public to believe that the US election was stolen.

Acceptance of evidence is socially constructed. If it's politically convenient to go along with the beliefs of your faction, and you're rewarded for saying increasingly ludicrous things in public, then people are going to do it.

Isn't this what Adobe is trying to do? [1] It was posted on HN a few times but never started a discussion that I saw. Personally, I'm scared of it although I can't put my finger on why.

1. https://blog.adobe.com/en/publish/2019/11/04/content-authent...

Deid solves this. Couple of projects are building this on various blockchains.

Lying is not a new phenomenon. Whether or not it's useful or good is a philosophical question with many answers which don't matter because if it becomes possible we will (have to) adjust and there will always be some upside if you want to see it.

Today, lying with video is really expensive, but if in the future everyone is able to do it and everyone knows that it's possible then we're back to he-said-she-said, which is how society has worked since the beginning, except the last 200-ish years. I can easily lie and say I saw Bill Clinton murder someone in 1990. Maybe in the future I'll be able to generate fake video evidence of it just as easily as I typed out that sentence. If everyone has a feel for how easy it is, then so what?

The danger is in the transition, when lying using video is affordable by a select few, and not everyone knows about it. Then it's powerful. If you're worried about that then we need to develop and teach this technology as quickly as possible. We can also do something fun like collect a time capsule of important videos before this becomes easy and timestamp it in a verifiable way (by posting the hash to a blockchain or some authority for example).

Very good post. Another factor to consider is that a technology can be powerful when everyone does know a bit about it, but the perceptions of around it can be manipulated.

To stick with your 'Bill Clinton murdered someone in 1990' example, if you are Alex Jones and you say that then I will be skeptical gien Alex Jones' serial unreliability. But if you deepfake yourself to resemble a (hypothetical) real person called Albert Johannsen who died in 1995, and manipulate the video to look like old VHS, then the authentic-seeming testimonial can be 'discovered' by someone clearing out an attic or storage unit, and then merely publicized by you-as-Alex Jones, who merely reports the claim of the discover (actually a collaborator of course).

There is an endless variety of of applications, eg you have really committed a crime and video exists, but you produce a deepfake of yourself committing the same crime multiple different ways or the like, such that everyone thinks You Did It but nobody can agree about exactly how or to what extent and you escape justice due to the ambiguity (albeit with diminished future prospects).

There are old examples with photos including newspapers/TV that used faked images, so it not something new. If there is a big punishment for intentionally faking videos and presenting them as true similar like for official documents or impersonating officials then this technology would be used a lot less in the countries where this laws apply.

Convincingly faking photos was, until recently, pretty damn hard.

This is currently possible by just using a screenshot from a game and an app from the AppStore: https://twitter.com/nillxzero/status/1369452664979943427

I think people's ability to detect fake photos have increased over the years. Consider the Cottingley Fairies photos (https://en.wikipedia.org/wiki/Cottingley_Fairies), to my modern eyes, the fairies are unmistakably fake: the lighting is wrong, the resolution is wrong, the texture is wrong. However there were people who were convinced that these were real photos of fairies back then.

> Convincingly faking photos was, until recently, pretty damn hard.

Stalin did it nearly 100 years ago.

But its not about evidence. Its about shifting public opinion. There is and most probably never will be a law that prevents people from sharing/liking/commenting or otherwise trigger the algorithms to spread something based on the fact that it is fake. Media also fakes stuff by simply not providing context. It need no "active" image manipulation. cutting and embedding it into other content works fine since years and apparently there is very very little that can legally be done against this. The fact that the original is out there is also rather meaningless. take for example the very popular "very fine people on both sides" quote, nothing stops you or me to listen to the full conversation. But even today from random people on the street who recognizes the quote probably 50% do not know the real context. Most probably because they simply dont care enough. They involuntary heard the quote but they will not involuntary hear the context, because one goes viral the other does not.

I was thinking at some tabloids, sure some anonymous can post fake stuff but if a journalists creates the fake image/video or documents then there should be more consequences. Also I am not from US so I am not targeting a specific camp, when I was watching news (I stopped years ago_ a lot of energy was spent on discussing insinuations, fake stuff or trivial things. I realized that politicians know how to throw the media some delicious bones to keep them busy with whatever they want.

Its called "accountability", something mainstream journalism doesn't seem to have in most places. Its just a fact that biased "news" get more clicks and thus more money. Its hard to define when "faking" starts and where its is just non-neutral reporting.

> it adds plausible deniability to anyone caught on camera doing anything.

I see this concern brought up frequently but I don't really think this is a big deal. In the grand scheme of things, ubiquous availability of video cameras is a fairly new development. Video cameras themself are a fairly new development.

We had a functional society before video evidence and we will have a functional society after video evidence.

There was a very short window of time in which we had somewhat reliable video evidence but it is now coming to an end and we will manage.

Functional? perhaps, but I would say that if one peer into history, what one finds is suspects convicted on evidence that would seem flimsy by today's standard, and so will the future look at us.

I would think that the number of false positive convictions declines with the advancement of new technology, and that this is perhaps a temporary setback.

There was a time before fingerprints and d.n.a. evidence as well, both of which have been very helpful not only in convicting the guilty, but in exonerating the innocent, and if ever come the time that it be feasible and affordable for a layman to plant fake d.n.a. evidence and fingerprints, that would be quite a setback for criminal forensics.

I think it's even simpler -- liars attracting any degree of press attention won't be able to get away with it.

"That picture was Photoshopped" doesn't work as a defense because it's not too difficult for experts to tell a Photoshop from a genuine image, nor to interview someone at the scene of an alleged event and learn the truth.

The same will be true with deepfakes, only much more so, because there are so many more ways to give away that it's a deepfake. Video adds new dimensions of scrutiny like how well the fake face tracks the head, matches lighting and expression, etc. Deepfake detection is in its infancy but you can bet it'll be even more accurate than Photoshop detection.

"If, say, a US adversary creates a great-looking video of the US president doing/saying something really heinous (...)" Did we not see enough fake stories about US presidents already? The DeepFakes are not really needed to falsely accuse a politician, or to wash away the compromising material. The media can use anonymous sources to claim anything bad about any public figure without consequences. Also, they can ignore and diminish any compromising material, effectively shielding the subject. It is all about who controls the media and subsequently the narrative. DeepFake may or may not be used in court though.

This isn't much different than the problem we've had with internet for many years, from inflammatory headlines with no substance in the story on down.

"A lie can travel half way around the world while the truth is putting on its shoes." - Mark Twain

>pedant alert<

"...there exists a family of expressions contrasting the dissemination of lies and truths, and these adages have been evolving for more than 300 years. Jonathan Swift can properly be credited with the statement he wrote in 1710. Charles Haddon Spurgeon popularized the version he employed in a sermon in 1855, but he did not craft it. At this time, there is no substantive support for assigning the saying to Mark Twain or Winston Churchill." [0]

[0] https://quoteinvestigator.com/2014/07/13/truth/

I like to look further into the past for ruminations that, while not as concise, explore the same territory. A passage from the Aeneid says -

Rumour the swiftest of all evils. Speed lends her strength, and she winds vigour as she goes; small at first through fear, soon she mounts up to heaven, and walks the ground with head hidden in the clouds. Mother Earth, provoked to anger against the gods, brought her forth last, they, say as sister to Coeus and Enceladus, swift of foot and fleet of wing, a monster awful and huge, who for the many feathers in her body has as many watchful eyes beneath – wondrous to tell – as many tongues, as many sounding mouths, as many pricked-up ears. By night, midway between heaven and earth, she flies through the gloom, screeching, and droops not her eyes in sweet sleep; by day she sits on guard on high rooftop or lofty turrets, and afrights great cities, clinging to the false and the wrong, yet heralding truth.

There's plenty to worry about with deepfakes. They enable what amounts to a man-in-the-middle attack in real life. Imagine an AI able to break into the conversation and maintain two different communication threads with each party.

Imagine what happens when you present someone with the voice and/or image of their loved one and pitch a product, plant an idea, or execute fraud? What happens when an emotionally unstable person is harassed by the voice of their lover or even a deceased relative.

Now imagine that you could rent some computing power and perform a million of those frauds in parallel. Psychological warfare will never be the same.

When people tell me the robots are going to kill us someday, I always reply that they won't have to, they'll just need to convince us to kill ourselves.

-- How often are politicians caught on camera doing illegal things now? I don't recall this happening very often. The bad behavior on camera I remember from way bad when ("Abscam") was an FBI entrapment scheme so you have more than just the video as testament.

-- The most common "guilty by camera" situation in recent years have been cops and when a scene from an event where other facts are known gets filmed, you need far more than face manipulation to make a fake that's going to be plausible.

Chinese deepfakes are going viral, and Beijing is freaking out

https://www.protocol.com/china/chinese-deepfakes-regulators-...

I'm comforted by history. Shopped photos don't mean we distrust photos entirely, but we're now generally "herd immune" to silly things in a photo because we all know that it came from somebody's edits.

This will hold true for deepfakes. It means that we might start judging the character of a person, their surrounding information (e.g., who they know who vouches for them) instead of the core information itself. It'll be a bit like how cybersecurity involves trusting the trust of trustworthy sources: not perfect, but doable.

Personally I think this might be a good thing. The alternative is a squeaky clean politician that has no character, mistakes, or experiences.

> I guess. If, say, a US adversary creates a great-looking video of the US president doing/saying something really heinous, that has the potential to inflame the world long before the truth can get its boots on.

By the time this becomes possible, it will be widespread knowledge that deepfakes exist.

It's already possible. Look at the Tom Cruise deepfakes. It just requires the power to get a great actor of similar build, and lots of processing power, both of which are within the reach of a state, easily.

And I don't think knowing the deepfakes exists will convince everyone. Plenty of fake photos already get passed around on Facebook and what-not, even though Photoshop has existed for decades.

Even with widespread knowledge that deepfakes exists, you can still make credible videos of people that damages them.

For example, picture a video of Trump saying in private committee 'I want to fuck that nigress'. That would be extremely heinous, and probably be a fake, but can you be 100% sure from a president that also said 'grab them by the pussy' ?

The line between real and fake news is becoming more and more tenuous.

Wait till we have real-time deep fake video and audio. That will twist the world, especially the remote work/schooling/presence one.

So we'll return to how guilt or provenance were determined for thousands of years. We've only had "reliable" non-human witnesses for a few decades.

And even with deep fakes, we'll still have fingerprints, DNA, phone and car tracking, facial ID systems in public places, etc.

The problem is much bigger, i can now use deep fake to present myself in a very relaxed way with a lot of confidedence and great facial expressions. Now i can deliver video presentations like a champ, just improve myself. Its me and not me.

Machines that write believable stories can fabricate entire realities. The deepfake stuff is just a very small part of it. How much of the information we use to make our daily decisions on come from unverified sources on the internet? How many of the "trusted" sources are just paid PR for the rich and powerful?

The deepfake stuff can be solved with public/private key crypto anyways.

How do you solve deepfakes with asymmetric cryptography, without just trusting the person the videos purport to be of to say (cryptographically) whether something is real or not (which is not a good solution)?

How will the adversary get the deep fake into circulation? If it suddenly appears on social media from some sketchy account, and in the meantime the White House press pool points out that they were with the President in a completely different place, it will blow up. And it would have to be technically perfect or it would fall apart on analysis.

The plausible deniability of camera footage always existed, even in the film days. There are lots of tricks to fake things on camera. Neural networks do make it much easier, but it's not new.

That said though, make forging evidence a crime of the utmost seriousness. Also, /more/ cameras always makes deepfaking harder.

Would better cameras circumvent the issue?

Sure a deepfake is convincing on a low resolution compressed video, but what about a 4k or 8k video where lens distortion and fine details are everywhere? I don't have a lot of confidence that facial pores could be convincingly simulated.

But are most secret camera recordings using high-res cameras?

Let's say there was video recording of a politician in a hotel in Russia, to pick a random example. We would already expect it to be low-quality video even if it was real.

>Let's say there was video recording of a politician in a hotel in Russia, to pick a random example. We would already expect it to be low-quality video even if it was real.

But why? The cheap last-gen iPhone in my pocket can take 4k videos, almost everybody has something like it in their pockets. Quality is cheap and will continue spreading, fuzzy video that looks like it was taken on a 90s camcorder will itself seem suspicious.

Generating those low level feature is actually a much easier problem. Here is an example I found using a google image search for "upscaling":

https://copyrightimage.com/2018/05/09/better-image-upscaling...

This is from 2018, which is an eternity for machine learning research.

But those kinds of things also could reveal the fraud - adding a blemish where the real person doesn't have one, features that move around frame by frame in a video. The difference is things which look believable might be details which are trivially verifiably false - sure you can simulate my face and my voice, but can you accurately simulate every pore on my nose.

Here's a random example:

https://ieeexplore.ieee.org/document/6619284

They can detect a persons pulse from tiny variations in head movement, I'd like to see a deepfake simulate that! And it's not about this one thing, I'm sure you could design against these tiny measurements one by one... the key is that there are a lot of them and a deepfake is going to have to simulate an enormous amount of data in order to pass what will become trivial tests for realism. This is made a much easier weapon against deepfakes when you have progressively better videos.

A single video itself would not meet a preponderance of evidence required to secure a conviction.

Also,as most states are 2-party consent for video recording the scenario you described would have a huge evidentiary rules hurdle regardless of the deepfake capability.

"News" organizations already edit video to maximize emotional impact to viewers, often in misleading ways.

I don't see this as a paradigm shift as much as a continued slide down the slippery slope of normalized propaganda.

It also has the potential to devalue real evidence in the form of existing compromising material unless people have done things like adding cryptographic hashes to immutable public records with attestations.

I wouldn't worry about it too much. Having so much on video in the first place is a pretty recent phenomenon. 30 years ago, politicians had a very low chance of being caught on camera at all.

It comes to min that I know of few politica scandals that were revealed by anyone caught on camera.

Any evidence relied on in court needs a chain of custody, and pictures and videos are no different. It's why screenshots are generally inadmissible in court: anyone could have edited them.

More megapixels!

I used to be more worried about this, but one of the things Trump proved is you can have unquestionably authentic video evidence of lies and many people won’t care.

The people that care about figuring out the truth and critical thinking will probably still figure it out.

Others will believe crazy things for stupid tribal reasons despite contrary evidence.

Reply on news.ycombinator.com