NeRF: An eventual successor for deepfakes?

I believe we'll go back to trusting local experts that you can meet in person to confirm that they are not a bot.

Because anything online will be known to be untrustworthy. Most blogs, chat groups and social media posts will be spam bots. And it'll be impossible for the average person to tell the difference between chatting with a bot and chatting with a human. But humans crave social connections and intimate physical contact. So people will get used to the fact that whoever you meet online is likely fake and so they'll start meeting people in the real world again.

I also predict that some advanced AIs will be classified as drugs, because people get so hooked on them that it destroys their life. We've already banned abusive loot box gambling mechanics in some EU countries, and I think abusive AI systems are next. We'll probably also age-limit generative AI models like DALL-E, due to their ability to generate naughty and/or disturbing images.

But overall, I believe we will just starting to treat everything online as fake, except in the rare case that you message a person which you have previously met in real life (to confirm their human-ness).

Surprise Plot Twist: Maybe we're already living in a post-truth society and you are still sure you know what the truth is. How would you even know that what you were ferociously defending as the truth wasn't a lie? What makes you think you're not smart enough to not fall for lies?

Largely, I think most people's means of finding the truth is just to take a vote of the information sources they find credible and go with whatever they say. I was talking with some friends about the California propositions a while back. Some of them were not clear cut which way we should vote on them. Instead of discussing the actual issue, people just wanted to know what various authority figures thought. These were not dumb people I was talking to, and I used to remember an era in the 90s maybe where you could actually have a reasoned debate and come to the truth that way. It seems that's obsolete these days since nobody seems to agree on the basic facts about anything.

The real problem isn't the veracity of the information, but the consensus protocol we use to agree on what's true. Before the internet, we were more likely to debate with their neighbors to come to an understanding. Now, with the large bubbles we can find ourselves in, afforded by the internet social media, we can find a community to agree on anything, true or not. It's that lack of challenge that allows false information to flourish and is the real problem we need to solve.

A skeptic (i.e. someone who cares to verify) not being able to trust media because it might be fake is only a minor problem as long as you have at least one trusted channel.

The president, say, can just release the statement on that channel and it can be verified there (including cryptographically, say by signing the file or even using HTTPS).

If you lose that channel, then you're pretty much screwed because you'll never know which one is the real president. But there are physical access controls on some channels, say the Emergency Alert System, which can be used to bootstrap a trust chain.

What will be much more possible is that someone who will not check the veracity of the message will take it at face value without bothering to validate it. This is your news-via-Facebook crowd.

At that point, it's less a technical issue than simply people don't want to know the truth. No amount of fact-checking and secure tamper-proofing of information chains of custody will help that.

There has never been a truth society.

This tech will certainly be used to frame someone for a crime, like I am sure Photoshop was used in such a way, and thousands of other techniques. And modern technology offers counters. It is an arms race but because of the sheer amount of data that is collected, I think that truth is more accessible than ever. The more data you have, the harder it is to fake and keep consistent.

I don't think it'll be way too much different than it has been for most of human history. We really only had a brief blip of having video, which was generally trustable, but keep in mind that before that for thousands of years it was just as hard to know truth.

Someone told you stuff about the outside world, and you either had the skepticism to take it with a grain of salt, or you didn't.

This doesn’t bother me that much because evidence isn’t required to convince millions of people that a lie is true. We already know this. Why make fake evidence that could be debunked when you can just have no evidence instead?

People with some degree of knowledge already know that any photo could be photoshopped. People that don't care will blindly trust a picture of someone with a quote or caption saying whatever, as long as it fits their narrative.

This has been the case for photos for almost 2 decades. The fact that you can now do it with video or audio doesn't change that much IMO.

> I don't think we'll have the ability to handle this kind of tech responsibly.

Makes you also think whether anybody from the Hacker News crowd working on any contributing tech is acting ethically responsibly. For myself, I have answered this question with "no", which rules out many jobs for me, but at least my kids won't eventually look me in the eye and ask "how could you?"

Sure it's cool tech. But so was what eventually brought us nuclear war heads.

I was listening to "This American Life" and they had a segment on someone who setup an site to give you a random number to call in Russia where you were supposed to give them info about what's happening in the Ukraine. It was someone shocking to hear their side of the story, that Russia is a hero for helping oppresed Russians in Ukraine.

But then I stepped back and wondered, I'm assuming that the story I've been told is also 100% correct. What proof do I have it is? I get my news from sources who've been wrong before or who have a record of reporting only the official line. My gut still tells me the story I'm being told in the west is correct, but still, the bigger picture is how do I know who to trust?

I see this all over the news. I think/assume the news I get about Ukraine in the west is correct but then I see so much spinning on every other topic that it's hard to know how much spinning is going on here too.

I guess I'm an outlier on this topic because I don't think deep fakes will continue to have significant societal impact after an initial disruptive period of perhaps a few years. During that time DFs will become so common that any impactful example will immediately be suspected and discounted by most people. Yes, there will always be those who choose to believe the unbelievable but the lunatic fringe already believe outlandish claims without the need for deep fakes, thus there will be little net change.

I think we'll solve it the same way we solved similar transitions when text and image faking became easy: provenance.

For many years now, most have understood that you can't take text and images as truth because they can easily be simulated or modified. In other words, the media itself is not self-verifying. Instead, we rely on knowing where a piece of media came from, and we associate truth value with those institutions. (Of course, people disagree on which institutions to trust, but that's a separate issue.)

I think that people will adapt. Humans are very clever and have been evolutionarily successful because of the ability to adapt to a wide range of environments.

Think about how devastatingly effective print, radio, and television propaganda at the time each medium was widely adopted compared to how effective they are now. They still work, but for the most part people have caught on to the game and adjusted their cognitive filters.

My guess is that we will see a bifurcation of society into those who are able to successfully weed out bullshit from those who can’t. The people who are able to process information and build better internal models of the world will be more successful, and eventually people will start imitating what they do.

Edit: I do think that these tools coupled with widespread surveillance and persuasion tech (aka ad networks), have set up conditions where the few can rule over the many min a way that was not possible before. I do think some of the decentralized / autonomous organization tech - scaling bottom up decision making to make it more transparent and efficient - is a possible counter. Imo, this struggle between technological mediated centralization and top down enforcement and control vs decentralization and bottom up consensus will be the defining ideological struggle of our time.

Here's the thing though: in theory, we should already be skeptical of video and audio evidence on its own.

Most of our institutions, in theory, do not focus on single mediums for assessing veracity of truth. The strength of claims and our ability to split the difference between noise and truth comes down to corroboration. How many other sources strength and work consistently with a claim? That's, in theory, how law enforcement, intelligence, and reporting should work.

In practice, there are massive gaps here and people's attention -> decision is lower than ever.

I don't think it's impossible for us to handle deep fakes, but I sense the same fear you have. I think ultimately it is more about our attention spans, and the "urgency" we feel to act quickly, that will be more of our down fall than the ability to produce fakes more easily.

You don't in fact need a convincing fake to create a powerful conspiracy theory. Honestly you only need an emotional provocation, maybe even some green text on an anonymous web form.

So there are information theoretic ways to certify that media is genuine, if you assume trust at least somewhere in the process. Basically just cryptographic signing.

For instance, a camera sensor could be designed such that every image that is captured on the sensor gets signed by the sensor at the hardware level, with a certificate that is embedded by the manufacturer. Then any video released could be verified against a certificate provided by the manufacturer. Of course, you have to trust the manufacturer, but that’s an easier pill to swallow (and better supported by our legal framework) than having to try and authenticate each video you watch independently.

There are issues that can arise (what if I put a screen in front of a real camera??, what if the CIA compromises the supply chain???), but at the end of the day it makes attacks much more challenging than just running some deepfake software. So there are things that can be done, we’re not destined for a post truth world where we can’t trust any media we see.

The concerns in your second paragraph can be mostly mititgated using a combination of trusted timestamping, PKI, cryptographically chained logs and trusted hardware. Recordings from normal hardware will increasingly approach complete untrustworthiness as time goes on.

The concerns raised in the first paragraph however... the next few decades are going to be a wild ride. Hopefully humanity eventually reaches an AI-supported utopic state where people can wrap themselves in their own realities, without it meaningfuly affecting anyone else. Perception of reality is already highly subjective, most of the fundamental issues are due to resource scarcity/inequality. Most other issues evaporate once that's solved.

"I don't think we'll have the ability to handle this kind of tech responsibly."

I do not think so either, but so far we survived 75+years with nukes around.

But you can argue, it was mainly by chance. Technological progress is awesome, but our societies cannot keep up yet. They will have to do heavy transition anyway, or perish. Or rather, we are in the process of transition. 20 years ago most people did not really know, what the internet is, now most are always online. Data mining, personalised algorithms for ad exposing, ..

So deepfakes are a concern, but not my biggest. Rather the contrary, when people see how easy it is to fake things, they might start developing a healthy sceptism to illuminating youtube videos.

I don't know exactly what you mean by "post-truth", but I think for the vast majority of human history we've lived in a time where it was rare to have easy/cheap to create and hard to fake records of what happened (photography, videos, audio recordings).

In terms of a single photograph though that's been very easy to fake in compelling ways for at least the past 30 years look at lochness and bigfoot photos.

Video's harder, but still pretty much all of this is not horribly difficult to fake. We've long been able to edit things out and mix audio to make a compelling video of a cartel leader meeting with them, no need for deep fakes.

Thankfully things that a real are very cheap to follow up on. Questioning some security footage? No worries there's 100+ hours of it to cross check with the footage and three other cameras too.

IMHO, consilience and parsimony will save us.

I think a bigger question is whether reputable sources --those people trust for whatever reason, would use this technology to prop up ideas and or to create narratives.

I don't think it's far-fetched. We've already seen where videos are misattributed[1] to stoke fear or to promote narratives --by widely trusted news sources.

[1] This was foreshadowed with "Wag the Dog" but happens often enough in the media today that I don't think use of "deepfake" technology is beyond the pale for any of them.

Reminds me of Stephenson's "Fall; Or Dodge in Hell" where all digital media is signed by their anonymous author and public keys become synonymous with identities. An entire industry of media curation existed in the book to handle bucketing media as spam, "fake", "true", interesting, etc.

I think society will adapt within 1 generation. The tech is already there (signing messages with asymmetric encryption)

The one unwavering thing about technology is that it doesn't stop advancing, and we can't use it responsibly.

The good news is that we've been going through rapid, rapid tech advancements the past 50 years and we're still here.

Wasn't this bridge crossed when Photoshop became popular?

The good news is that public awareness of potentially manipulated media is on the rise. Coupled with good laws, good detection tech - public awareness and media literacy is important. At Metaphysic.ai, we created the @DeepTomCruise account on Tiktok to raise awareness.

We also created www.Everyany.one to help regular people claim their hyperreal ID and protect their biometric face and voice data. We think that the metaverse of the future will feature the hyperreal likenesses of regular people - so we all have to work hard today to empower people to be in control of their identity.

I think the uncomfortable truth is that we're already there and it just doesn't matter because everything works on trust anyway. Sure, quality faked media has a ways to go before you're watching an Avengers quality deepfake porn but for groups that care to a professional degree (intelligence agencies, militaries, digital media companies, etc.) the quality is already good enough for intelligently planted media. The real limiter is access to trusted distribution channels for your faked media. A homunculi media consumer who gets their information from AP News/Reuters/Fox News/CNN/LA Times/New York Times/Verified Twitter Professionals/etc. is probably not at all in a position to judge the veracity of the material presented. If they were in a position to make such judgements they wouldn't need to care so much about whichprestigious media outlets they favor. But they do care, precisely because they know that they're gullible and so they want the media outlet that will give them the real truth (which usually means either the truth they like or the truth they don't like but from a brand that conforms to their ideals of what a news media brand should be). If CNN and Fox News both ran the same deepfaked video of Joe Biden shooting his veep with supplementary text stories and such then I wager a good portion of the American populace would believe it had happened and have serious conern.

you could never trust it, now you'll know you can't trust it

> What will life be like when you can't trust any video or interview you see because it could be completely fake?

I don't understand your point. This has been the case for a while. People were editing photos to remove people in the time of Stalin. And even before that, you can lie, write false records, destroy them.

>> There are huge swathes of people who live in their own make-believe world, like those believing wholeheartedly that the 2020 election was stolen

There are also those that wholeheartedly believe Trump colluded with Russia to win the 2016 election, or that the Steele dossier was factual.

Honestly, when people have gone so far as to redefine common words it makes it really difficult to have conversations with people.

1. Hate going from one of the most visceral and obsessive emotions that exists to being tossed around at everything

2. The advent of your truth instead of the truth

3. Constant injections of "x" into every existing word apparently?

"Womxn in Mathematics at Berkeley" - https://wim.berkeley.edu/

This is all before we get people to understand that having a discussion where some of their points might not be a strong as they think they are...somehow means you're attacking.

The world that we have created for ourselves over the last 20 years is weird.

Blackrock sells passive index funds. There's no room for an AI to make decisions there, so it probably doesn't do anything interesting.

That's not an AI problem, it's a concentration of wealth problem.

Giving that power to a person or group of people would be almost as bad.

> It feels almost like a new primitive is being discovered.

That's a really nice way of putting it. The impact on the field of computer graphics is tough to predict, but ohh my am I excited.

One thing we should be doing is supporting critical thinking at the high school and university level. Unfortunately, it seems we have been dedicated to the opposite of this for about 50 years, at least in the US.

I suspect we'll need to return to the idea of getting our news from trusted sources, rather than being able to rely on videos on social media being trustworthy.

Technically, we could try and build a trusted computing-like system to ensure trust from the sensor all the way to a signed video file output, but keeping that perfectly secure is likely to be virtually impossible except in narrow situations, such as CCTV installations. I believe Apple may be attempting to do things like this with how Face ID is implemented on iPhone, but I suspect we'll always find ways to trick any such device.

I think most of those people believe that humans will no longer exist in a "full blown unchained AI world".

What about nft-based camera recording?

Very much agreed. It just results in a terrible UX when 98% of other websites have a standard scrolling mechanism

Agree. It's a jolt to scrolling expectations

One thing I've always wondered about that, they explained where those are running, but how they still have autonomy? Are they producing they own power? How about when they need to replace hardware?

I am not saying it's impossible, but I would like to see that part being explored, even if it's in the tie-ins comics.

> videos of ARMA 3 being used as war footage

I am genuinely disappointed in the public for falling for that.

Neural Radiance Fields are a technique from the neural rendering research field, while photogrammetry is a research field on its own. However these are just turf wars and in practice there is a lot of overlap between both fields.

For example, most NeRF implementations recommend the use of COLMAP (traditionally a photogrammetry tool) to obtain camera positions/rotations that are used alongside their images. So this multi-view stereo step is shared between both NeRF (except a few research works that also optimize for camera positions/rotations through a neural network) and photogrammetry.

After the multi-view stereo step in NeRF you train a neural renderer, while in photogrammetry you would run a multi-view geometry step/package that uses more traditional optimization algorithms.

The expected output of both techniques is slightly different. NeRF produces renderings and can optionally export a mesh (using the marching cubes algorithm). Photogrammetry produces meshes and in the process might render the scene for editting purposes.

Nerfs can represent reflection, speculars, refractions.

They also are proving to be faster, more accurate, etc

The input data is the same. Nerfs have the chance of requiring less input data.

Technology to detect fakes necessarily lags a bit from technology to create fakes.

In general, you need to have examples of a type of fake to detect it.