People really don't see where this is going, do they? Replicas are the last of their worries. Actors and singers should be more worried about being replaced by perfect, completely artificial people. It's like when extras were worried about being replaced by body scans. No, ThisPersonDoesNotExist will replace you all, silly.
Hollywood productions have already begun shifting back toward using extras instead of CGI actors.
It turns out it's a lot cheaper to just tell a real person to do something different in real time a bunch of times over the course of a day then it is to tell animators to animate a bunch of CGI people differently a bunch of times.
Extras don't get paid that much to begin with (a fixed, flat rate which varies depending on the type of production i.e., commercial vs TV vs movie). Digital artists don't get paid all that much either, but the disparity is large enough that you actually lose money (compared to hiring a real-person extra) if you need to re-animate a CGI extra more than once. And quite frankly, given the limited VFX budgets most productions have to work with, it's better to spend the VFX budget on actual visual effects than on background filler. Otherwise, you end up with Dr Strange 2, Secret Invasion, and the Book of Boba Fett...
Not generative AI, if that's what you're implying. Just because companies are offering it for free right now as a loss leader means they're going to keep it free when the accountants come home to roost.
An extra costs $100-$1000. A digital artist's time costs about twice that. A license for a generative AI image will cost at least 4x what it costs to pay the extra, and that's not including the subscription you'll need to pay to access the AI service.
Wait until the LLM interface to the behavioral model of your virtual extras means you can tell the CGI actors to do something different a bunch of times without having to have an animator involved... and they don't need you to provide craft services and wardrobe or complain about being stood on their feet for hours.
I'm not sure anyone alive now will still be alive when this magical feat happens.
After all, people were saying that robots would talk like humans and be indistinguishable from sentient life before you or I were ever born.
Decades later, even with CPUs with more than 1,000,000,000,000x the processing power it's still news when a company releases a robot that can walk in a straight line without falling over, which is something that human babies can do before they're able to figure out how to control their own bowel functions.
The Hollywood writers got major protections against AI exploitation in their new deal, and surely SAG will follow with their own version. Hollywood can’t exist without its writers or actors and they’ve made it clear to the executive class that they will not take these changes lying down.
If generative AI can create a whole movie from scratch won't that mean the end of Hollywood as they would get replaced by OpenAI and co? Just because Hollywood won't do it does not mean no one will.
Unions protect labor from the forces of capital who take advantage of them. See: the recent WGA strikes and the protections from AI work that they now work under.
I think the point is it becomes irrelevant how good of a contract hollywood actors get if hollywood is cut out of the picture entirely by smaller independents using AI.
People like celebrities and personalities. Perfect and artificial is the opposite of what they want. People like Taylor Swift because she's Taylor Swift, not just because she's a great singer. (Not to mention that, for music in particular, live shows are a thing.)
There will definitely be a role for fully AI characters when the technology gets there (people love vtubers and vocaloids already!) but I doubt they would ever really replace human performers.
> Taylor Swift is a generated personality designed by a branding and marketing company
Interesting. When did she turn from an actual person into a generated personality, and how? And why can't we get more of them on tour to lower the ticket prices?
people want authenticity, and ai is just not gonna fly
sure - you can suggest that Taylor is the product of branding/marketing. its cynical and somewhat true. but aside from that, she's an artist. in the deepest meaning of the word.
people are not going to resonate with a generative ai. instead - people will continue to resonate with actual art and artists expressing actual feelings
They _want_ that. But they rarely have that. I argue most if not all music artists in particular, are "fake" already, in the sense that, like you say, branding+marketing, and scripted stunts and controversies, etc. People are already fooled by this in today's age. I think there might be pushback, but I think in the end people will accept/be fooled by ai generated artists.
Or maybe the two truths can coexist: some people will worship SpongeBob and others Taylor Swift, all will attend both live and virtual shows of both, with the only negative consequence that SpongeBob will now eat (more of) the cake of some other human artist. It's still displacement just slightly less dramatic. As for myself, I'm a simple man, going only to cellar live shows which cannot be replaced by AI as the very reason of being there is... being there with the artist. So the AI seems to only threaten the pop industry, not undeserved if you ask me.
That whole train of thought not only requires that long-form virtual actors are within reach (we’re close but not there), but that all this recent generative work can escape having a signature at scale.
There’s ZERO evidence of that right now and little reason to expect it.
What’s dramatically more likely is that generative AI will be an incredible new medium, disrupting current arts and reshaping almost all art markets the way photography and sound recording each did a few. And its disruption applies to film, illustration, music, coding, writing, etc
That’s no small achievement and its an exciting thing to witness, but it’s not the same as replacing the arts that already exist. Arts have shown themselves to be cockroaches, that simply adapt themselves around these sorts of disruptions.
There will still be HUGE markets for human (and hybrid) arts even as they are smaller and differently shaped than what we saw in the 20th century.
We're much closer for virtual "voice actors" particularly with the advent to do emotional transferrance to map inflection/feeling from a simple recording onto a custom trained TTS.
Here is evidence: large crowds of extras have already been replaced by CGI. This is a list of 10 movies with the most number of extras: https://www.tasteofcinema.com/2017/10-movies-with-the-most-n.... The most recent entry on the list is from 2003 (~20K extras). Why would anyone hire 20K people for a movie today?
> Why would anyone hire 20K people for a movie today?
To make better art. I've seen most of the films in that list are they're incredible films that stick with me and give me a sense of awe today. Watching 1000s of real life people in a scene just creates a sense of spectacle that is just not achievable with current CGI - the brain is just not fooled or impressed like it is when you watch those old films and see the amount of work and effort that must have gone into them. Sergio Morricone hired the Spanish army to build the cemetery at the end of The Good, The Bad and The Ugly and it took them several days to make. Nowadays it just takes someone a few clicks of a mouse but it fucking shows as well. Maybe with AI we'll actually get to a point where the brain can be tricked because it gets the lighting right, but at the minute it just doesn't and the films are worse because of it.
It's even more awesome when you watch the film in it's totality because it builds up slowly, with bigger and bigger scenarios until it culminates with this.
And even in potato cam resolution, how impressive is this scene from Cleopatra:
They just hit different when they're actually real.
What's hilarious is that the execs still don't fucking see it. Christopher Nolan is one of the only ones out there still pushing for practical effects and he rakes in the money at the box office more consistently than just about any other director other than Cameron or Spielberg. It's because the films are actually worth watching. Before Oppenheimer, what was the last big film? Top Gun, which also used a ton of real life flight footage.
If I want to watch an animation, I'll watch an actual, awesome animation that takes full advantage of the medium like Into The Spiderverse or something from Studio Ghibli. If I want to watch a movie, I want to watch a real flesh and guts movie. Yet 99% of the stuff Hollywood puts out is some weird hybrid shit that is the worst of all worlds.
If Hollywood starts making better art, people will start going back to the movies. But they've got greedy. In any other industry, if you made a product, and you got 100% more for it than what it cost to produce you'd consider that a success. But these arseholes don't. They want 2x, 3x, 4x, 5x and they want to milk it for all it's worth and turn it into a fucking "universe" until it just collapses into a heap of shit and no-one even remembers why they liked the original.
Oh I don't argue about art. Most of these decisions are driven by financial constraints, not the artistic vision. And this is not going to change.
Christopher Nolan can afford extreme artistic choices like not showing ads or previews before Oppenheimer. That's such a rarity that I wouldn't make any extrapolations even in today's world. Betting that gen AI will result in an alternative type of art and not just replace multiple roles and production techniques in existing industries is not something I can see happening considering that professional art is business.
E.g. we didn't see CGI creating an alternative movie industry that exists in parallel to non-CGI movies. CGI is used by everyone, including artsy independent creators.
I will forever love Studio Ghibli animation. But if in the future the same level of expression can be achieved with CGI/gen AI - I just fail to see them not switching their tools (possibly in the post-Miyazaki world).
> What's hilarious is that the execs still don't fucking see it
Why would they? Very few people can notice and appreciate the level of detail you are describing. And I am not even sure we will never be able to replicate it computationally. In 10-20... 50 years? Don't see why not.
Yeah I don't disagree that if the AI can get it spot on which I think it will be able to do eventually it will take over. It will take over even if it's not spot on, because the bottom line is all that matters to the major studios now.
> Why would they? Very few people can notice and appreciate the level of detail you are describing
People might not consciously appreciate the detail but I think almost everyone subconsciously appreciates the overall vibe. Everyone can watch that The Good, The Bad and The Ugly clip and see there's something special about that location.
The reason I feel like the execs should be able to see it is that they're cutting off their nose to spite their face. If you watch a movie and come out seeing like you've seen something breathtaking, you're going to tell your friends and ticket sales are going to go up, not just for that movie, but for future movies as well. Whereas if film after film after film is a dud, which seems to be more common for me than not nowadays, you stop watching films and spending your money on films. For me this has already happened - when I was younger I used to go to the cinema once every week or two. The state of cinema is so bad now I might only go once a quarter.
So the execs take these cost cutting measures, but they have the overall effect of losing them money in the long term. The film may end up pulling in 200% over production compared to 100% if they'd have used practical effects. But that "missing profit" isn't missing - they're instead spending it on raising their overall brand value and reputation by offering up a higher quality product.
Like the last film I watched at the cinema was Gareth Edwards The Creator. He'd built up a little bit of brand value with the original Monsters and Rogue One, but The Creator was so monumentally, insultingly bad I don't think I will ever go watch a film he's made again.
I do agree that soon we'll have best movies with human actors, best movies with some AI actors, best movies with all AI actors. And then the academy will resist movies with AI actors, then maybe have a special category before they give up because all the movies have them.
I really do believe this. software will eventually change the very nature of human language and hence of human thought
I picked up this line of thought from an academic Walter J. Ong who wrote a book "Orality and literacy" in which he argues that literacy (written word) changed how we think.
so I just extend that reasoning. I think that software constitutes a larger change in humanity's thought than the printing press; but that's like my opinion man
I wouldn't say they're being silly no. Think of what is going to happen when anyone on planet earth can make any kind of photo realistic propaganda materials they want. Imagine somewhere else on the planet with tensions between two communities like Israel-Hamas. Now imagine you're someone with an agenda. You can now create videos and news reports in the blink of an eye showing one faction murdering some kids from the other. Target some unstable "pizzagate" individuals with the videos and then wait for one of them to go on a vengeance rampage with the hope that it actually sparks a real riot or war.
Or imagine there's a presidential election and one candidate wants to take down the other. So they buy your psychological profile from some data broker, then create a deep fake video of the candidate reading a speech tailored to make you dislike them.
You don't even have to go so large scale. Imagine you're a teenager and you're mad because someone else is making moves on the person you fancy. So you decide to create a video of this love rival kicking a dog outside the school and upload it to Youtube under a fake account and then spread it round the school.
It will get to the point where you cannot trust anything you see on the internet anymore. Laws are absolutely going to have to be created and rewritten to take into account this new reality. I strongly suspect all social media accounts will need to be tied to ID numbers like passports or social security at some point within the next 20 years.
Your first two examples are the only ones that are consequential and you don’t need AI for either. You could do all that a century ago and you can be 100% sure that intelligence agencies have.
> Your first two examples are the only ones that are consequential
Er no. That poor kid might end up getting socially shunned at best and lynch mobbed and killed at worst. I'd say that's pretty fucking consequential. And that's just one example of an infinite amount of scenarios. Every "big" event is just an aggregation of many individual stories.
> and you don’t need AI for either.
You are massively underestimating the scale of change. In the past you would have needed an entire team working many hours to create one short video of high enough quality to fool people. Now anyone, anywhere, will be able to create them in an instant. Not only that, they will be able to create videos tailored to an individual and delivered to them for maximum effect. That is 100% only enabled by AI.
> You could do all that a century ago
Television wasn't even invented a century ago. You could make up some fake story and put it in the newspaper, but people trust their eyes more than words, and they're also moved to action by
them more as well.
And no, a government absolutely could not run any kind of deception campaign in 1923 anywhere close to what they will be able to when this is perfected. A motivated sole individual with a bit of capital to spare, let alone a government, will be able to run a deception campaign thousands of times more effective than any 1920s trickster's wildest dreams.
The technology inherently tunes into commonalities that associate its training data with incoming prompts and contexts.
An operator can push this pretty far with elaborate and carefully designed inputs, but the system is always going to be gravitating towards the patterns it represents most well.
When you or I look at our first bunch of outputs, we don’t immediately pick up on the nodes it likes to gravitate towards because there are many many of them, and we meed more of our own training data to become better discriminators.
But as these tools become ubiquitous, people will start to get to get this inkling sense of when something feels a little AI-ish. You can already see people do that with ChatGPT and SD/Midjourney output that wasn’t carefully produced. You probably do it yourself.
So far, there’s no real way to conceive of these technologies as not being subject to that kind of signature.
Like you and me, they’re still capable of incredible variety, but also like you and me, each model will have its own signature that can be more or less discerned.
For the arts, that kind of signature means that there will be a continued attention to human artists alongside these systems, because they all bring their own recognizable qualities and aesthetics.
I mentioned this in the thread about the RIAA getting worked up over this:
Should that be illegal, as long as the result isn't described as being the real person? To repeat my analogy, say I want to cover Guns N' Roses. I can hire a guitarist that could copy Slash's guitar parts (but can't say it's Slash playing them). Is that fundamentally different from getting an AI to sing Axl's vocals (as long as I don't say it's the real Axl singing)? It feels different, somehow, but I'm not sure that's philosophically defensible.
Help me think this through. Why are AI replicas that don't claim to be the original different from a cover band or an impressionist?
Thinking out loud, I think the governing aspect of this is fidelity. Cover bands, and impressionist are always going to be delta away from the original. Current-gen replicas can pass the generic test of "is this indistinguishable from the original by laymen?".
This gets especially messy with movie actors. Specifically -several extremely high-profile movie actors _specialize_ in the sort of roles they take to project a certain persona, and be _the_ guy for that role. It's also known within producer circles, that a single actor can move the sales needle +-~$X0M revenues; which is part of their valueprop. Present-gen replica tech poses an existential risk both for brand building, and exclusivity of these actors.
Just doing a little googling, it looks like cover bands actually so need to license their songs, or the venue does, and if they want to record their covers they need a license for that as well.
That seems more like a “let’s not sweat the small stuff” type decision to me. It seems like exactly the sort of decision that would have to be revisited if huge industries got built on AI.
I mean, ultimately if you had a “virtual musician” that could do “live performances” of any cover with 100% fidelity on demand while running on your phone, I think we wouldn’t treat that like a human cover band. Doing so would just kill off the idea of music IP (which, I’m definitely not going to defend the current mess that is IP law, but people have been arguing about it for a long time, so I suspect we as a society won’t just toss it out now).
It also occurred to me while writing this that this sort of virtual musician wouldn’t even really be like a cover band, because they don’t even play at a venue. They play just for you. I suspect there are no laws, at least being enforced, that cover what sort of songs a musician can play in private, in their friends’ houses. So, I think this is really going to break our intuition.
Is there a reason that this isn't covered under existing Personality rights (which are, admittedly patchwork in the US[1], but do generally exist in some form throughout the union)? I can't make a video game with Brock Purdy as QB, without paying, so a deepfake of him playing football is probably also covered?
1: Indiana has the strongest in the US. The name, image, likeness, signature, photograph, gestures, distinctive appearances, and mannerisms of Woodrow Wilson are still protected in Indiana (100 years after death, and he died in '24).
How big a deal is harmonizing things across state lines?
If no federal constitutional dictates are being violated does harmonization just amount to a convenience for big companies, at the expense of local self rule as per the constitution?
Worst case you end up with multi-nationals bribing one set of federal politicians to get national laws enshrined. One stop corruption.
There's a constitutional question. Congress does not have the authority to create new forms of intellectual property. Copyrights and patents, that's it.
That's why the US doesn't have database copyright.
States can go further. States have a "general police power" - they can make almost anything illegal.
it can recognize any forms of property including IP
the stuff about copyrights and patents isn’t mentioned in the constitution, those are forms of IP Congress passed as instructed by the article of the constitution to promote arts and science by granting temporary monopolies
Sigh. Yes, it is mentioned in the US Constitution.
U.S. Constitution, Article I, section 8:
The Congress shall have Power To ...
Promote the Progress of Science and useful Arts, by securing for limited Times to Authors and Inventors the exclusive Right to their respective Writings and Discoveries;
That's where copyrights and patents come from. It's an enumerated power, and Congress is limited to what's written there. Congress can't create entirely new kinds of intellectual property. See Feist vs. Rural Telephone.
yeah we’re quoting the same article. we agree on that. and new kinds of IP for a temporary monopoly is not the argument. That article is only about temporary monopolies.
outside of that, congress can recognize and regulate any aspect of interstate commerce, including the trade of assets it didnt previously know about or recognize. which is a different enumerated power.
Not quite true; it is constrained by the constitution's use of the term "Writings and Discoveries" which is why copyright requires the work to be in a "fixed form"
the regulation of interstate commerce is an unlimited dictate thats also part of the constitution, as all commerce is interstate commerce for the last century there is no federalism left in this country outside of what the federal government merely tolerates from the states, and in the parallel judicial systems against persons
there is no prioritization of “local self rule”, an ethos of the original republic and 10th amendment, the federal government has subjugated all states since then and throws them a bone here and there since it technically derives support from the collection of states
with that in mind, this specific industry of image and likeness releases has been long overdue for one framework to rule them all. the issue isnt big companies, its small operators and enthusiasts that dont really know what text to put in their release forms and always have been operating in multiple states.
I also don't understand -- from the article, it's not clear what is changing at all.
My only guess is that maybe this is applying to non-commercial use as well?
Like maybe this is intended to make deepfake videos illegal altogether? Posting them to YouTube or social media or anything? If they're not presented as parody, but rather being claimed as real-life footage?
But the article doesn't say that or even imply it.
If this kind of small-minded anti-intellectual and anti-science act had been put in place in 1900, the "No Fakes Act" would have meant no phonographs or recording of music. You must listen to actual music, not fake replicas recorded on cylinders or disks.
Just like we have a revolution in music because of recording technology, and a revolution in art because of photography, we will have a revolution in movies because of replicas.
People like this want to stop progress at all costs. Society should wither away, never changing, never adapting to new technologies, never discovering anything new. Until some country that isn't so small minded ignores these arbitrary rules and eventually takes over as the preeminent scientific and cultural power in the world.
We have always had the right to impersonate people and clone their voices. Since the dawn of time. This is taking away yet another fundamental right of everyone in order to commercialize it for massive companies. Just like copyright became this multi-lifetime shackle around the necks of society recently, this too will further restrict the development of our culture.
AI grifters realizing why Meta paid celebrities to clone their voices for their AI chatbots.
These same celebrities are ready to take anyone to court to collect their money for unauthorized cloning of their voices for commercial purposes and it costs them almost nothing with a greater chance of them winning.
This just means that companies will only work with people that agree to have their likeness be replicated by AI. I really think that we'll start to see celebrities that are entirely artificial within 20 years.
Every day draws us closer to The Right to Read world.
If someone is copying and selling your work as created by you, that is one thing. Sounding like you, well guess what, your voice isn't copyrightable, and really in I see no world in which we'd want that.
Moreso, this is going to end up in lawsuits like we saw in the open source world in the 2000s in the SCO vs Linux debacle. "That voice sounds kinda like me, so it must have been trained on my data!"
Trying to legislate digital content generation seems like a waste of time and resources. I am not much of a libertarian but this is one area where I agree with minimizing laws and regulations.
There's already a kpop group that goes in this direction https://youtu.be/1wGOHbcQKIc?si=hKti2i2gKq_pQ16T
Obviously not 100% ai generated; there's motion capture/ modeling and a real singer is recoding, but yeah. Going in that direction imo
It's a tough issue, but I believe that there's a need for some sort of regulation.
Joe Rogan has mentioned a few times that people are using a deep fake of him to sell penis enhancement products.
Should that be ok?
What about character assassination stuff? Should I be able to make a damaging deep fake of a person I'm competing with for a position, post it to YouTube, and then arrange for a link to be sent to HR?
Should I be allowed to make a deep fake of a rival for a romantic interest, putting them in a compromising position?
I'm not sure what the solution is, and I'm not a fan of knee-jerk legislation - but in this case I understand the need for something.
I believe in self-ownership, protecting my rights to my own image and voice seems reasonable.
Of course, as always, the devil is in the details. As others have posted, this could end up being messy in court - how similar does an image need to be to be considered a copy? How do you develop objective standards for that?
Isn't there already legislation around this? Random people can't use Micky Mouse to sell penis enhancement products either. Seems like the same laws should apply for deepfakes.
Also, fabricating incrimination emails or writings is defamation. I assume making a damaging deep fake would also apply.
Well Mickey Mouse is trademarked. The earliest cartoons fall out of copyright next year, but Mickey Mouse himself will still be an identifiable mark of the Disney Corporation and therefore you can sue.
A deepfake is a recreation of the likeness of a person though, and you can’t trademark your likeness so the same laws wouldn’t apply. You would need different laws.
Oddly, no, as long as the product actually does what it is advertised to do, it is not false advertising.
Currently, it's a misappropriation of likeness issue, which is currently a state-level civil action in some but not all states. The point of the proposed law is to make stuff like this regulated by a single, federal law instead of a patchwork of state laws.
No, just as using a lookalike or (for radio ads) voice actor with a deliberately similar voice and phrasing wouldn’t, unless they actually claim it is Joe Rogan endorsing the product.
Bette Midler won a lawsuit over a soundalike in a Ford commercial. Ford first approached her to do the ad and when she turned them down, they hired an imitator.
