Fandom can't decide if leaked songs are real or AI-generated
404media.co
404media.co
Media is turning from a product into a service.
The personalization impact of AI on media is going to be more profitable for the surviving media companies than static media ever was.
You'll still have many of the roles currently in media, but the work will shift from creating a singular static media product to creating a media framework that's further extended by AI products.
People have turned out to have rather binary thinking on the topic. Everything is cast as either human or AI, with relatively little consideration for the much more likely scenario where it's both in nuanced applications.
We're nowhere near saturation in meeting demand for most things as much as we are constrained in supply which tempers and limits demand. AI supplementing human labor is going to result in much more tailored product offerings at similar costs, not the same offerings at lower costs.
And any companies that invest in the latter instead of the former are going to be losing market share to those who do the opposite.
I really hope people stop trying to forecast the future of disruption against the context of the status quo, and instead better recognize the ways that the status quo is going to be changing alongside that disruption.
It's got that scammy smell about it.
I open the page for a split second, saw the 404, and hit the back button then started skimming the comments
Smooth and identifiable already (plus the deepfake video which is just crazy). Everyone can wring their hands about copyright and authenticity and royalties, but I am excited by the opportunity for creators to use AI to create new things.
Eventually, humans will get better at sniffing out fakes and get bored with simple emulation of vocal tics.
If AI can be so convincing, then your publisher won't we wanting to pay you to make any new songs if they can just pay an AI guy for a few days to crank out a new song.
I think a person or artist should own their voice, but what is the threshold for an altered ones that is different? And for 'real' human voices someone can sing and sound just like you, that's not a crime. Would it be the same for an ai-generated voice, anyone can copy any of them?
With the troubling to me decision that you can't copyright ai generated images, it feels natural to me that you could probably not copyright / control ai generated voices the same way. This is a mess, then I could take my ai generated video and voices of presidential candidates, politicians and they can't stop me maybe, because it's all got that magic adjective "ai" in front of it.
Record companies can make money off of it -- so they will support this in some way. And they will fight to make sure people can't easily do it on their own systems (which is the way it's going right now).
Last month, there was an incident where a prominent voice actor had their voice cloned for an AI parody video and the actor asked them to take it down; in response, the voice actor was harrassed off of social media for a bit.
I don't think these are as distinct as you think. Especially image-generating AI uses seem driven by the same impulse as voice cloning, virtually always "I want a picture like this picture I already have of X, but drawn like artist Y" (or "I want a picture of X for purely prurient reasons"), very rarely is it people struggling to put their own ideas into the world.
Ah, I get it, AI tools which earn you money good and creative, AI tools which don't bad.
With Stable Diffusion and ChatGPT you can argue that it's a massive amalgamation of inputs with no specific one having a huge impact, but voice cloning currently requires specific example(s).
In the case of voice actors being unable to get their voices taken down from AI sites, they can't take legal action because the IP owners have the rights to their performance.
... but I can't imagine why that'd be attractive to them.
Many of the most popular artists today lean away from heavy electronic control, and go for a more acoustic/natural sound.
I think that the AI vocals becoming indistinguishable from the real thing has more to do with the quality of the AI, and less to do with modern pop music sounding robotic.
The autotuned vocal style generally peaked in popularity a couple decades ago, and is much less common now (with many of the current top pop artists opting for more organic vocals).
we're about 3 years away from ai making 5 star sounding tracks with vocals and lyrics.
This is also somewhat of a commentary on just how banal and formulaic radio pop has gotten. Sigh.
So let's game it out...
Most of the pop hits of the last 30 years come from a small handful of songsmiths. Most pop stars already don't really write their own songs, they are more the 'Figurehead' of the whole experience. Their overall look and stage presence are the major selling point (some of them can't really sing without autotune, looking at you Britney).
So a handful of songwriters will be out of work, and nothing much will change. Mayyybe a few artists will sign away their likeness and voice to churn out album after album, hoping the public is none-the-wiser. Pop music may become still more homogenous, since part of the huge pop hits is the unexpected twist worked in that makes it "pop".
It'll be kind of a reverse Hatsune Miku, but things won't change that much. I think the appetite for a completely synthetic pop star is not there in the West - they'll have to be tricked into it. Pop stars are already arguably artificial in some form.
It would be more than that, though, if AI truly becomes as good at writing pop songs as that handful of songsmiths.
Currently, those songsmiths are the limiting factor for pop stars. As you said, there aren't many, and they can only write so many songs per year. Because of this, choosing which pop stars get to sing the songs they write is a big factor in determining who becomes successful. If suddenly AI can pump out as many pop hits as we like, that limiting factor goes away and the whole market changes.
Anything to support that conjecture?
Pop songs take off because they're marketed properly AFAICT. Writers choose from many songs, all good, producers largely decide which songs and groups will make it. Songs can be around for years before being given to a star to make into a hit.
Of course marketing now can be 'going viral', and that can be various degrees of organic and paid promotion.
To that, I'd respond "scarcity."
It's hard to imagine highly valued art (either culturally or monetarily) without scarcity.
Consequently, I expect the next few decades will see similar progressions to the past, when technology encroached on scarcity (e.g. painting after the photograph, live performance after consumer video/music playback).
Scarcity will reestablish itself in whatever guise remains technically feasible, and again become valued.
In the case of LLMs and diffusion, I expect it will be creating things that are so novel they could not have come from AI.
Hopefully skillful, deep parody and the absurd will reassert themselves, as post-JS-Daily-Show I think that's been missing in culture.
But it sure as shit isn't generic pop music, which will be the first thing to be churned out of humanless hit factories and flood the market.
It is people's shared experiences of a song that makes a song most valuable (it's not the only value, of course). Having more songs means less shared experiences, more is less.
Imagine going to a club where the DJ can play any of a million songs but only one person knows that song; compared to them playing a setlist of floor-filling bangers ...
Imagine a handful of objectively "good" songs.
Play them for a crowd where nobody knows them.
Or, in a more humorous example, imagine the Beatles played tracks only from their second album, while touring for their first album. Would audiences be disappointed?
I struggle to imagine an overly negative reaction. People love hearing songs they love, but they also fell in love with those songs for a reason. Well executed music is well-executed music.
But shared love of music, or shared experiences of music transcends this value IMO; and is largely orthogonal.
"That song we danced to", "what we sang at the campfire", crab dance for LTT watchers, 'easy for ENZ' for CSGO players, a national anthem to a nationalist, ... the music matters a little but it's only really a rallying point for shared experiences.
You are definitely right that abundance leads to a decline in shared experiences. I’m the only person I know who listens to the music that I do, but of course I treasure that music. But beyond my anecdotal experience, from sites like Last.fm it appears that listeners really began to fragment by the 2010s; with so much on offer now, people don’t necessarily listen to the same music as even their closest peers.
DJs, too, have spoken about the decline of the well-known banger when they are being flooded with hundreds of new tracks every week. Moreover, computer mixing today means those tracks might get so cut up by DJs (e.g. taking a bass line from one track and a synth line from another) that they become well-nigh unrecognizable to even the savviest trainspotters.
I disagree entirely. The Mona Lisa's value is greatly expanded by its free availability -- the one-off image on my wall is scarce, but it will never be a cultural icon, that is antithetical to scarcity.
A value of art is reflection (the mental process), but this is magnified enormously if society can reflect on the work, reference it, abstract from it, view it and develop freely from it.
Great pop songs are great because everyone has heard them and we have shared experiences around them. There is no such thing as a scarce pop song.
(This is one of the great crimes that arise from copyright terms being great than a couple of decades, society doesn't get to riff on the important icons of its recent past.)
Arguments against against scarcity that point to copies of a singular work seem circuitous. There are innumerable copies because the single, canonical version is so valued.
Similarly, great pop songs are valued because everyone knows them, which is why individual song value peaked during the dawn of mass distribution (CD, then early digital) but before the market was flooded with volume.
The scarcity in the pop sense is the limited number of songs that everyone knows. (Largely through sophisticated media campaigns and forced radio/channel placement, but I digress)
>Arguments against against scarcity that point to copies of a singular work seem circuitous.
It's a good point, but I think it misses something. It is experiences of the effect of an artistic work that matter, not the work per se. Experiences of the work are 10-a-penny -- tens of millions have probably seen one of the 'originals', billions have probably experienced the effect of the artwork (which I will deftly avoid defining ;o)) through photos, copies, videos, imitations, and emulations.
> To that, I'd respond "scarcity."
What is your definition of valuable here?
If you're referring to value to culture/society, I think you're very far off-base. The most culturally valuable artistic works are ubiquitous, the opposite of scarce. Art isn't really able to have any culture influence if it only impacts a small number of people.
If you're referring to monetary value, you're also dead wrong lmao. Look at the top 100 most-paid artists of the last decade, and tell me how hard it is to find and appreciate their entire artistic catalogue for yourself.
The argument that scarcity = artistic value doesn't have any basis in fact, and is the sort of thing that would only be shilled by someone trying to con you into buying an NFT.
> It's hard to imagine highly valued art (either culturally or monetarily) without scarcity.
hahahaha what? Compare the monetary and cultural impact of that one "ultra-scarce" Wu-Tang album (monetary: $2m, cultural: none) to the impact of Taylor Swift's last album, which is available on every streaming service (monetary: $200m+, cultural: very high)
> The most culturally valuable artistic works are ubiquitous, the opposite of scarce.
Not so. Copies of those works are ubiquitous, but there is a singular, definitive work.
Name me a handful of world-famous works for which there are multiple, almost-indistinguishable but distinct copies.
The Mona Lisa has a few original alternates, and yet they pale in value to the famous one. Which itself, ironically, became popular famous mostly through being stolen (scarcity).
> Look at the top 100 most-paid artists of the last decade, and tell me how hard it is to find and appreciate their entire artistic catalogue for yourself.
> [Once Upon a Time in Shaolin] vs [Speak Now (Taylor's Version)]
Total artistic renumeration, especially in the modern period, is dominated by distribution volume.
But if we're talking about single work valuation, the Wu Tang album costs $2M.
Taylor's album costs $15.
That's the premium for scarcity.
When referring to recorded music, this isn't a distinction that has ever actually mattered in the real world, just a fiction made up to shill NFTs.
Are you going to pretend that anyone actually cares about a "singular, definitive FLAC file" that all of the streaming services' FLAC and MP3 playbacks are based on? This is pure fantasy, the copies are the same thing as the original piece.
The idea that Mona Lisa's (or any other artwork's) cultural influence comes from its scarcity is hilarious. Literally anyone can visit the Louvre and appreciate it for themself. Do you think it would have anywhere near as much influence if it was hidden behind closed doors and only 1 person was able to see it?
> But if we're talking about single work valuation, the Wu Tang album costs $2M. Taylor's album costs $15.
Last time I checked, the sum of revenue from their discography is how artists and labels get paid, not based on the maximum amount that 1 person is willing to pay.
Speak Now is a single work, and it generated like 100x as much monetary value as Shaolin (with like 10,000x as much cultural impact). And those estimates are extremely conservative, when you consider that you can tour and sell merch off an album that people can actually listen to lol.
This doesn't provide any more credence to the falsity that art's scarcity is the source of its value (when overwhelming evidence proves that the exact opposite is true)
Those replicas and recreations are recreations of a... single, scarce work.
That's famous precisely because there is one original.
But if that's not true, it should be possible to point to, say, a series of similar paintings or musical compositions that are all famous.
Generally, that's not the case though.
Because people want one thing.
The one Mona Lisa. The one officially-blessed Taylor Swift album. The one version of Beethoven's Fifth.
Complexity and variety confuses simple people and the market.
> The one officially-blessed Taylor Swift album. The one version of Beethoven's Fifth.
Millions of people have bought vinyls/CDs/MP3 replicas of these albums. These are not scarce works. The demand is not for the scarce original (who even knows what that means when it comes to music), it's for a faithful recreation of the artist's work and creative output.
On top of that, nobody who listens to these albums really just wants one album... they want to listen to music that moves them. Taylor Swift fans will listen to other albums that she puts out, and music from other artists that they enjoy. When Beethoven released his 6th symphony, it didn't make his 5th any less valuable to those who enjoyed it.
Because for most people, the value of art comes from its intrinsic beauty, not its "scarcity". Unless if you're someone whose only attachment to art is as a vehicle for financial speculation, which is sad but unfortunately common in NFT circles, etc.
> But if that's not true, it should be possible to point to, say, a series of similar paintings or musical compositions that are all famous.
Have you heard the term genre before? It's literally a word for a group of similar musical compositions. Most genres contain plenty of similar compositions that are all famous.
If people enjoy a work of art, there is almost always other popular art made that's similar to it. Turns out, you get bored if you just listen to one album over and over again.
When it comes to art, most people enjoy variety, it doesn't "confuse them" lol. This is clearly reflected in the market, as there are hundreds of unique genres of music, each with thousands of unique artists.
The only meaningful difference between it and the one in the Museum of Modern art is that it was painted by the Van Gogh directly. And sure it has a hefty price tag attached, but everybody who buys a copy buys it because they like it.
Growing up music was rare, but with access to Spotify I can listen to so much more music. That has not meant I appreciate music less. Sabatons "Primo Victoria" is still going to pump me every time I hear it and Bethovens ninth is as uplifting as it ever was, no matter how many times I listen to it.
- content marketting is already AI writing, it's just humans churning through the SEO optimized algorithms.
- Mass digital stock photograph is AI generated "art". People that do that for a living have an exact formula for reproducing images that will sell en masse.
- As you point out, a good chunk of generic pop music is churned out by musicians following algorithms behind the scenes and then manufactured to look like the work of a public performer.
The parts of our content work threatened by AI are already cold, sterile, and machine generated.
The AI revolution is actually just the next increment in automation of all these things.
I don’t really agree with this for two reasons:
1. It is understood that vocal synths (e.g. Hatsune Miku) are not humans and that humans are not vocal synths. There are also humans covering songs originally sang by synths and vice versa, but everybody still understands what’s going on. (Heck, there are even synths based on the voice of real human singers, e.g. KAFU; it’s still understood that the singer and synth aren’t the same thing.)
2. Vocal synths are not fully AI and require a significant amount of ‘tuning’ to sound good (with various producers generally having their own ‘sound’ due to differences in tuning style). They are also just a voice, they don’t generate any instrumentals. Therefore, a human-produced song sang by a vocal synth is still a long way off from a fully AI-generated song.
> Hatsune Miku is a red herring I think.
I don't think we'll get more of these.
> I expect we'll see more Milli Vanillis too.
Completely agree, and definitely what I was saying.
> Possibly there will be curated such avatar/music generators, and more than likely, many of them will be naked.
I would make the case that pop musicians would be less popular if they freely got naked. The teasing of maybe being able to see a bit of rarified (and real) flesh is definitely a big part of the equation, which you're not likely to be able to reproduce with what people think is an AI.
If the system knows you are a prude, then they won't get naked for you. They'll get very good at learning "how much is too much" for any given person, for any given "what" (skin, private messages, flirting, etc).
Human pop stars can't have relationships with every fan. AI pop stars can.
I don't know if it will go that way, to be honest. But certainly, the film Her does show us what that might end up looking like. (Even though she was just a disembodied voice)
Edit:
I am torn about how I feel on the matter but here’s a story worth sharing: I stayed up all night the day before the release of a much-awaited book in a series to read, leaked blurry jpeg by leaked blurry jpeg, hundreds of pages of the book that had just appeared online (I also went out and bought a copy the next day).
While reading it, I knew there was no guarantee it was “real” and it could have been an elaborate hoax but the quality of the storytelling was such that I didn’t care (at least in the moment) if it was fan fiction or the real deal: it was good and I wouldn’t have regretted my time or enjoyed what I had read any less if it turned out to be a “fake.”
If what I'm getting is just a synthetic, semi-random découpé, then that is lost. Similarly, looking at randomly generated maps can be fun, but it's a pretty shallow fun unless there's a real territory represented there.
That said, it can also be interesting to look at the machinery that's generating it, in the same way I enjoy videos of how things get made. But it's a different kind of interest and engages different parts of my mind.
But unfortunately now that this type of comment can be the rambling of an LLM, I don't want to put in the effort when it could simply be a mechanical fever dream...
I find the story of a human who makes an excellent fake pretty fascinating. There are noteworthy examples of forgery-as-art throughout history. I think art, as a concept, is a big enough space to house both things.
You wouldn't be wrong, you would just be post-modern. As a post-modernist viewer, I decide the context that I want to place media in.
It certainly comes down to perception, interpretation, and a blind assumption though, right? Before generative AI, the blind assumption was that some human made a piece of art. That might be changing now, but if you didn't know (and couldn't know) a piece was artificial, feelings may still be evoked and you might still infer some meaning from it.
With that said, not all art need be about human connection. I still enjoy mindless dubstep music on occasion despite not caring one bit about the artist or any meaning behind it (if any even exists beyond "it sounds cool"). The only value there is my enjoyment of it, but then again, I think enjoyment is the most valuable aspect of any piece.
Let’s put AI aside for a minute, then. When a suspected {Van Gogh, Vermeer, Picasso, etc} is found to actually be from a temporally coincident painter, perhaps one of his pupils, why can you no longer do that?
I don’t really care about having some equivalent “conversation” with an LLM.
There are likely thousands of people who can paint extremely high quality Monet-alikes, but they weren't there at the beginning of the impressionist movement.
Granted, there's also name weight and limited supply due to creator death, but that "first!" can't be ignored.
The artistic-economic tangle around originality wasn't something I'd thought about deeply until I saw some of Warhol's "X of Y" prints in person. Because he used an inexact process, there are subtle differences from frame to frame, which to me begged the question of whether a specific selection of frames and composition of them together was sufficient to re-establish a unique, original artistic work.
Let's imagine Hitchcock's compatriot releases a film that is Hitchcock-esque in every way it can be (camera, lighting, themes, sound... the whole package), while featuring a novel plot.
Then let's say Hitchcock also releases an equally-equivalent film to his first work.
There is no non-plot innovation in either work, and the plot is as similar as can be while remaining distinct.
What are the values these works could likely command?
To me... I think the Hitchcock work would be more valuable, by virtue of his name, which in turn originated its value from his first masterful and original work.
His compatriot's film would be valuable. After all, Hitchcock's original work was acclaimed, appreciated, and popular, so it stands to a reason an extremely similar work would be as well.
But I can't see it approaching the value of Hitchcock's second, despite them being functionally identical artistic works. It's instead discounted by the lack of Hitchcock's name, itself valuable from the link to his first, original work.
Interesting thought exercise!
The follow-up would be what would happen if the market were flooded with first-alike films, either from Hitchcock or his compatriot!
I'm not denying the appeal of "authenticity" but I can't for the life of me understand it.
I think what is new is that there are certain signals in text and creative work in general that would ordinarily demonstrate that some care and effort and resources have been used in the process of creating it, which makes it seem more likely that it's legitimate, and that now those signals are no longer valid. I think that people will rather quickly adjust from default credibility to default skepticism for text online, and will rely more on "provenance" for determining what to trust.
> no one is really sure what’s real, what’s fake, whether they’re being scammed, or who or what made the songs that they’re listening to.
This is close to the definition of technologically induced societal madness Jaron Lanier sketched in an article about a year ago.
Pop will eat itself. We always knew that. Pop is the spirit of whimsical, ephemeral, self-devouring, involuted art. It's to be expected.
But so much else has become like Pop; news media, food, medicine, clothing, even the cars we drive... At some point AI combined with appalling cyber-security and our over-reliance on tech will start to tear all of these things apart. This kinda started in the pandemic, even before AI. Is it really government advice to drink bleach? Or a terrorist plot?
Philip K. Dick would not know how to write this stuff.
These guys have come out of the gate strong. I think they only launched in the last couple of weeks.
Remains to be seen if they can sustain it.