Without Content, OpenAI’s Sora and Generative AI Tech Are Useless. So Shouldn’t Artists Get Paid?
What do generative AI tech “startups” OpenAI (the creator of ChatGPT) and rivals Anthropic, Cohere and Perplexity all have in common? They are privately held companies valued (or soon to be valued) in excess of $1 billion, making them “unicorns” in the lexicon of Silicon Valley.
Here’s another thing they all have in common. They built potentially transformational technology that is essentially useless without the content on which it is “trained” to deliver its results. And they get that requisite content by scraping the vast Internet, sucking up millions of copyrighted works in the process — all without consent.
These generative AI leaders — and other leading genAI tech innovators well on their way to unicorn status, such as music-focused Suno and video generator Haiper — defend their relentless non-consensual “training” practice as just being part of answering to the higher calling of social progress. So they respond to any qualms about copyright infringement by claiming fair use as a defense.
But is their training — and what it represents — really “fair” when Silicon Valley continues to rack up billions from its dependence on massive amounts of copyrighted works, while the artists and media and entertainment companies responsible for creating those works continue to be pummeled by Wall Street and face financial uncertainty?
Where are the media and entertainment unicorns that fuel this relentless march in the name of tech progress?
The issue of generative AI infringing copyrights is playing out in the courts right now, and so far Silicon Valley is winning. Federal courts that have addressed the issue have essentially concluded there can be no actionable infringement when an artist’s copyrighted works are included in a training data set of millions. The rationale seems to be that generative AI’s end result – its output – will not be substantially similar to the allegedly infringed work and, therefore, cannot cause direct harm to its creator. It’s kind of an “only one small screw in a machine with thousands of parts” rationale that, upon first blush, sounds kind of logical. If that machine spits out plastic fasteners, they look nothing like the screw.
But wait a minute. Doesn’t the machine’s manufacturer need to pay for that screw no matter what that machine makes? Of course it does! The screw is an “ingredient” in the overall manufacturing process. It creates value – and it also took value (and resources) to make.
So let’s zoom back out to the macro issues at play here. Isn’t Big Tech’s “fair use” refrain just one big justification for another massive transfer of wealth from artists and creators to big tech companies like Apple, Google, TikTok and Meta that have built their multi-billion and trillion dollar valuations on the backs of content creators for decades?
Enter Suno’s latest music-creation version
Let’s take hot new generative AI music startup Suno, which launched its jaw dropping latest version last week. Just by typing in a few words as a prompt – for example, create “a sad folk song asking why there aren’t any media unicorns, because tech unicorns are everywhere” — Suno generates professional sounding music tracks in seconds, complete with song title, lyric sheet, and thumbnail artwork. You can listen to that song, which Suno automatically titled “The Song of the Unseen Unicorn,” here. It’s scary good, and its lyrics are scarily apt.
How does Suno’s AI music producer do that? I certainly don’t know its special tech sauce. But I know that its real secret sauce is the wide world of copyrighted songs and recordings that it apparently sucks into its training vortex. From what I can see, the music it uses for training hasn’t been licensed, despite the company’s recent “Trust & Safety” message that rather obtusely states that it doesn’t “recognize references to other artists” because they “are not here to make more Fake Drakes.” I’m quite sure that only means that Suno generated tracks will not replicate the voice of any specific artist. It doesn’t mean that Suno doesn’t include hordes of copyrighted works in its training data set.
(I reached out several times to Suno for comment but never heard back.)
If my understanding is correct — and I’m quite sure it is — isn’t this essentially, in the provocative but completely defensible words of one music writer, just one big heist? If all of us can simply create high quality music in seconds and with just a few keystrokes, then won’t there inevitably be less room for human created tracks? After all, Spotify and other music streamers are already being overrun by AI generated tracks.
The simple answer is “of course”! Some material fraction of market demand for human-generated music will be lost to our new artificial composers that received their music training for free, unwittingly from us humans.
Some will argue that this is nothing new because artists have always been inspired by and “copied” the art of others – the only difference here being that we are substituting machines for humans. But wholesale vacuuming of entire copyrighted songs and recordings is fundamentally different than artists building on top of the creative building blocks of others before them. Individual artists aren’t creating entire systems that are designed to enable mass market commercialization by millions. And oh yes, let’s not forget that copyright law is there to rope in individual humans if their “inspiration” crosses the copyright line.
Many others will point out — correctly this time — that new technology has disrupted the media and entertainment industry from the beginnings of time, decimating some jobs while creating entirely new categories of jobs.
And it’s true. It is a tale as old as time. Synthesizers enabled music pioneer Gary Numan to create an entirely new musical genre, for example, and he initially received significant blowback for destroying demand for session musicians.
But Numan was no tech behemoth. He didn’t create the tech, he just used it.
YouTube, an original unicorn
A more direct comparison is YouTube which was, from its beginning, a revolutionary new (and very cool) technology. But it was nothing without the content that passed through it. Originally, that content was expected to be user-generated content, enabling anyone to broadcast themselves (“broadcast yourself” was its original tagline, in fact). That meant, of course, that users gave their consent when they uploaded their videos. But it was professionally produced content, much of which was copyrighted, that really broke things wide open. Remember that SNL skit “Lazy Sunday?” That was a critical inflection point.
At first, YouTube looked the other way at this reality, building a massive customer base in record time in the process. Silicon Valley bros Chad Hurley and Steve Chen took that base – and the value created, much of which was built on top of unlicensed copyrighted works — and sold it to Google for $1.65 billion (yes, YouTube was one of the original unicorns).
Ultimately, the glaringly obvious fact that creators were receiving no compensation for the priceless value generated by their content led to major infringement litigation that culminated in today’s Content ID system that pays creators for any unlicensed use of their works.
For similar reasons, it is glaringly obvious that creators should be compensated for the fundamental value they bring to otherwise essentially useless generative AI tech.
This isn’t some anti-tech rant. It’s meant to be a simple declaration of artistic independence to be able to choose to participate fairly in a new generative AI world that builds its value on the backs of their works. Artist “opt in” systems from the likes of generative AI companies like Lore Machine (a story visualization platform I recently featured) and Rightsify (which says its Hydra music generator trains only on music it owns) point the way.
Given this essential core value creators bring to Silicon Valley, shouldn’t there be some media and entertainment unicorns too? There seems no reason why some privately held media startups or content franchise makers can’t be’ worthy of that badge of honor, given the reality that content is the essential ingredient that fuels the generative AI machine. After all, it’s the content that is transcendent and feeds the soul – the necessary “ying” to technology’s “yang.”
Think about it this way. Not even the great Bruce Springsteen – one of the most successful and prolific music generators of all time – comes close to unicorn status. He sold his massive catalog of music, including all of his songs and recordings, for $550 million a few years ago. To us, he’s The Boss! His music will inspire lives for generations to come. But in the eyes of Silicon Valley, he’s just another wannabe unicorn startup (who is only half-way there).
Maybe Bruce would sympathize with these lyrics from the chorus in my Suno-generated song, “The Song of the Unseen Unicorn”:
Oh, where are the media unicorns?
(where are they now?)
Lost in a world that doesn’t see their crown,
In a society of screens, clicks and endless streams,
The media unicorns, unseen, it seems.
Reach out to Peter at peter@creativemedia.biz. For those of you interested in learning more, sign up to his “the brAIn” newsletter, visit his firm Creative Media at creativemedia.biz, and follow him on Threads @pcsathy.
The post Without Content, OpenAI’s Sora and Generative AI Tech Are Useless. So Shouldn’t Artists Get Paid? appeared first on TheWrap.
