Applied Science #11: On Elf.Tech, GrimesAI-1, Hive IP, and a more resilient future

This piece was entirely written by a human being. The next one might not be.

A Midjourney creation in response to the prompt “a gathering of elves deciding the fate of the music industry but it's at the end of the movie 8 Mile”

“No idea’s original/ There’s nothing new under the sun/ It ain’t what you do but how it’s done” - Nas, “No Idea’s Original”

---

I’m going to attempt brevity on a subject that demands extensive thought and constant work on solutions from people far smarter than I am. 

In an age of infinite media—whether created by humans, generated by artificial intelligence, or a Frankenstein’s monster of the two—copyright law needs rethinking. In America in particular, the current regime lacks the flexibility or imagination for a present that once seemed like a distant future. We have reached the end of the age of copyright enclosure and must prioritize collaboration, attribution, and compensation.

What could that look like?

First, an acknowledgment that all media now exists in a public commons, even if it is protected by strict licenses. When the entire history of media is accessible on the other side of a search bar or voice command, it is also limitlessly composable—remixable, sampleable, coverable, usable. Any piece of content provides a springboard for creative collision and inspired generation alike. Stopping someone from sampling a song or finding a clever workaround to incorporate copyrighted audio into a TikTok presents an impossible task, a game of whack-a-mole on an endless field. Music IP law has been devouring itself for years in the face of this dizzying reality; increasingly specious lawsuits and opportunistic “new” business frontiers highlight the inherent contradictions in ways one can view and confront copyright infringement (or perceived copyright infringement, a rabbit hole that calls for its own book-length exploration). This paradoxical paroxysm points to the inadequacies of current copyright regimes: At present, law primarily grants the power to punish creative infringement, enrich corporate interests, or prop up dubious claims—not to truly protect individual creators.

Second, the music business needs to embrace the chaotic tide of this moment. The taming of music piracy in the wake of Napster was not the product of the major label IP fortress exercising its strength against consumers who didn’t want to pay. Piracy dissipated because streaming made it easier to access music than ever before, while displacing the risk of illegal downloading. Labels (and all IP owners, for that matter) will not simply be able to say a prayer before bed, take a deep breath, and litigate AI-driven tools and creations to the fringes. Universal Music Group chairman Lucian Grainge already seems to have admitted as much on a recent earnings call, saying: 

“We are open to, in terms of licensing, any business solution. Obviously we have to respect our artists and the integrity of their work…So yes, we’re open for business with [AI companies] which are legitimate, which are supportive, and [with] which we can create a partnership for growth.”

Of course, that flavor of conference-call-quote doesn’t mean lawyers won’t be sending scary letters to companies training large language models and computer generated voices on copyrighted material. It does implicitly concede that AI’s superbloom will require a more malleable response from the music industry than the rusty hammer taken to illegal downloading nearly a quarter century ago.

The spirit calling from deep inside Grainge’s tacit admission is so simple as to sound ridiculous: Make it easier to do an activity legally than illegally, and a majority of people will flock to the legal means. In microcosm, Spotify beat Napster (and an army of far better clones) by making the entire history of music readily available. The inverse partly explains why film and television piracy—time consuming and difficult for the average person—never matched the scale of music piracy and rarely gets spoken of in the pantheon of problems facing Hollywood.

Consider the complexities of AI-generated creation another way—we are no longer battling the bootlegging of existing intellectual property, but rather the mass birth of new intellectual property based on the voices and styles of established artists (to say nothing of the creation of hybridized voices trained on a mixture of different inputs). The fake Drakes and Weeknd’s will erupt like hydra heads for a time—chop one off, two new ones appear in the bleeding neck’s place. 

This moment calls for building around concepts of reciprocal intellectual property—what could be called “hive IP,” or a networked collection of intellectual property derivatives that radiate out from, point back to, and properly remunerate an original source. 

A healthy, solidly-constructed world of AI-aided music tools and users would do well to take inspiration from actual beehives: There is a queen who never leaves the hub; larvae are incubated and birth new workers bees on a periodic basis; worker bees leave the hive, setting out into the world to collect pollen, a definitive mission that can follow infinite paths to completion; worker bees return with pollen to produce honey; an implicit understanding exists among the bees that the fitness of the hive supports the fitness of each individual bee.

Mat Dryhurst articulated the bedrock of a concept like hive IP in a series of tweets after the US Copyright Office released a report on new guidelines in response to AI’s runaway development:

“Sympathize with the US copyright office, but an honor system approach to declaring which parts of a work were human created vs AI feels brittle.

Progressive policy would begin looking at artist ownership of bespoke ML models and the co-authorship possibilities therein.

We've already articulated the best path here:

- at the very least, enshrine the ability for artists to opt-out their data from training other people's models

- enshrine the ability for artists to own their own models

- enshrine the ability for artists to earn from generations from their models, by permitting ppl using those models to sell those works and earn $ for the original artist.”

At CreateSafe, our first approach to an untamable beast takes root in these principles: Build products that move with the tide rather than against it, protecting original creators and building better compensatory rails for derivation. This work stands on the shoulders of trailblazers like Holly Herndon, whose Holly+ presented a resilient model for licensing vocal likeness. 

Over the weekend, we launched the elf.tech beta in conjunction with Grimes. It delivers on her tweeted promise:

We developed elf.tech with Grimes based on her fundamental vision to open source her creative self. In that spirit, elf.tech is an artist API that allows creators to work easily with the IP and vocal likeness of an artist (in this case using inputs to create what we refer to as a GrimesAI-1 voiceprints), while also directly remunerating that artist with their new creations. Elf.tech presents a strong paradigm for near simultaneous creation, collaboration, consumption, and compensation—the notion of IP made valuable through networks of creators and consumers, rather than solely through monolithic broadcasters. 

With elf.tech and the introduction of GrimesAI-1, we are helping flesh out business solutions for this new era, while also introducing new modes of creation. 

These are the stipulations we laid out for elf.tech users:

CREATE

Registered http://elf.tech/ creators can upload pre-recorded acapellas (or record their own set of fresh vocals) to be transformed into a GrimesAI-1 voiceprint.

Users will receive the transformed Grimes AI -1 voiceprint back in a WAV file. They can incorporate these GrimesAI-1 vocals in the creation of new sound recordings.

Creators will own these new recordings. GrimesAI-1 does not claim any ownership of the sound recording or the underlying composition (unless that composition is a cover of a Grimes song).

SUBMIT

Songs can be uploaded for distribution to DSPs via http://elf.tech for $9.99/year. For artists looking to release a sound recording featuring a GrimesAI-1 vocal on a label, distributor, or platform that is not http://elf.tech, please email grimesai@createsafe.io to request approval for distribution before you begin the process.

Creators who use GrimesAI-1 voiceprint in their recordings agree to share an equal % of the master recording royalties distributed on streaming services, onchain marketplaces (primary and secondary sales) and other DSPs using GrimesAI-1.

For digital streaming platforms, royalties will be collected from all DSP and paid out by elf.tech. For on-chain sales and royalties creators agree to add the GrimesAI.eth wallet to the collection when minting.

Creators must have the legal right to all elements of their distributed recording with the exception of the GrimesAI-1 voiceprint. Neither Grimes nor CreateSafe, Inc. are legally liable for creators that upload elements that do not belong to them.

Credits

Featuring GrimesAI or adding GrimesAI as a main artist, secondary or featured artist is the approved method of crediting Grimes when using the GrimesAI-1 voiceprint.

STEMS

Elf.Tech is open sourcing Grimes audio stems to music creators and AI voice model companies. We want improvements to the Grimes voice model to be the best self-replicating pop star voice on the market and that can only happen if it’s a community effort.

We are still in Beta, we will be making consistent updates to the GrimesAI voice model, stems and distribution service, so remember to enjoy the CHAOS.

We see Elf.tech accelerating the fusion of historically disjoint parts of music making and business. It combines creation, collaboration, distribution, curation, and consumption; it is an ad-hoc DAW, a pop-up record label maker, and fan network all in one. Crucially, it invites anyone to experiment and generate audio, while also stipulating clear, yet pliable rules for commercial exploitation. As our CEO Daouda put it to Rolling Stone:

 “What’s the difference between this and what people have been doing with properties like League of Legends, Harry Potter, Star Wars, etcetera, where they’re making all of this amazing fan art and they’re also monetizing that amazing fan art? That’s a beautiful relationship between the consumer and the fan and the creators of that IP. And so that’s what we’re jumping on. That’s what we’re excited about.”

[Incidentally, you may attempt to make fan art and create something entirely new, as I did when I recorded my voice on elf.tech last night and received a voiceprint that sounded decidedly unlike Grimes; think of that, perhaps, as opening the door to the kinds of “fanfic” relationships that IP like 50 Shades of Gray has to Twilight. In that sense, elf.tech can be a synthesizer and distributor in one, incubating new IP loosely derived from a central inspiration. We are only scratching the surface of its potential and the pursuant economics that would fairly reflect these novel creations.]

Tools like elf.tech slather new paint on ancient creative natures—digital exploration of music as social fabric, oral tradition linking humans and saving stories through performance. While I don’t completely agree with Grimes’ contention that “copyright sucks,” I do wholeheartedly back her Jungian musing that “[a]rt is a conversation with everyone that’s come before us. Intertwining it with the ego is a modern concept. The music industry has been defined by lawyers, and that strangles creativity.” With generative technologies, you can, of course, just make a shitty Drake simulacrum, or you can use tools to unlock dazzling pathways of imagination, share what you make with others, and unite communities around creation (which may end up being communities of people who enjoy hearing Drake cover various songs throughout history—if that’s what gets you going, enjoy it before the UMG boot crushes your fun!). These concepts are as old as communication itself and stretch beyond the bounds of some corporate business affairs team or governmental copyright bureau. Tools like elf.tech can serve up entertaining distractions, or imbue impactful social function and form into music. 

Returning to the concept of hive IP, we see tools like elf.tech (and Holly+, for that matter) as adjunct treatments for an ailing creative ecosystem in which technology has already enshrined sampling and recontextualization in our media interactions, but legal regimes have strained to reflect reality. Elf.tech presents a framework for easy, consensual exploitation of IP at a time when it is far simpler to take the same actions without so much as a clear path to approval, let alone commercial exploitation. It is likely evident, but worth noting that we are fully experimenting here—with economic models, with tools, with web-based experiences, with simplified legal frameworks. Elf.tech will grow, change, and be refined over time as we learn from failures and successes alike.

The reasons for fear and criticism mount as various generative tools and AI ghosts spring from the shadows. The protection of artists, the primacy of human inspiration over machine generation, the rapid implosion of long held truths about creativity and how to compensate it—AI promises the greatest sea change in our field since the dawn of recorded sound and the dreaded introduction of the words “master” and “publishing.” Everyone who pretends this transformation is a mirage is either rich enough to escape and not care or dangerously delusional (or both). 

In late April, Peter Gabriel wrote the following:

“We are entering a world that is about to be fundamentally transformed by AI. Many people see AI as the enemy, but along with extraordinary scientific, functional and creative tools, it can provide great education and better healthcare to billions. It also has many inherent potential dangers that we urgently need to address.

Like the wheel, or the industrial revolution, I believe the changes coming with AI are unstoppable, but we can clearly influence them.

...

Should we defend artist rights and copyrights from AI? Absolutely. Should we defend human rights and democracy from AI? Absolutely. AI is a product of our species and we need to find ways to build the ethics, compassion and wisdom that we value, directly into the algorithms to protect and defend what is important to us.

For the arts, I think it’s very important that there are smart algorithms for recognising copyright infringement, whether in the work or in the prompts, and automatically embodying that information into the attached data.

There are amazing creative possibilities opening up with AI that are really exciting and transformative. I felt the same sort of buzz when computers came into music giving us samplers and rhythm machines, which, in turn, opened up new worlds of music making.

When the future has shown itself so clearly and is flowing as fast as a river after a storm, it seems wiser to swim with the current. AI is here. Let’s learn what we can and how we might adapt and evolve it to better serve everyone.”

As I see it, there is no future but the one that has already presented itself. We can move with it, we can help shape it, but to deny it is death.

Check out Elf.tech at this link or one of the many links scattered above. Transform your voice with it. Upload random audio to it. Make stuff with it. Break it. Create weird shit with us. Email me your creations, questions, complaints, thoughts, and concerns at jon.tanners@createsafe.io. Whatever you do, have fun experimenting.