The unofficial Silicon Valley motto of “move fast and break things” may be an inspirational rallying cry for innovation, but it cannot apply to the fabric of society itself. It is high time our legislative bodies acknowledge the danger of that motivation and respond in kind, as we are in the midst of an unparalleled technological revolution.

Read more The Sphere Is Adding Enhanced 4D Effects to ‘The Wizard of Oz’

Art is a process, not simply a product. Be it writing, filmmaking, composing music, or countless other creative pursuits, the artistic process is integral to bringing a spark of inspiration to life. Any artistic undertaking may be rooted in innate talent, but it also requires diligence, discipline, endurance, creativity, and patience, ultimately producing a product of human expression unique to the artist’s experience.

Having art reduced to a “prompt” is an indignity to all creatives of the past, present, and future.

As president of the Motion Picture Sound Editors, I believe we must offer protection to those who dedicate their lives to these endeavors. Legislative action is vital to securing the necessary protections.

It is important to delineate between prompted generative AI outputs and the use of tools that incorporate AI components. To be clear, when I refer to prompted generative AI outputs, I’m talking specifically about outputs or assets in which the AI engine, not a human author, generates the specifics and details of the expression (e.g. video, audio, music, script, image) based on natural language input or other means, such as video-to-sound. This is separate from using tools to modify or enhance pre-existing human-authored creative materials, as is often done in post-production.

Preserving the value of human creativity is not as difficult as it would seem. I propose that the following two provisions be the framework of legislation for the protection of artists:

1. Legislation prohibiting prompt-driven outputs or assets of Generative AI models, and whole works that contain prompt-driven outputs or assets of Generative AI models as any part of said whole, from being copyrighted.

2. Legislation stating that all prompt-driven Generative AI outputs, assets, and their derivatives be clearly labeled as such: Visibly – pictures, images, and videos; Audibly – songs, scores, sound effects/design;  Metadata – trackable for ALL generative AI outputs.

The goal of a law, or laws, built upon these foundational principles, is to disallow the:

Under such regulations, individual prompted outputs would be ineligible for copyright protection, as would whole works that contain them. Examples include films that use a prompt-generated score, sound effects, visual effects, or VoiceOver, and songs that use prompt-generated horn sections, backup vocals, or other core elements that contribute to the song’s final presentation. Beyond the ethical and moral reasons for these provisions, there is also a legal basis for them.

The language must also prevent corporations from hiding any prompt-driven Generative AI usage under Trade Secret Law.

U.S federal law defines a copyrightable work as “an original work of authorship that is fixed in a tangible medium of expression.” Works are considered original when they are independently created by a humanauthor and have a minimal degree of creativity; a “spark” and “modicum” of creativity per the Supreme Court in Feist Publications, Inc. v. Rural Telephone Service Co.

Generative AI models are inherently derivative. Large Language Models (LLMs) are massive datasets that can be accessed and amalgamated by computer technology faster than at any time in human history. This makes these machines true marvels of innovation, not artists or creatives.

Ultimately, prompt-driven generative AI outputs and assets are the highest level of statistical calculation and speed, algorithmically predicting the most likely next step in a mind-boggling series of seemingly infinite steps to produce an asset. These assets are the logical end of those statistical calculations. At the end of the day, they are math. There is no “spark” or “modicum” of creativity as a result of a prompt.

A critic may ask, “But why do I get different outputs after making the same request or asking the same question of an LLM? Isn’t that an indication of some creativity in the generative process?”

In those situations, statistical variations occur in the LLM’s next-step calculations as confidence levels decrease. It is still a calculation, a random deviation in the process when confidence in the next statistical outcome, or token, is reduced. There is no a “spark” of creativity.

Or question, “But the Copyright Office already says that generative AI outputs are not eligible for copyright protection. Wouldn’t these measures be redundant?”

Not exactly. There are exceptions to this rule, such as “significant human editing” and “creative selection and arrangement,” that open the door for prompt-driven Generative AI outputs to be eligible for copyright. That loophole is like a small frayed thread on your jeans. It will get picked at until a hole opens up that is too big for any patch.

If there’s one thing we should have learned over the past decade or so, it is that loopholes beg to be exploited, and once that happens, the exploitation becomes the norm; so much so that we can’t remember what the original norm was.

Read more ‘Off Campus’ Star Stephen Kalyn Joins ‘The Housemaid’s Secret’

One such case has already chipped away at this subjective criterion. The U.S. Copyright Office granted a copyright, upon appeal, for the work “A Single Piece of American Cheese” by Kent Keirsey, the CEO of Invoke AI. This work and its appeal followed copyright law, exploiting the ambiguous exception of what constitutes “significant human editing” to a work whose original source is a prompt.

It should be noted that Invoke AI’s self-described purpose, “wasn’t to create a masterpiece (obviously). We wanted to see where the ‘floor’ was for the U.S. Copyright Office to grant copyright protection to an AI-assisted work.”

This was a company-driven probe into the limits of copyright law. After the ruling, Invoke AI , referring to artists, “We hope that those people can now feel confident that their work can be protected, at least in the United States.”

This floor was found only because more stringent laws like the one(s) proposed above were not in place. Existing copyright law left the side door cracked and Invoke AI found their way in. They were the first to pick at the frayed thread in the jeans, and they certainly won’t be the last.

Legislation is the only way to close these loopholes.

Tens of thousands of AI-generated pieces of “music” are uploaded to Spotify every day, some of them hitting the charts and reaping profits. Artists on Spotify get paid from a pool of subscription and ad revenue. The dollar amount paid is dependent on an artist’s total number of streams past a minimum threshold. It follows that any prompt-driven Generative AI “song” is taking money out of the pockets of human artists.

Generative AI models are being hotly contested in courtrooms across the country over the legality of their training. Failing to account for and disclose their training materials has left them vulnerable to accusations of theft. By all measures, that theft is voluminous. While unsettled, this fact alone should relegate their outputs ineligible for profitability as they are currently “fruit of the poisoned tree.”

Would there be any amount of “modifying” I could do to a stolen car that would then make it mine to sell?

Many of the defenses around accusations of these companies’ scraping of the internet and other methods of garnering training content, rests on the Fair Use Doctrine. There are four pillars of Fair Use, all of which are weighed in the application of the doctrine as it pertains to copyright violation. (Section 107 of the Copyright Act):

Arguments against these companies’ use of works as training materials could be made under several of the pillars. But let’s look specifically at number 3, which seems to be the most egregiously violated tenet.

By and large, under the Fair Use Doctrine, the amount of a copyrighted work used is determinative of whether or not copyright infringement has occurred. The larger the piece of the work used, the less likely Fair Use applies. In the case of companies feeding their models training data, they ingested whole works – complete books, movies, songs, pieces of art, among other things. This “copying” and usage of innumerable complete works severely tests their Fair Use argument.

While the following points are grounded more in ethics and morals than in law, they are fundamental to understanding the full scope of this issue. They should absolutely carry weight in this conversation, as it is the artistic efforts of many creatives that produce a product ultimately owned by a few. The theft of talent and artistry by LLMs goes significantly deeper than simply the impact on legal rights holders.

For every song ingested into these LLMs, ALL of the creative elements that went into the track have been surreptitiously appropriated. The law is very thin on allocation of rights when it comes to recorded music. Yet, the final tracks that impress upon our hearts, minds, and culture are the sum of the skills and creativity of – at a minimum – the producer, songwriter, engineer, mixer, and, of course, the musicians. The tone of the guitar, the arrangement of the horns, the groove of the bass, the nuance of the drums, the timbre of the background vocalists are all now critical information inside of an LLM to be coalesced and synthesized into an output. All without permission. All without compensation.

The theft in filmmaking goes even further. Every movie assimilated has stolen the artistry of the Writer, Director of Photography/Cinematographer, Production Designer, Prop Master, Costume Designer, Picture Editors, Production Mixer, Sound Editors, Foley Artists, Re-recording Mixers, VFX artists, and many, many others.

The law must adapt to address this systemic exploitation. Decades-old “work for hire” contracts never anticipated that an artist’s own works would be weaponized by computers, replicating their talent for the profit of corporations and users who lack the skill, or even the interest, to create it themselves.

The magnificence of the AI renaissance has blinded our society to its pernicious creep into the arts. Unregulated, prompt-driven generative AI stands poised to initiate and accelerate a profound regression of human creativity, if not its cessation. While rightly hailed as a “democratizing” agent in some circumstances, in practice, it serves as a means of redistributing stolen creativity and artistic talent en masse. It needs to be curtailed.

This is a call to arms for humanity. We need the men and women of the United States Congress to be the resistance against the millions of corporate dollars that would undoubtedly fight this legislation, and to stand with the people and enact laws with these two logical, ethical, and moral provisions.

Which of you will be the champion of human creativity?

David Barber, CAS, MPSE, is a supervising sound editor and re-recording mixer. He is also a musician and is currently serving his second term as President of the Motion Picture Sound Editors.

Read more ‘Tires’ Renewed for Season 4 by Netflix