Report by Dawoud Kringle
On Tuesday, December 12, 2023, MFM held its most recent Music is Essential Talk. The subject was the impact of artificial intelligence on the music industry. This was a first for MFM in that the Talk was in a roundtable discussion/think tank format.
Traditional notions of music authorship and attribution are being challenged with the advent of creative/generative AI. These AI models are much more complex than directly sampling a song. The legal and regulatory landscape around generative AI is a controversial topic. This interactive roundtable discussion explored how AI is impacting the music industry and how as music professionals we can have our say in how these new tools can grow and evolve without negatively impacting our livelihoods.
The participants were Adam Reifsteck (MFM Board Member, webmaster and podcast host), Barry Heyman (MFM Board Member and Copyright Attorney/ Herman Law), David Newhoff (Copyright Advocate and Editor of Illusionofmore.com), Roger Blanc (MFM Board Member, member of NY Chapter leader of the Recording Musicians Association and Local 802 Musicians Union), and Jerome Harris (member of Local 802 and President of the Music Workers Alliance).
After Reifsteck opened the talk and introduced the participants, each participant introduced himself. Barry Heyman was the first to speak. After Sohrab Saadat Ladjevardi interjected with his thoughts about the subject, Jerome Harris introduced himself, adding that his organization (Music Workers Alliance) is formulating a stance on how to handle the problems that will inevitably arise from generative AI. David Newhoff and Roger Blanc introduced themselves. Newhoff made mention of not only the legal but also the cultural impact of AI. Blanc spoke of two contracts in negotiation with SAG-AFTRA that potentially deal with the factor of AI as applied to the music business.
Reifsteck then reiterated something Ladjevardi mentioned; namely that AI is, strictly speaking, not as new as many would think. It’s just that in the last few years, the technology advanced with generative AI in such a sudden and startling way that people are still trying to understand it and figure out how to deal with it.
These concerns range from concerns is how AI can effectively mimic anyone, i.e. deep fakes, harmful cybersecurity attacks, and Large Language Models (LLM) that have billions of parameters they can plug into to open a larger aspect of generative AI that can write engaging text, create images, video, and audio, and the inevitable copyright issues (we are still fighting to preserve the rights we have to our music). Ultimately, the changes that happened in the 21st century (mp3 downloads, streaming, etc) have already made things difficult for the professional musician. The aforementioned aspects of AI could make things worse.
At this point, Reifsteck turned it over to Heyman. He began with an overview of what copyrights are and what works are copyrightable (basically, music and lyrics, and sound recordings). He went into considerable detail on how copyrights work. Then he went into what Newhoff had mentioned regarding the cultural aspect of AI and its status as a copyrightable work.
There are three requirements to have a copyright: 1. The work must be original, 2. It must be in a fixed, tangible form, 3. It must have authorship. Copyright gives the owner reproduction, distribution, and publicly performs the work. This leads to the question of whether the product of a generative AI, such as a poem/lyrics made by Chat-GPT or a video made by Fireframe.ai, is copyrightable and who owns it. He brought up the propensity for artists to be lackadaisical in approaching copyrights, and that it’s advantageous to take a more detailed and decisive approach.
Heyman brought up an uninteresting legal precedent. Some years ago, a photographer gave a camera to a chimpanzee. The chimpanzee took its photo, a “selfie,” in a mirror. The photographer attempted to claim a copyright on the photo. He was denied the copyright because he did not take the photo. This brings up the requirement that copyrights can only be given to human authorship, not authorship by an animal – or a machine. Therefore, if a work is generated by an AI, then theoretically, it wouldn’t qualify as copyrightable authorship. However, utilizing AI as a tool to create something of human authorship may or may not qualify as copyrightable material.
He then brought up the question of copyright infringement. AIs are trained on data derived from the products of human intellect and imagination; all of which are copyrightable. The products of an AI response to a prompt could be substantially similar to an existing copyrighted work that could qualify as copyright infringement. There are also questions about AI-generated works that are in the close likeness and resemblance of existing work.
After Reifsteck offered a brief discourse on derivative works created by generative AI, he asked Heyman if the main crux of the legal battle over AI copyright is the tech companies using existing copyrighted works for generative purposes without permission. Heyman explored this, saying that while individual components of what is used to generate an AI work are not copyrightable, the resulting output should not reflect or resemble what was imputed to create it.
Ladjevardi offered an opinion, saying that the greatest challenge facing AI is the ethical use of the technology. The technology itself is ethically and morally neutral; it’s human use of the technology that determines its ethical quality and legality. He wondered if there is a way to regulate it to ensure its ethical application.
Christ North brought up the question of quantum computing and the potential of unprecedented computing power. This leads to. the possibility of AI producing work that a majority of people will find at least sufficient, if not preferable. He brought up two music groups in Japan that are blowing up as of this writing, and they are all AI-generated. He asked if there would be an entirely new form or model of compensation for artists as a result of the presence of AI.
Newhoff took the floor and offered his views on the questions that had been raised. He feels that we will probably not deviate from the long-standing doctrine that copyrightable works need to be human-authored. A more difficult question is that of the human author who collaborates with an AI tool, and where we draw the line. The Copyright Office has come up with guidelines on how to disclaim AI-created elements in one’s work. Newhoff explored the question of whether the development processes infringed on existing copyrights, “in the style of,” and other concerns. He added that he believes that AI developers should not claim copyrights on anything outputted by their AIs.
Reifsteck brought in Harris and Blanc, using the National Producers Association’s recommendations to the US Copyright Office as a jumping-off point. In a nutshell, the National Music Publishers Associations hold that works created through AI should not be copyrightable, and that AI developers should be required to license materials and keep records of the materials they use for training AI models. They also say that they are opposed to the opt-out regime and that the training on copyrighted works is not fair use.
Harris responded by saying that the Authors Guide and several other organizations gave recommendations of laws, policies, and regulations that the companies that produce generative AI should require. This includes getting permission for the use of the creator’s work in training their systems and compensation for these creators, that the data sets, that the companies disclose what they are using to train their systems, and that AI-generated content should be labeled or watermarked. He went on to say that the number of music, literary, and artistic-related organizations are coalescing their views on what is necessary to protect human livelihoods. Harris also mentioned that the inclusion of new and radical technology tends to shift the dynamics of what people can do, and reiterated what Ladjevardi said about human’s ethical use of AI.
Blanc told an anecdote of a friend of his named Karen Toffler, who is the daughter of Alvin Toffler, the author of the 1970 book Future Shock. 53 years later, we are “drinking from a firehose” in terms of the rate of technological development. He said that the rise of hard disk recording and personal computers to make music gave rise to (among many other things) hip hop, which traditionally used pre-existing music to build from. This created copyright problems, which were eventually solved. The use of other people’s intellectual property in music is more commonplace than many realize. He noted that in his experience when a resource is threatened and a collective group acts to protect that resource, the larger financial interest prevailed over our comparatively insignificant cultural interests. Expecting the rich and/or powerful to take care of business for us is unrealistic.
Blanc had spoken to Mark Ribot about the upcoming recording contract negotiations. Ribot is concerned that something may be slipped into the language of those contracts that is taken as official policy that could do irreparable harm. He went on to say that we (New York City) are a heterogeneous environment with many big interest groups. The diversity that New York enjoys also makes attacking and dominating the bigger interests more difficult.
The floor was opened up to the audience. The questions covered a wide variety of concerns and ideas. I recommend the reader to access the video recording of the talk.
Ultimately, the problems and the unanswered questions that arise when generative AI is factored into the creative and legislative process of music and art have no historical precedent. We are in entirely uncharted territory and have no model to work from.