A major concern with the recent development of artificial intelligence in many industries is the risk of losing jobs to AI, which many claim could complete simple tasks with ease and more complex tasks with greater efficiency than humans. One of such industries is the contemporary music production industry.
The primary concern comes from the rapidly evolving text-to-music generative AI models such as Suno or Meta’s MusicGen, which have advanced significantly, even compared to the start of this year. First gaining traction in 2023, these generative models are much like large language models such as ChatGPT. They break down existing pieces into tokens, like individual words in a book, then determine the conventional order of these “words”. Finally, it interprets the user’s prompt to select certain tokens and the general “plot” of the “book,” and groups those tokens into a “sentence” of “words” that makes the most sense.
Initially, results were extremely generic and unassuming, and the AI often misunderstood the user’s intention and parameters when interpreting the prompt; the classical music generations, which required more details and intricacies, were even less appreciated by the community, and sometimes were made fun of.
However, recent advancements in generative artificial intelligence are enabling researchers to train and develop more sophisticated AIs that produce more human-like music. Suno, for example, has developed rapidly since the beginning of the year, varying from many earlier models. A survey done by music magazine Pitchfork concluded that although AI-generated music still lacks nuance and a human touch for music experts, the level of authenticity is more than enough for the average ear to enjoy. They noted that text-to-music AIs are a great tool in lowering the skill barrier for non-musicians to produce contemporary music, but also expressed many concerns regarding their use.
Though convenient and helpful, the text-to-music AIs also raise ethical concerns regarding the training of specific models and the copyright of the music they produce. Although some models are trained on copyright-free music in the public domain, others are not. Critics of this approach suggest that the lack of transparency in the training behind these models may lead to unintentional mimicking and even copyright infringement. Many others are also worried about who owns AI-generated music; the copyright ownership is contested. While some argue it ultimately rests with the user, debate persists between the AI company's claim to legal rights over the AI-generated music.
With this in mind, people should be careful when using AI material in their work, not just in music, but in their daily lives, as the debate regarding copyright ownership is still raging and up to interpretation. This also raises a question for Concord Academy’s own music production and composition classes: should text-to-music AI be treated the same as other generative tools, or should it be viewed simply as another instrument in the creative process?