Do AI music generators pose a threat to the music industry?

AI music generators are a hot topic in the music industry right now. Some are excited at the prospect of music generators being a tool to aid in portions of the creative process. Others are salivating over the possibility of generating entire creative works. Then of course, there’s the poor songwriters, who may fear this is the end of their craft.

AI music generation is following a similar path as in other industries, with a goal to increase potential production and decrease costs. Yet, it presents many or more legal and ethical issues: recent lawsuits have explored originality and infringement in AI-generated artworks, and those concerns are more complicated (and more aggressively defended) in the music industry.

What Are AI Music Generators?

AI music generators work by producing clips of music somewhere between a few seconds to a couple of minutes long based on a descriptive text input.

The AI models have been trained on thousands of hours of existing music, and those inputs often include metadata that tells the AI the genre, artist, lyrics, style, and so forth associated with a given song.

This makes it possible to call up and transfer styles, specific instrumentations, and even vocal effects to apply to new tracks. The same goes for mood, genre, tempo, key, specific voices, and more — the possibilities are vast.

The process of deconstructing existing songs and reconstructing them into new material is being referred to in the industry as machines learning how to “destroy” and “recover” data sample inputs. This process can be used to:

• Produce new music theoretically

• Remix current music

• Interpolation (re-recording a pre-existing melody and using it in a new song)

As with other similar technologies, the results are both impressive and incomplete. For example, while Jukebox can spit out what one might consider listenable music (even with vocals), the AI music generator was incapable of writing lyrics that were anything more than nonsense.

Ethical and Legal Issues of AI Music Generators

Many ethical and legal issues have arisen related to creative arts and AI: after all, much of the training data for these systems is copyrighted and protected. Even if an entire song or phrase isn’t reproduced, artists and labels own every single building block used by the AI music industry.

Some platforms try to mitigate the risk of copyright issues by allowing users to experiment and upload with AI music generator tools only in a noncommercial context, theoretically prohibiting the sale of “AI songs.” Some only accept inputs from public domain works, Creative Commons-licensed works, or works contributed directly with permission of the original artists and copyright holders. 

Many of the minds behind the AI music industry argue that training AI models should fall under fair use. They cite parody and community usage and assert that this research is beneficial for everyone in the long run. (You won’t find many songwriters who agree with these assertions, though.)

One big difficulty is that these AI music generators are being developed without any plan or structure for identification or mitigation systems. Without an industry-standard way to identify AI-created music, artists are left with little recourse. YouTube-style DMCA takedowns and automatic detection systems are simply not built to handle the complexity of AI music.

Google discovered that around 1% of the cases studied, their AI model used existing music, unchanged, in the production of AI songs. The company has not released its tool to the public and says it will continue developing ways to detect violations and protect copyrights.

What Are the Potential Future Uses?

Despite the negative elements and the technology’s lack of maturity, AI promises to improve aspects of music production. For example, perhaps you want to work with a particular artist, but in a language he doesn’t know. AI could potentially merge that artist’s voice into new material. 

Synths and MIDI software have already opened up music creation to people without a certain level of music training. That’s where much of techno, hip hop, post punk, and new wave music have grown out of, incorporating beats and vocals and various elements into various patterns and even aiding in editing. AI tools could extend that concept, lowering the barrier to entry even further. 

The less complex the need, the better AI does. Samples, loops, uncomplicated atmospheric music, and sound design are all relatively simple tasks that AIs may be able to handle far better than composing and rendering a Top 40 hit. 

There is certainly fear that the AI music industry could interfere with human composition, especially at the lower end of the pool (such as inexpensive royalty-free music for marketing and commercial licensing). But perhaps the better way to look at AI in music is as an aid to human creativity, not a replacement for it. The ability to generate parts of a track with text prompts could free up an artist to unleash creativity elsewhere. 

Despite the valid concerns being worked through, there’s decent hope that AI music will not replace musicians and threaten their careers, but become another tool in the creative toolkit—something along the lines of composition software, MIDI programming, synths, drum loops, and the like.