The first time I heard “Hold Me Through This Moment,” I assumed there was an artist behind it.

However, as I motioned to add the song to my favorites, captivated by its ethereal bridge, I noticed the album cover’s suspiciously AI-generated appearance. Indeed, a quick Google search corroborated what I had suspected: The song and its “creator,” Imani Dawn, were nothing but products of algorithmic generation.  

Until then, I would have adamantly argued that AI was not, and would never be, capable of replacing true artists. 

The music we have produced thus far contains multitudes. From Delta blues to Native American powwows, music bears the weight of centuries of history and culture, providing a safe haven for those who haven’t found the community elsewhere. Without the experientiality songs are seasoned with, what difference would there be between music and noise? 

Others similarly highlight the ineptitude of AI creation, with Anne Ploin, a researcher from the Oxford Internet Institute, arguing in a 2019 report that AI was incapable of replicating the creative process involved in art.

Yet, perhaps we have been mistaken in equating AI’s ability to create art with its ability to drive engagement; despite my distaste for the algorithm, it has managed to will itself onto my Spotify recommendations. Thus, it raises the question: Exactly how competent is AI-derived music and how has it managed to gain traction? 

Going back to the basics, AI generates music using algorithms that analyze massive datasets containing millions of audio files, MIDI files and musical scores, accessed via the “open internet,” as defended by Suno co-founder and CEO Mikey Shulman in an August 2024 blog entry posted to the AI music company’s website. 

Of the two components of music composition, melody is easier for AI to replicate, as it essentially comprises scales, intervals and timing — quantitative data. While the output may be miles away from the works of Hans Zimmer or Ludwig Göransson, it is adequate. 

Lyric generation, on the other hand, resembles text generation more than melody generation in its success. Text is analyzed by breaking it into a block of words — labeled as tokens — and determining their meaning by observing them in the context of hundreds of sentences, using transformers to predict the positions of tokens within a sequence of words. 

Although the process appears practical, it is important to note that prediction does not imply understanding; AI does not know why the words are where they are, what effect they will have or how the narrative within the lyrics can develop and advance. The result is often meretricious: nice to look at but without any significant value or meaning. 

How, then, has AI music managed to invade our platforms? Well, there are two main reasons. 

First of all, some AI music is “good” in the sense that it can entertain consumers, as it has captivated me, as mentioned above. Most of us do not initially seek details in the content we engage with, looking for quick satisfaction instead. This need is easy to fulfill; even a mediocre melody can sound good if it’s on-key.

Beyond this, AI is a lot cheaper than human labor. 

Traditional music production entails significant costs: studio bookings, musicians, sound engineers, royalties and more. 

In contrast, not only does AI production incur none of the aforementioned expenses, but its products can be used commercially without recurring copyright fees. As a result, it makes sense that large for-profit corporations would gravitate towards mass-scale AI music production.  

Some consider this a positive, arguing that AI music will weed out the human artists who produce lazy music. Contrary to this belief, however, the oversaturation of the market will not only exclude the described artists; it will dilute audience engagement to the point where it becomes impossible for new artists to enter the industry. 

Ultimately, with significant barriers to entry for artists, we may soon find ourselves in an entirely AI market with no room for human creation and the solution is complicated. 

It would be naive to assume anything would dissuade the business side of the industry from prioritizing human artists. Especially considering that people do listen to AI music when it pops up in their feeds, it would be easy to argue that AI music provides a valuable service. Notably, the point is not to eliminate AI music entirely; it is to foster an environment where we still have the opportunity to choose what music to listen to. 

First and foremost, to sustain our right to choose, we need access to our options: tags that tell us which music is AI-generated and which is not, as well as additional digital data-filtering features. Only if we advocate for differentiating between anthropogenic and AI music can we preserve our culture while exploring the new.