Up close and personal: when the algorithm goes too far

Today’s entertainment world is characterised by hyper-personalisation. Everyone has a unique TikTok feed tailored specifically to their own niches. DSP playlists have names like “For You” and “Made For [username]”. Consumers have come to expect everything they are served to be in their exact niche, a concept taken to its eeriest extreme in the Black Mirror episode ‘Joan is Awful’. 

Artificial intelligence (AI) is deepening this trend, and not only by making algorithms even more sophisticated. AI text-to-music generators like Songburst and CassetteAI allow users to whip up whatever kind of music they wish. Universal Music Group’s Endel partnership reimagines albums for specific activities, like sleep and study. 

The major label has now entered a new partnership with YouTube for an “AI incubator”, meant to explore how AI can “responsibly” empower artists. Here is something for the incubator to think about: if we take this notion of hyper-personalisation to its extreme, the endpoint is a world where every DSP user receives a slightly different version of every song release. Is this a dream, or another episode of Black Mirror?

A song for every user

Three trends are now colliding:

• Consumers expect hyper-personalised content

  Featured Report

  State of the music creator economy The consumer era

  The pandemic triggered a surge in the music creator economy, which brought with it an influx of external interest and investment. 2022 saw growth moderate, but the long-term outlook is not only strong,...

  Find out more…

• In today’s remix culture, a song’s initial release is just the beginning of its lifecycle

• AI is accelerating the above

So far, the content feed is hyper-personalised. But in the future, it could be the content itself. Imagine if a single release pack included both the official song, and a version ever-so-slightly tailored to things like the user’s BPM preferences, their favourite style of singing, their language, or their heart rate. Maybe the song is slightly sped up, the bass made slightly louder, a piano melody switched for a violin. While this all seems far off, the beginnings of it are already happening in some spaces — like fitness, and gaming, where the company Reactional Music enables music that reacts to gameplay in real-time. 

The NFT boom of 2021 saw a wave of generative music NFT projects, where artists generated thousands of unique audio files based on the same input, making each version rare. The concept of “generative music” comes from Brian Eno, who coined the term in the 1990s — he saw a shift from composers acting as “architects” who oversee a specific plan, to “gardeners” who plant seeds that grow into something all of their own. Amid the 2021 boom, Water & Music published a deep dive into the challenges of generative music NFTs, which range from sociocultural to legal, concluding that consumer demand around generative music NFTs is not yet proven. But as AI enters the picture — with a quicker, more palatable experience for the average consumer — the sociocultural wall may begin breaking down.

Dream or dystopia?

Undoubtedly, some artists reading this are fuming. Hyper-personalised music would arguably destroy the art form by turning something intentional into something that feels more arbitrary. Besides, consumers don’t always know what they want — which is why, in Rick Rubin’s studio, “the audience comes last”. Some would argue that the music industry is already doing too much to try to please consumers, by making decisions largely based on what gets engagement on social media. 

Thinking about it from the eye of the consumer, the entertainment experience is quickly turning into a battle between the user and the algorithm’s version of the user, prompting a new feeling: “algorithmic anxiety”. And a song for everyone would also only further saturate streaming platforms. Maybe it makes more sense for generative music to exist in specific spaces, like games and meditation apps, but not on streaming platforms. 

Perhaps the biggest lesson here is the double-edged sword of hyper-personalisation: how it can be used to drive fandom and better consumer experiences just as much as it can end up isolating listeners. As we have seen with social media virality, algorithms that are overly focused on discovery can have the opposite effect intended — end up pushing creators away from their built-up fanbases. The industry is speeding down the path of hyper-personalisation, but care should be taken to ensure we do not end up in the Black Mirror version of this story.