Why Algorithms Suck 1: Is Spotify’s ‘Related Artists’ Feature Doomed to Failure? - And Why You Might Like the Music You Do

This is the first in a series of posts I’m planning about the way algorithms do or don’t shape our music preferences and our ability to find music. Admittedly it’s a little biased towards the negative perception of them, but this is in keeping with the niche I hope to fill with the Audiomnivore blog- to move beyond genre and go deeper into music’s role in modern culture, how it can stimulate discussions, and why it’s good to get out of your comfort zone.

This first post, at least, is also heavily biased towards Spotify, as that was my main platform to listen to music before trying to branch out for blogging, as well as the fact that I spent a considerable amount of time trying to work out why it sucked so much. For this post, I’ll just focus on the process of finding music by ‘related artists’, something which I’ve tried in the past and always found somewhat underwhelming. Without further ado-

Why do you like the music you do? It might be the beat, rhythm, or groove. For another person, it might be the lyrics. For someone else, it may be the chord progression. There are almost as many unique draws to a specific piece of music as there are people who listen to it, with each person having an individual reason why they like that particular song. In light of this, it’s very strange, in fact, that algorithms such as Spotify’s don’t have the option to find similar artists by these categories- instead, it roughly groups artists by genre, as well as factoring in things such as artists’ side projects or other bands from a particular scene.

Ok, ok. Not everything has to be super iconoclastic and outside the box. It’s totally reasonable to group things by genre, after all, this is what brick and mortar record stores have been doing since time immemorial. The trouble arises when you realise the areas these algorithms are missing out on- the human element of music listening, the emotions, cultural contexts, and meaning behind the sound which are a huge part of what draws people to listen to what they do. Spotify also doesn’t allow for the fact that ‘similar artists’ can mean a host of very different things to different people. After all, it is not only genre which makes artists similar.

Take Marvin Gaye’s seminal classic album, What’s Going On. Dealing with anti-war and anti-racism themes, lyrically this could attract a broad variety of listeners, ranging from pacifists to activists. Amongst the ranks of listeners may even be those who are simply attracted to songs with a deeper or more socially conscious meaning. Then take the music- this same album would add to the list soul music aficionados, 70s kids harking back, producers (drawn in by Gaye’s revolutionary arrangements and instrumentation) as well as those who appreciate something mellow to listen to on a musical level.

But surely genre brings with it all these cultural connotations, doesn’t it? Yes, and no. It’s as much psychology as sociology which makes people listen to one song over another. While punk may have the cultural connotations of rebellion which make it easy to classify the more abstract reasons people are drawn to it, where do things like trip-hop or game music fit into the mix? If you first got drawn into heavy metal by it’s relatively dramatic edge, no algorithm would ever lead you to an obscure but similarly dramatic game soundtrack which you might just love. Ultimately when looking at user generated playlists, they show slightly different patterns than the algorithms.

One of the biggest things which stood out to me when thinking about this was how Spotify doesn’t seem geared towards music geeks, which is somewhat odd, considering what it deals in. The closely related, musically and rhythmically similar artists it offers as ‘related’ seem tailored not to musical discovery, but only to a specific type of musical discovery- more of the same, keeping it safe, nothing too different, whilst ignoring the wider links between artists which draw the same people to listen to incredibly disparate genres.

Whilst you could argue there’s no way Spotify could try to guess at the likes of all these individual groups of listeners, it still seems a shame. However, touring musicians break their audience down into micro demographics all the time. Why can’t Spotify do the same? Yes, Marvin Gaye is similar to the other soul artists offered by the algorithm, but to someone drawn by his social consciousness, he’s also similar to artists as diverse as Kendrick Lamar, Bob Dylan, and Aretha Franklin. The holes in the algorithm show even more when you get into greater abstracts like emotion or juxtapositions. Female, alternative pop artists of the younger generation- think Melanie Martinez, Billie Eilish, and Haley Kiyoko- tend to find a common fanbase, or at least one with considerable layers of overlap. By Spotify’s genre based algorithm, however, a fan of Eilish couldn’t easily discover Kiyoko without going down precisely the right rabbit hole in the platform’s twisting labyrinth of choices and clicks. Melanie Martinez is offered on Eilish’s related artist’s page, but so are artists as diverse as The 1975, Julia Michaels, and Lorde. What if some Eilish fans are sucked in by the edgier elements in her music, and therefore also enjoy something like Evanescence, or go for her trap influences instead? Who knows?

Whereas sometimes the algorithm seems based on some of the most superficial characteristics- sound, beat, an artist’s technical genre, at others it’s offers seem like little more than stabs in the dark, lumping together artists who happened to be popular at the same time, or who maybe share a few superficial characteristics. In the former approach, it suggests much lesser known emo bands with the same punk pop, power-chord based songwriting for My Chemical Romance, but leaving Nirvana out- despite the fact that for those born well after the grunge heyday, the two bands often go hand in hand. With the latter approach, Lily Allen is on Amy Winehouse’s related page, but so is Solange.

The real judgement is whether anyone actually listens this way- and some people certainly do, with geeks of particular genres having extensive record collections which go right down into the depths of their respective underground- but even these people have emotional and cultural preferences attached to the music they listen to. Many of the characteristics which could link artists are extremely difficult to pin down scientifically, such as the feel of music, or emotions such as sadness or dramatics, the latter of which has been the link found in well publicised studies showing the similarity of classical music fans and – yes, that’s right - metalheads. Spotify, in it’s defence, does have playlists for particular moods or specific crossovers, even including Classical Music for Metalheads, a spin on this particular concept. However, the fact remains that when finding new artists via the algorithmic method, there is extremely little option to tailor your searching to criteria such as mood or lyrics, let alone something extremely abstract like the contrast between soft and heavy, which nevertheless is popular enough to be used by acts as different as Swedish death metal band Opeth and internet meme turned performing artist Poppy, albeit in very different ways.

In some ways, the algorithms are in a lose-lose situation. The causes behind something as complex as human music taste cannot be pinpointed so simply- as much as the algorithm designers may want them to. Going on the basic sound alone doesn’t seem to cut it. Alternatively, when Spotify doesn’t classify things by genre- and suggests Foo Fighters, Audioslave, and Rage Against the Machine for Red Hot Chili Peppers- they miss pure funk fans who enjoy RHCP’s earlier work and slip through the gaps. Perhaps on the streaming services of the future there would be an option to find similar artists by mood, or by lyrics, or by production, using separate algorithms for each. Nevertheless, this still wouldn’t be able to broach the depths of subtler emotions. Maybe you can classify sadness, but there’s still a world of difference in the fan bases of Elliot Smith and Adele. Ultimately, is any sort of algorithm able to replicate the human element of music choices as well as another human can? In the end, it might be best to go to the record store…

Note: This is an opinion piece and is open for interpretation. Music listening habits are changing all the time and my own experiences on Spotify will naturally colour what I write.

However, if you have any strong feelings either way, feel free to leave a comment- this blog is aimed at debate/discussion!