Find music like the music you love, with AI

Originally published at: Find music like the music you love, with AI | Boing Boing


Funniest one I got is this one.


1 Like

depeche mode sounds more like new order than joy division sounds like new order.

1 Like

Well, this post led to a depressing discovery.

My first thought was that this reminded me of every noise at once , a similar project that used spotify data to make the connections and sort music into related genres.

Then I read that the guy behind this has been laid off from spotify , and that it’s not going to update any more, and they might in fact kill it.


Well, that sucks. I was about to make the same observation. is a treasure. But, once again, it seems that we can’t have nice things, or at least not for very long.


It reminds me of the user-created tag clouds that used to have. They were great.

1 Like

Based on the associations, it seems like at best it works off “people who liked X also liked Y,” or “these bands were all sometimes given a particular (genre) label” as the basis for deciding that two bands are similar to each other - when they absolutely aren’t, from a musical perspective. Bands, especially those outside formulaic pop, usually end up pretty arbitrarily stuck in a particular genre, more often because of an impoverished genre vocabulary used to describe them rather than because of meaningful similarities.

Oh, you think that’s bad:


There’s an awful lot of bands popular at the same time, or which came from the same place, etc., none of which has anything to do with musical similarities… I suspect this is due to an incredibly simple reason: the system treats labels associated with the band as equally significant, whether they were “gothic-industrial,” “'80s” or “Finnish.” (And is probably privileging/based entirely on texts from the US, where a band is only going to get a country label if it’s outside the Anglophonic world.) But then it’s made weirder than that, because it’s also using other associations with a particular band to group them, which means everything floating around the search item can be completely dissimilar to each other, with no connection between them at all.


Why is this not the same as Pandora? Which never pretended to be “AI”, just an interesting datamap that used statistical models to produce predict… oh, wait, that’s what “AI” is these days, isn’t it?


I entered Hugo Largo (art rock/dream pop) and got hit with The Sorts (instrumental rock). Anyone you knows anything about Hugo Largo, would have expected Hetch Hetchy instead (produced by two Hugo Largo bandmembers: bassist Tim Sommers, and violinist Hahn Rowe, with Hetch Hetchy’s lead singer Lynda Stipe sounding uncannily like Hugo Largo’s Mimi Goese). AI? Aye yai yai! As bad as Jethro Tull being awarded a Grammy in 1989 for a category/genre they did not inhabit.

1 Like

I suspect it is entirely text-based, grouping artist with other artists that it found close to each other in lists/websites/etc. I entered Canadian band “Triumph” and it was surrounded by other Canadian bands of the roughly same era but that were in entirely different musical genres.


Is there any actual AI used in this? It’s not new, I remember this website showing up three or four years ago when I Googles “Bands like ____”.

1 Like

This seems pretty out there.


They sit side by side in my CD rack because that’s how the alphabet works out.
I’m a bit underwhelmed by the software failing to come up with some kind of overlap here.

That mostly seems pretty straightforward compared to some of what I’m seeing (where I cannot, for the life of me, see any connections at all), and reaffirms my notion of how this must be working - it’s just grouping by labels. If you look up “An Emotional Fish” it’s described as an “alternative rock” band popular in the '90s. So in this case, it’s organizing based on the labels “alternative rock,” “'90s” and a few other things, none of which say anything much about the music itself. And in this case whatever other labels it’s using really aren’t relevant - it looks like it’s struggling and only has a small number of meaningless labels, so it can only make the most meaningless associations as a result. As a music recommendation engine, it becomes totally worthless, when it’s (secretly) saying, “Hey, do you like a band that was popular in the '90s? Here’s some more bands that were also popular then!” (Although in this case, about a quarter are… something else entirely.)

The baffling bit is how Rocio Jurado, for example, fits into that, though. Or Milladoiro, or Kylee Henke… who seems to just be a random Youtuber? (Not even sure to whom “Tamara” refers.) I suspect the label set is so small in this case that some sort of corrupted data is playing a significant role in the organization. (I even wonder if the relevant label here doesn’t just amount to “has a video on youtube”…)

Knowing An Emotional Fish’ body of work they are worlds away from Nirvana, Pearl Jam and Marilyn Manson. So yes, I expect the site is using the most tenuous of connections.

1 Like

… it’s not the same in that it is a competing product

Right? It’s remarkably similar to a site I can’t recall the name of from at least like, ten years ago. :thinking:

Why is Billy Strayhorn further away from Duke Ellington than, uh, Iron Butterfly?

Looks like a good tool for a game of “Six Degrees of Separation”, but for musicians.

Worth following the link to Gnod, which leads to a similar tool for authors at I’m getting the feeling that this one gives a lot of points for “has co-authored a book with”, though it’s possible that it just looks that way because authors who collaborate tend to have similar styles.

Edited to add: Also, look at, which explains in broad terms what the prediction is based on. Basically, a large survey or individuals’ music preferences.

1 Like

Kind of a cool idea - it feels limited by the selection of a single band, though. Surely AI should be able to accept a list and suggest interesting new choices.
And I’m not put off by the fact that it seems to be based on listener preferences. That seems way more reliable than trusting an AI system to understand musical taste!
The comments here have already suggested some avenues to investigate. :grin: