So I just went to YouTube, as I often do when looking for some quick entertainment, and one of the first items suggested was a song called “Coyotes” by a group I’d never heard of, “Public Library Commute.” It was just recently posted and only had something like 60 views (with something like 78,000 subscribers to the channel), so out of curiosity I clicked on it.
I immediately recognized that the song (which I enjoyed), had some chord changes/melodic similarities (starting at 0:41) to something I had recently listened to, although I didn’t remember what at first. By some miracle, my memory did kick in, and I realized it was Daddy’s Car. Which I don’t ordinarily listen to, but which I dredged up not too long ago to post a link in a thread here on the SDMB.
Listen to both links, and you’ll probably see what I mean. My once-excellent command of Western music theory has slid away from me over the decades since I’ve exercised it, so I can’t explain it analytically, but to mangle what Supreme Court justice Potter Stewart said about pornography, “I know it when I hear it.”
So, are the algorithms really that good? Did “Coyotes” get a priority spot on the YouTube videos presented to me because some AI recognized that it bore similarity to Daddy’s Car?
Those two songs don’t seem all that similar to me.
Google does seem to use a bit of AI in their algorithm these days. Their mysterious algorithm also looks at your viewing history and looks at what others with a similar viewing history have also viewed. Google also gives videos that they think they can make more ad revenue off of a higher priority in your recommendations.
I personally find Google’s algorithm more frustrating than anything else, but once in a while it leads me down a rabbit hole of things that I find really interesting. I also find that I frequently have to remove videos from my watch history because youtube’s idiotic algorithm occasionally latches on to things that I don’t like and gives me absolutely horrible recommendations based on it.
You are missing something that is definitely there, but like I said in my OP, my ability to analyze it fails me due to having spent decades away from the Western musical training I once had. But there is DEFINITELY a similar shift. It isn’t that the songs are similar, it’s that they both utilize the same trick from the “bag 'o tricks” that composers with a command of the theory behind what they are doing can use. Like shifting up a key, which (to the best of my fading knowledge - glad to accept corrections/clarifications/illuminations from people who spend more time on this than I do) is a common technique in pop music.
I don’t doubt that you are hearing some sort of similarity that I’m not. Maybe you can give some time stamps from both songs where you think they are similar?
The first time at about 0:41 in Coyotes and around 0:38 in Daddy’s Car, although the latter seems saturated with whatever the heck it is.
But, this is nothing I’m intent on proving. Your first explanation is almost certainly what’s behind the suggestion, and while the similarities in the pieces may be driving that (no pun intended), that doesn’t mean that algorithms noticed, just humans.
I’m happy to grant that there’s some similarity there that would be apparent to someone trained in music theory, but not to a layman.
But I think Google’s algorithm is mostly based on what humans watch and link to, based on whatever criteria we use. If the algorithm suggesting this for you wasn’t a coincidence, it’s probably because some other music-theory-trained human put both in a playlist illustrating those particular compositional techniques, or a human who was a fan of those techniques listened to both songs repeatedly. Or more likely, several humans did things like that, because even Google isn’t likely to notice a pattern from one instance.
What I find frustrating about YouTube suggestions is that yes, it does occasionally give surprisingly good suggestions for things I didn’t even think to look for, but enjoy-- like train travel videos. One was suggested to me, and I went down a mini-rabbit hole there.
BUT…watch more than one of any subject, and 98% of the videos YouTube starts serving up after that are all on that same subject. Hey Google, I’m not that one-dimensional! I do have varied interests.
Right now for example I’ve been watching a lot of cooking / recipe vids, often along the lines of “How to make the best _____” and 99% of my YT suggestions are food-related. That’s not so bad, but in the past I’ve watched just one or two, say, shuffle dancing lesson videos, then my YT feed is like 90% suggestions for young girls shuffle dancing. C’mon, Google! Mrs. solost or my kids could be looking over my shoulder when I open YT and get the wrong idea! Then I have to search for other topics just to try to get YT to diversify.
YouTube has been really good at recommending new music for me. I suppose overall there are more misses than hits, but there have been a lot of great hits. Also they keep recommending certain videos over and over until I give in or click Not Interested. But they are usually good if I give in.
I remember when the Pandora music service came out (20 years ago?!) and the whole idea of the Music Genome Project was mind-blowing but also mind-blowingly simple. Get a “genome” from elements of songs, classify them all, use the classification to put music in front of people who have curated a list of what they like.
The Music Genome Project is exclusive to Pandora now, but for a while it could be licensed. I wonder if YT has licensed it? I’m sure they have done what they can to reverse-engineer it and make their own version.
But anyway, the technology to give you recommendations of music based on the actual musical content of a song has been around for 20+ years. I’ve only put it to the test when making my single Hot Holiday Jazz Pandora station but it’s RIGHT on the money. I love that damn station!
If I ask it to play a genre, YTM absolutely sucks at playing the songs I want to hear.
But if I ask YTM to play a specific song, it’s pretty good at guessing the next five or six songs I want to hear (or am in the mood for)
But then after those five or six songs, the music starts to suck again. So I have to come up with another specific song to request and start the whole process over again.
I will sometimes pick a random YouTube song to play in the car and then just go Next, Next, Next to see what comes up. Typically, to my disappointment, it usually goes back to the same songs after hitting Next two or three times. And it usually doesn’t matter what song I start out with. So, I don’t think it’s that smart. But, I’m not sure what its doing exactly.
I also noticed this behavior, of including a brand new music upload from somebody I’d never heard of near the top of the suggestions, started within the last couple weeks. I’ve been wondering how it could possibly be able to match things to my tastes based on so few views. I’m usually listening to things in my already curated playlists, so I haven’t clicked on any of these newbies to listen to immediately, but a couple days ago I decided to just start adding them to a new playlist and I should give that a listen soon to see how well its new algorithm is doing.
I listen to YouTube at work, and it’s pretty good at making playlists of the songs I’ve listened to before. But it adds The Verve’s “Bittersweet Symphony” every damn time. No matter that I’ve hit the “Not interested” flag each time it shows up - doesn’t matter.