August 24, 2018
Thanks to Alissa Simon, HMU Tutor, for today’s post.
How do algorithms know which options are right for you? They are purportedly a mathematical calculation based on personal tastes, previous preferences and your own interaction. I will use examples from Pandora and Netflix to express my meaning, but really, I could broaden this discussion to any number of entities. Also, I am using a very broad understanding of algorithms for this general discussion.
Recently, the song “Pachelbel Meets U2” popped onto my classical channel. Regardless of the song’s merits, however, I was immediately annoyed. I wanted this channel to be purely classical. For me, U2’s “With or Without You” came through so strongly that I could not focus on Pachelbel and it totally distracted me. I explain this only because it demonstrates taste’s incredible caprice. I like U2, I like Pachelbel, I like instrumentals of contemporary music, so, really, isn’t this just an example of me being picky? And I answer, yes! Of course, but isn’t that what taste is?! All I know is that I gave this song a thumbs down on my classical channel, for no really good reason. Sorry, Pandora, that was a swing and a miss.
My favorite category on Netflix is “Because You Watched.” This category bases suggestions off of something that you recently watched. These selections are not restricted to genre. In fact, they almost defy genre. Sometimes it links by actor, or comments by other viewers. And Netflix has nothing to lose with this process. The more content they recommend, the better for them. In fact, all of the companies that invest in complex algorithms have everything to gain. And consumers react by giving them data that they need to run the algorithms. If Pandora throws in instrumentals to my classical, and I vote thumbs-down, then Pandora responds with another selection. It also simultaneously removes this song (and perhaps some song group) from my category.
Broadly defined by Merriam-Webster, algorithms are a “procedure for solving a mathematical problem in a finite number of steps that frequently involves repetition of an operation.” Could that also be a definition of taste? There are many reasons that I might remove something from a playlist. Here are only a handful:
1] I don’t like the song
2] it doesn’t fit my current mood
3] I like it, but it is outside of the station’s intended purpose
4] I don’t like U2 and/or Pachelbel
5] I don’t like mixing genres
6] I don’t like remakes
7] I don’t like pianos or guitars
So, how does any mathematical equation break this nonsense down into bits of actionable information? How could an algorithm match infinite experience? Netflix and Pandora answer this by including other people’s recommendations. So, perhaps you gave a thumbs up to a movie that happened to be in the science fiction genre. Instead of recommending only sci-fi movies, Netflix will populate a handful of sci-fi and also some random films based on what other people liked. So, if another person liked the sci-fi movie you just watched, you will probably see a recommendation that has nothing to do with science fiction. And this seemingly random selection comes from other people’s tastes. Netflix, Pandora and others gain a lot by incorporating this feature. The more you interact, the more accurately they recommend, but also the more user-specific data they gain, which reinforces the whole system.
Does this type of system function differently than, say, radio in the 1960s, 70s and 80s, when top Billboard hits drove the radio songs that we all heard? Radio offered choice mainly by genre: country, Spanish, pop, etc. Though they did compile data, it pales in comparison to the amount of data that is available by these new devices. Radio offered music and we listened or not. I never thought twice about how many times I heard the Eagles or Michael Jackson on the radio. But now, I wonder why my Pandora Spanish station continues to play songs by Latin artists in English. Why are the ads in English, whereas my friends’ ads are in Spanish? I wonder if my behavior prompts Pandora to believe that English is my first language.
As we invite these devices into our homes and lives, it is worth truly thinking about taste. (Per a previous post, taste according to Merriam-Webster is: “a] critical judgment, discernment, or appreciation b] manner or aesthetic quality indicative of such discernment or appreciation.”) Why does Pandora (or any service) recommend something to you specifically? What do they know about you and are they making the critical judgments for you? I do not ask this because I am worried about some cyber conspiracy (although I’m sure there is data to support that too). But rather, I am worried about how taste interacts with culture. How individualized is the Pandora community and does it in any way reflect community as we currently define it?
With constantly changing technology, I wonder if something is being mistakenly hidden, missed or suppressed. I go back to the idea that Pandora thinks my first language is English, though I have given no data to support this. The algorithm seems to be making critical judgments about me, not just my music.
To read more posts about about taste, try these.
Taste defined in art and music
Taste according to Gibbon and Brillat-Savarin
To leave a comment, click on the title of this post and scroll down.