What are recommendation algorithms doing to our sense of taste?
The MinefieldPlay episode
Show notes
There are few things more peculiar to a person than their preferences. Why it is they enjoy one genre of music over another, or a particular artist within that genre but not others. Why they derive specific pleasure from a certain type of fiction (romantasy, say, or Scandinavian procedurals) whereas others (like Agatha Christie’s Poirot crime novels or dystopian sci-fi) leave them cold.
And then there’s that whole undergrowth of what we might call “guilty pleasures”: low-brow books or formulaic television series or lowest-common-denominator movies that we secretly enjoy but would be mortified if anyone found out.
Which suggests, of course, that the network of preferences we call “taste” most often has a class dimension to it. Having specific tastes, and finding certain things distasteful, signals our belonging to the social stratum that has learned how to appreciate those cultural objects. It’s not that taste is altogether emptied of its subjective dimension — its ability to evoke authentic feeling, real enjoyment — but rather inner appreciation is in a kind of performative dialogue with the expectations of others.
And yet even within the realm of taste, there are subtle distinctions. Immanuel Kant one between “the taste of sense” (what is pleasant to me) and “the taste of reflection” (which may not be immediately enjoyable, and which may require effort or patience or instruction before yielding its treasures). According to Kant, what is truly “beautiful” is only available to the taste of reflection — a form of enjoyment that we want to enjoy with others.
In our world of endless digital reproduction, we increasingly rely on recommendation algorithms to curate our encounters with culture — algorithms that work along the lines of, “If you liked that, you will probably like this …” Algorithms, in effect, attempt to make our preferences legible, which is to say, predictable, offering us more of the same in order to keep us interested and engaged.
In this way, algorithms can only work at the level of what Kant called the taste of sense — they can operate along the lines of “likes” or “dislikes”.
But algorithmic recommendations cannot read the subtleties of our preferences, they tend toward massification, and they rule out the possibilities of both aesthetic achievement — learning how to appreciate, even love, what we didn’t initially “like”.
