Sort-by-controversial as a proxy for information content

John Ohno on 2018-12-27

A common defense of trendism is that unpopular content does not necessarily have more information than average — some things are unpopular because they are boring & some things are popular because they’re interesting, so a naive reader with approximately average priors will get more out of being exposed to popular things than unpopular things. I’ve previously rejected systems like curation that attempt to harness trendism by injecting interesting content in a broadcast model, because my problems with trendism come down to a criticism of the tendency to accept a broadcast-like model (with all its downsides) as unavoidable. But, there’s an alternative to ‘sort by new’ and ‘sort by random’ that’s worth serious consideration: ‘sort by controversial’.

There’s a model, promoted by a piece of Scott Alexander’s short fiction but fairly natural, that suggests controversy is a good signal of high information content (marred by its tendency to also signal high emotional engagement). Alexander presents extreme controversy as being inexplicably divisive — we cannot articulate why someone would or how someone could disagree with us about something with this attribute, even though we might know a great deal else about them, & being on the other side of this line makes them Other and unknowable. This is almost the pinnacle of high-information: extremely surprising, and requiring the analysis of unknown-unknowns (the awareness of which is itself information-rich) to justify. I think this characterization does not account for classes of division that are easily explained.

Media that is controversial for political reasons is rarely interesting in of itself. In the united states, we have two major parties, and the country is more or less evenly split as to which is worse (and although almost nobody considers either good, almost everybody considers the difference between them substantial), so any piece of media that supports or identifies with one of the two major parties is controversial. Nevertheless, identifying with a major party is low-information: the major parties agree on almost every policy, so actual governance is similar between them in the worst case, and those disagreements they have are well-known and well-understood. Party affiliation is well predicted by geography, age, race, and cohort (in fact, it’s well predicted by each of these alone), which also makes it low-information. Mainstream party affiliation is not usually the product of careful thought; careful thinkers end up affiliating with obscure political ideas (georgism, left-accelerationism, deep green, anarchosyndicalism, posadism) or having political ideas that aren’t easily categorized at all. The affiliation of a figure with obscure political ideas is of substantially higher information value, but is less controversial, because controversy in this case is an artifact of tribal competition rather than disagreement about ideas or ideals.

Another example of high-controversy low-interest media is when controversy is explained by a cohort effect. In anime fandom, most people become anime fans for about 2 years, during which they consume all available anime & give inflated ratings; after 2 years, almost everybody drops the medium entirely. The two year mark is the point at which someone develops enough familiarity with the medium that the novelty of common tropes drops off & shows can be evaluated on their actual quality — which is all over the map. Shows that are extremely controversial are often popular-but-bad: the half of the audience who has had less than two years of experience with the medium is wowed by a novel-to-them use of cliched tropes and ranks the show highly, while the other half of the audience has already seen the same kind of show done substantially better & ranks it low. (In a general audience, we see this same tendency with children’s television: if we had three year old children and thirty year old adults watch and rate Blue’s Clues, then aggregated those rankings together without accounting for the age difference, Blue’s Clues would seem like a very controversial show.)

Nevertheless, I think that with a sufficiently diverse audience or a sufficiently un-diverse one, sorting by controversial is a better way of optimizing for interest than sorting by popular, sorting by new, or sorting by random.