Also Liked

I’m in Part 2 of David Sumpter’s Outnumbered. There’s a discussion on superforecasters and how they continuously update their expectations by weighing and incorporating new information. Sumpter also compares the performance of prediction markets versus algorithmic crunching of polling data from FiveThirtyEight. There’s no clear winner. Crowd-sourcing sometimes helps, but sometimes leads you astray.

 

The chapter that most grabbed my attention analyzes the “Also Liked” effect, named after Amazon’s algorithm showing you that “customers who bought this item also bought…” or “you might also like…” It’s clear that users find this helpful to narrow down the plethora of choices which would otherwise be overwhelming. The fact that many other apps do the same thing underscores its effectiveness. The algorithm recommends. You’re welcome to ignore its recommendations, but let’s face it, most of us don’t want to wade through gobs of irrelevant stuff.

 

Sumpter runs a few simple simulations to see how these algorithms, over time, filter what you see and don’t see. The results are interesting. All it takes is a few early “likes” that make the appropriate connections to catapult something into the few vaunted choices that are displayed. That’s even true for a small data set. If you considered thousands or millions of products, the early advantages compound exponentially. Slightly-better-sellers become best-sellers in a self-fulfilling prophecy, algorithmically controlled. When you upvote a news story or website in social media, the upvote quickly cascades into more upvoting.

 

A related phenomena relevant to academia are citation statistics. Sumpter provides a data plot showing proportion of articles versus number of times they are cited. The data follows a power law (you see a straight line in a log-log plot). Sumpter writes: “Power laws are a sign of vast inequality.” When examining the data over time, it starts to resemble the “Also Liked” effect. The rich become richer in fame. The poor get relegated to obscurity. We now have the h-index (for authors) and the impact factor (for journals) and a vicious cycle ensues. But more broadly with respect to ubiquitous social media, Sumpter writes: “Inequality is one of the biggest challenges facing society, and it is exacerbated by our lives online."

 

From there, Sumpter moves to examine echo chambers and filter bubbles. There’s some interesting old data looking at political blogs from 2004 in the leadup to the U.S. presidential election that year. Democrat and Republican blogs linked nearly exclusively to their own writers with little cross-over. But all of them still linked to the same limited mass media. By 2016, this had changed. Mass media divided into and joined their respective echo chambers. But with the implementation of “Also Liked”, what you were presented when you logged on to social media became filtered more and more towards what you liked or what your friends liked. That’s the filter bubble. It also explains why conspiracy theories proliferate and gain steam. But a strange thing happens if the conspiracy theory becomes too popular too quickly. When enough naysayers and doubters “like” each other’s responses, their views can start to eclipse those of the original conspirators.

 

Sumpter presents analyses by Michela Del Vicario and her team looking at the echo chambers of scientists versus conspiracy theorists on Facebook. Both are echo chambers and filter bubbles. Most of the general population (at least in Italy) tune them out. The features of the posts are interesting. Analyzing the words, “the general rule is that the higher the activity of the user, the more negative words they use… the effect was stronger for scientists… and the more active they were on Facebook, the more negative they became. Becoming a dedicated member of an echo chamber is not a route to happiness.” Worse in my opinion: “Not only are conspiracy theorists less grumpy than scientists, their shared posts are also more popular than those made about science news. This is particularly worrying since many of the conspiracy theories are about science.”

 

The increasing isolation of individuals within modern society, and the speed at which the internet proliferates ideas, have made it very hard to stop and think. We’re too busy reacting. I used to “like” posts from Facebook friends, back when I used it. But if I ever get back to it, I might stop doing so. Instead, maybe I should spend more time making conscious choices about personal connections and not feed the filtering algorithms. “Also liked” seems like a boon, but it can also be an insidious curse.