Let’s start with an assignment.
Take a look at the graphic below. It is a network graph visualization built using the software from Quid, much like the network maps we publish all the time. A dot (or node) in this visualization is a single news article. Articles with the same colors have been arranged into a cluster by linguistic similarity. Connecting lines likewise indicate linguistic similarity, only between nodes in different clusters. Up, down, left and right don’t have any meaning, and the colors for clusters are arbitrarily chosen. Distance and connection are the only dimensions that matter.
Which cluster would you say is the most similar to everything else? Where is the beating heart of this network visualization?
If your eye is drawn to the lavender, purple, red and orange clusters there at the center, you’re on to something. But a 2D visualization is an inherently crude way to present these kinds of relationships. After all, this network graph is trying to represent about 12 million relationships between blocks of unstructured text, and you’d need a lot more than two spatial dimensions (and the sort of pseudo-third dimension you get from the wormhole effect of connectors) to do so faithfully.
So I’ll give you the answer: it’s the lavender cluster, and it’s not especially close. Here are our own mean normalized harmonic centrality scores by cluster, which I acknowledge is a complicated-sounding mouthful. It’s actually pretty simple. It is a measurement of how linguistically similar the average news article categorized in a cluster is to every other article across the network. The units are not necessarily intuitive, so I’ll put it in context: that’s a damn big gap between the top cluster and everything else.
So what IS that cluster? And for that matter, what is the dataset being used to populate the network map itself?
The network map is an arrangement of 4,804 articles from the Lexis-Nexis Newsdesk database published by broad-circulation, US-based outlets between January 1, 2021 and July 15, 2021 referencing violent crime. And the most influential lavender cluster…well, that’s the thing. It’s a cluster defined by people writing about how other people are writing about violent crime.
In short, welcome to Metaworld, in which articles which explain the news increasingly ARE the news.
The easiest way to get a sense of this is to read the articles. Here is a pretty thorough sampling of what you’d find in that lavender cluster:
Horowitz: NYT admits homicides in 2020 likely topped 20,000, most since 1995, but can’t figure out why | The Blaze
New research shows homicides rose in cities that experienced Black Lives Matter protests | The Blaze
Factcheck: “Crime is escalating to a level we haven’t seen in decades as a direct result of Democrats’ defund the police movement and Biden-Harris open-border policies.” | Politifact
NY Post Editorial Board: Murders soared in 2020 – and the anti-policing movement is clearly to blame | NY Post
Most of those articles are opinion journalism – editorials, op-eds and essays from established and well-known opinion journals. Many are vaguely noted as “analysis” pieces, some are of the “fact-checking” genre, and a few are marketed as hard news. And there’s nothing wrong at all with interested citizens and journalists writing opinion pieces, analyzing the events of the day or checking facts. I mean, you may think that some of the opinions are really bad, but unless you live in one of several European countries or Asian metropolises with Epsilon Theory readers where this isn’t the case, bad opinions are still legal.
What IS concerning to me, however, is that in the last 18 months, this emergence of a central meta-cluster dominated by discussions of discussions of the topic has become the rule for practically every topic of any social importance. Network graphs visualizing coverage of America’s COVID response, the January 6th riots, the environment, the future of remote work, voter suppression / voting integrity, CRT / privilege education, and now violent crime, have all demonstrated this property. It even showed up in an analysis we conducted as part of an ad hoc Epsilon Theory Forum discussion about changing narratives of American geopolitical dominance.
I am concerned because it hasn’t always been this way. This isn’t a natural feature of “meta” content being central because of topical language, or because of some tautology of self-referential language spiking linguistic similarity calculations – or at least it isn’t only because of that. This is a more recent phenomenon.
I am concerned because our data indicates that the language patterns, phrases, framing structures and arguments of opinion, analysis and explainer content are increasingly making their way into nominally hard news content.
I am concerned because our data also indicates that opinion, analysis and explainer content are becoming a higher volume of published journalistic content, even before taking into account the disproportionate distribution bias toward such content on the social media channels used as primary news consumption vehicles for most Americans.
I am concerned because the combination of those factors is steadily removing feasible ways to consume real news that isn’t debased by fiat news, our term for opinions stated as facts.
I am concerned because this trend exacerbates and very likely prolongs political bi-polarity. It continues to support a world with two wholly distinct sets of ‘facts’.
So what can we do?
Same as ever for news consumers: never stop asking “why am I reading this now?” Look for hammers and nails. Be mindful of attempts to auto-tune facts and opinions to archetypes. And support your local and regional news outlets. Our work on fiat news has consistently found that well-staffed city papers, even those with a heavy political slant on their editorial and opinion pages (in either direction), do a much better job insulating their news content from their opinion content than national and principally web-based outlets.
Influential individuals and institutions in the media who are interested in stemming the tide of fiat news could do a lot, too. More news organizations electing to be news-only outlets would go a long way, but the economic model on that may have long since sailed. More realistically, a commitment from a group of large news organizations to clearly mark and separate opinion, analysis, feature and ‘fact-checking’ content from hard news where they can control its distribution would be a big step. A set of protocols and standards for doing so that would allow other full-hearted publishers to join in would be a bigger one.
And come to think of it, that sounds like an interesting use-case for a distributed, non-government controlled system for emphasizing and rewarding positive externalities.