Economics 101: The demand curve slopes downwards. That means that when the price of something increases, the quantity demanded decreases. Supply curves the other way: if the price increases, the quantity supplied increases. Prices thus communicate information between buyers and sellers, allowing the system to maintain equilibrium. If demand gets too high, prices go up, both slowing down demand and increasing supply: a negative feedback loop.
But what if the supply and demand curves sloped upwards? The price would go up, causing both to go up, over and over: a positive feedback loop. This is the current state of the attention economy, the reason the internet is driving people insane.
The market economy is just one of many self-regulating systems, but it’s by far the most central to social science, the one we first think of. Cell biology, climate science, and evolutionary bio are all based on the same principles.
And so is the market for digital media, the “attention economy” in which we participate so greedily. Here, though, the market analogy is more confusing than enlightening: there are still people producing and consuming, but the information greasing the wheels isn’t prices but attention.
As I wrote in Wither, Economics, that discipline is in terrible shape; they’re generally no longer doing economics because their institutions and culture have become decadent and insular. I think that media economics is essential to understanding social media today, and have published academic papers applying this framework to Facebook, TikTok, and YouTube—even have a book forthcoming on the latter.
So I was initially excited to read a working paper that’s been circulating called Paying Attention, by a genyoowine card-carrying economist (who is, to be clear, a grad student; the paper is excellent by the standards of the discipline, so my criticisms should be passed directly through the paper to the larger system). A huge amount of statistical and theoretical work is devoted to answering the following question:
Is the desire for attention from other people a quantitatively important non-monetary incentive?
Answer: yes!
Specifically, the author estimates whether additional attention (in the form of quantified audience feedback) causes posters to post more.
Answer: yes!
What makes it Economics-y is that there’s an actual elasticity estimated:
“Producers who go viral produce 183% more posts per day for the subsequent month”
and
“Adding three comments causes Reddit producers to supply 15% more posts.”
I argue that these numbers are essentially meaningless; the problem of temporal validity means that they could change at any point for any reason. A “post” is simply not a well-understood or stable category in the way that a “dollar” is, so attempting to build up a structural framework is simply untenable.
In my book, I take for granted that consumers respond to consumer-facing incentives, and producers respond to producer-facing incentives. This is literally the backbone of social media. But social media is too varied and too rapidly changing!
What I will not do is attempt to provide a quantitative answer to the question of the relative effect of supply and demand in this system. We could find, for example, that for a given point in time and population of channels, that 75% of the variance in what is talked about on YouTube is driven by audience demand and 25% by producer preferences
These numbers are intrinsically conditioned by both the larger sociotechnical context and by the specific affordances of the YouTube platform. We cannot plausibly hope to generalize these numbers to either other platforms or to an uncertain future. For this reason, I do not believe that careful causal estimation of a single or pair of effects is an efficient use of our scarce resources: the resulting knowledge is too expensive and too low in temporal validity.
In other words: who cares if it’s 183% or 147%?! Instead of the extravagantly wasteful display of rigor economists are forced to perform to get a top publication or job, I wish they actually used the intellectual tools developed by economists to understand these systems. Though my expectations are low, I’m still shocked that a paper titled Paying Attention fails to even cite, let alone engage with, the foundational insights of Econ Nobelist Herb Simon.
Attention is indeed the key to understanding the economy of social media. But attention is really not the same thing as money, and this has profound implications for modeling these systems. So let’s explore those implications.
There’s a link shared on Facebook. As people Like it (attention goes up), supply goes up: the recommendation algorithm acts as a central planner, collecting all of the information about the attention given to the link and then multiplying the supply through the magic of zero-marginal cost digital reproduction.
But because attention information accretes to the link (the little numbers on the post go up), the logic of credibility cascades means that demand also goes up. People want to see it if other people have also seen it; not just because they are cuing off of the decisions of others (the logic of information cascades) but because one of the ultimate goals of media consumption is to be able to display that you are up-to-date, informed.
Let me restate this in alternative intellectual framework. Walter Benjamin’s famous essay The Work of Art in the Age of Mechanical Reproduction discusses an aesthetic crisis brought on by technology: if artwork can be copied with photography, why should we care about authenticity? Benjamin argues that the copy cannot possess the original’s aura, its uniqueness in space and time, but that the proliferation of photographs changes the way that art operates in society. Instead of private objects of veneration or sublimity, mechanically reproduced art acts as a symbol.
Digital reproduction on social media allows for the re-invention of the aura through a postmodern lens. The quantification of attention, accreted to digital objects, creates unique value in the eye of the viewer — not through uniqueness but through ubiquity, through the experience of viewing the same thing as millions or billions of other people. The aura is constructed collectively by expanding the audience.
At the heart of the attention economy is thus a positive feedback loop. It is not self-regulating, so at some point this internal cycle runs into fundamental constraints outside of itself. There are only so many people in the world, so there’s a hard cap on the total amount of attention; up-to-datedness naturally decays over time.
So, the other key difference between the economy and the attention economy is that each unit can only be consumed once. (The exceptions prove the rule: if your primary motivation for consuming media is not display but pure diversion, you might not care about who else has consumed it. You can play PacMan all day and it’s just fun.) Time proceeds and new events occur, and each unit of media is ultimately unique.
The larger system, then, is the accumulation of attention, encoded in the brains of the consumers, the databases of the producers, and the weights on the neural networks the central planner algorithm uses to map these impossibly high-dimensional connections. It is worth discussing each of these actants in turn.
Algorithms
Most often implicated in the overheating of digital media, their task is ultimately the simplest: they accelerate existing trends by more quickly and efficiently sending information throughout the system. Some critics have pointed out that algorithms play an additional sociological function in diffusing agency, but I’d argue that this too is an already-existing feature of the system that algorithms accelerate. The market for attention is thick, with many producers and consumers, so agency is already diffuse.
Algorithms can of course have a massive impact on individual stories, producers or consumers, and do indeed structure the entire system. It would also be possible to change the algorithms to be so contrary to the revealed preferences of consumers that the credibility cascades couldn’t get off the ground. In the presence of competition from other platforms, however, unilateral action by a single platform would have little long-term effect on the online media ecosystem.
So algorithms learn the relationships between the tastes of all their users, distributed across all of the media producers. These are stored as weights in some massive matrix, and they don’t change overnight; they are stocks of what we might call “attentional capital,” in this case encoded explicitly in the data. If the algorithm notices you spending more time on/engaging with content from a give producer, that weight will go up.
This is why algorithms themselves can’t cause demand to slope upward: the weights for each outlet (for a given consumer) have to sum to 1. This mathematical identity means that the algorithm can’t push a consumer’s consumption beyond the limit of only showing them content from a single producer. In other words, reproducing that dreaded polarization machine of the late 2000s: the home page.
Producers
Producers are, counter-intuitively, far more susceptible to influence than are consumers. Profit-driven media have always sought to give their audiences what they want, and the quantified attention economy allows them to observe the audiences’ responses immediately and in high resolution.
When producers are large corporations, they are somewhat insulated from these tendencies; in particular, the rapidly eroding journalistic ethos developed in the US over the 20th century still acts as ballast, restricting what the firm can produce. The technology of physical reporting also imposes constraints: Amber Boydston’s work on “media storms” models the production of news with explicit attention to the temporality and physicality of reporters travelling to location. Once they’ve flown out to cover the corruption scandal, they’re likely to uncover more stories and stick around until public attention is captured by some other event. The instant and universal internet does not have this property.
When producers are economically precarious or emotionally needy young people, however, there is almost nothing that they won’t do. Media commentator Ben Smith tells a poignant story of a young, extremely online reporter who started with a reputation at Buzzfeed for “doing anything for the Vine,” progressed to vile public stunts livestreamed to his YouTube channel, and culminated with attendance of the storming of the US Capitol.
These “micro-celebrities” or “influencers” are nothing of the sort; they are puppets dancing to marionette strings held by thousands of viewers. And the feedback loop keeps getting faster. TikTok videos can be given thousands of pieces of feedback in minutes, and the young people involved are deeply invested in the approval of their peers.
Livestreams are the limit of contiguity: the focal streamer(s) looking back and forth between what they’re doing and the chat window scrolling by. They address and personify this “Chat,” speaking to their collective audience as if it were a single entity and they were in a one-on-one conversation. Collectively, then, the Chat controls the producer. This sounds like the plot of an overwrought episode of Black Mirror, but this dynamic is the reality for thousands of creators and millions of consumers. The explosion of personalized erotica on OnlyFans extends the scope of commoditization, laughing past older norms of propriety.
In general, then, producers get thousands of individual signals about the successes and failures of their creations, and respond to them intensely. During the boom years of YouTube growth, successful creators were riding the positive feedback loop, getting more and more views while putting up with more demands from their audiences. The more attention they get, the more is demanded of them. It is very difficult to self-moderate; the psychological effects of getting fewer views or likes are intense. Around 2018, many YouTubers experienced intense burnout from constantly posting. The most successful may have gotten rich enough to opt out, but there are an entire generation desperate to take their place.
Consumers
Consider the origin of fandom culture. Fandoms provide a much-needed source of in-group community identification for their members; this comes, of course, with out-group hostility. The only battlefield for conflict, however, is symbolic. The weapons are attention, attempts to accumulate positive associations and “Likes” or “Views” for the focus of the fandom. The “battle” to make PewDiePie the most submitted-to YouTuber was one of the biggest stories in the world in 2019 (regardless of whether you’ve heard about it); it was explicitly about the little numbers, manifestations of attentional capital.
There is also significant energy invested in making the ratio of positive to negative associations, and thus attentional capital, of your symbolic opponents go down. The participants—especially those from younger generations, raised to understand the attention level at the “pre-cognitive” level that capitalist subjects understand prices—want to increase the supply of their preferred media by increasing the demand for their preferred media…by reducing the supply of their symbolic opponents, thereby reducing demand for their symbolic opponents’ media.
The positive feedback loop goes both ways; less attention due to lower supply then comes to mean lower demand, rather than the homeostasis we expect from market economies.
The positive feedback loop for a given symbol continues to accelerate upwards until it bumps up against some other system, some friction to slow it down or some barrier that stops it. In both cases, significant energy is produced that must be absorbed; the other system has its own homeostatic negative feedback loops, but these can only handle so much before they are denatured or yanked out of equilibrium.
The algorithm doesn’t care; the attention weights in the matrix are all the same to it. But other systems still have to absorb the caprices. The relatively inflexible world of newspapers has met a well-publicized demise as viral publishers and content took off, denaturing the 20th century news ecosystem. And the corporate masters of the algorithm are happy to let this happen, until the degraded democratic system threatened them.
Producers are still connected on one side to the economic market, allowing them to exit the attention economy in one of two directions: positive feedback loops could crash them to zero, forcing them to do something else to survive; or they could skyrocket them to the moon, allowing them to cash out or settle into a steady state. The half-life of media producers will thus be dramatically shorter than in the previous technological era.
Consumers, though, have no place to go. There's no collapse, just more and more consumption, running up against and eroding the limits imposed by biology or society. Unless people cease to be fed or sheltered, there's nothing stopping them from applying more and more intensity and attention to the pixels on their screens, from attaching deeper and deeper importance to the producers and topics on the screen.
The controllers of communication always attempt to expand their influence, from the church to the radio broadcast, the university to the demagogue. This was clear to father of the discipline that studies control and communication, Norbert Weiner, who in the foundational 1948 text Cybernetics wrote that:
Of all the anti-homeostatic factors in society, the control of the means of communication is the most effective and the most important. (p151)
The issue that we now confront is that everyone controls the means of communication, both as producers and consumers. Producers or memes that gain forward momentum keep gaining forward momentum; if they start to fade, or even if they try to stand still, they decay quickly to nothing. There is no homeostasis.
The dynamics on the major platforms might change; they could ban more people, they could be broken up by the government, they could have better content moderation. But as long as the central mechanisms are positive feedback loops, online communication will continue to spin and spin and spit out energy into our culture and politics, overheating our locally robust systems until they are denatured and nothing is left.