Garden Path Content

Noticing a change in incentives

June 4, 2023

Vocab: in game theory, an iterated game is any game you play over and over with the same players and without any known end. A game where asocial behavior is a winning strategy when played one-off, like the Prisoner's Dilemma, changes dramatically when played repeatedly by the same players.

I’ve fought with this essay for ages, partly because the vocabulary in it makes me sad. I want to talk generically about telling stories starting in prehistory, and continuing through printing political pamphlets in the Enlightenment, publishing novels in the 19th century, writing scientific papers in the 20th century, and making TikTok videos last week. The modern vernacular for all that might be “creating content” and a literary theorist might say “constructing a text” and repeating either of those phrases out loud makes me want to hide under a rock and never emerge.

So, just temporarily, allow me to use the verb “writing” and the noun “story” to describe the whole human history of creation, in all media, fiction and non-fiction. Allow this for the sake of my spirit. For reasons that will eventually become apparent, don’t let my word choice fool you into assuming that means solely written words, or even a human author doing the writing.

The oldest and simplest reason to write a story is to communicate an idea. You want someone else to know or feel something, and when they understand or feel that thing, you have your reward.

Writing for money is most often an iterated game, in the sense that when an author sells a story, their audience knows who they are and who their publisher was (if they had one). To convince the audience to buy more, the author must provide them with at least some satisfaction.

(Just like “writing” is standing in for a spectrum of activities, by “satisfaction” I mean any one of a whole rainbow of emotional and intellectual reactions; it doesn’t have to feel “good” so long as it feels purposeful, needful, valuable. Readers like to be tortured, just a little, if they understand in the end what the pain was for, and believe the author intended their experience.)

To be more beguiling in attracting readers to their next work, they might pull a Scheherazade and hint at the beginning of the next story as they end this one; or they might pull a Dickens and break one story into many parts; but each part still needs to produce enough satisfaction on its own that people are willing to continue paying to get the next. A soap opera with too many twists and too few resolutions eventually loses its audience.

There were advertisements in stories long before there were stories paid for by advertisements; my casual research suggests that a consistent form of that innovation took until 1836, with La Presse – a French newspaper that was sold more broadly and more cheaply, with some of the cost to readers offset by the advertisements within.

Despite the addition of a third party – the advertisers joining the authors and the readers – stories paid for by advertisements have much the same iterated-game constraints. If a publisher wants people to see ads, they have to want to read the story, and if the author or publisher hasn’t delivered satisfaction in the past, readers will pass them by. The only change is that whatever kind of satisfaction the author delivers mustn’t lessen the needs your advertisers are trying to fill. A story about being content with the simple life might sell garden planters but not luxury vehicles.

Until very recently, that covered the vast majority of the story-making world: free stories that only want to communicate; stories paid for directly by readers; and stories subsidized by advertisers. (Through this lens, propaganda is just stories with advertisements for the government.)

And so, I think, readers came to believe unconsciously that they have a tacit contract with authors: the stories we pay for with our attention will be written with intention, and will eventually resolve in some satisfactory way. If they don’t, readers have recourse: they can pan the story, and not read things from that author or that publisher. The gap between this tacit contract and the explicit one that goes “readers give writers money for whatever the writers have written, the end” is perhaps why there is tension and frustration from both sides for writers suffering from writer’s block, like George R.R. Martin or Patrick Rothfuss; the readers sense that a story that stops without a satisfactory ending violates the tacit contract, while the writers point to the explicit contract that says they don’t owe more than what the readers have already paid for. (Of course, the publishers may have not-so-tacit contracts to complain about.)

Even early algorithm-driven content platforms obey those standards. Satisfying stories get upvoted, unsatisfying non-stories get downvoted; writers who are effective get followed and writers who are not, do not. Admittedly, the attention economy led to the expansion of “satisfaction” to include “blind rage” – more things got watched and shared and commented on purely because they made people angry – but the tacit contract was tattered but still intact. Authors write stories with intention and readers reward those stories that affect them.

Over the last few years, the TikTok-style of infinite-scrolling algorithm-fed content treadmill has become widely adopted; either in its purest form of content-firehose or blended, like Twitter and Facebook which show you stories from people you follow mixed in with stories from people you don’t. And the small but significant change is that content can now get boosted not just by explicit voting, subscribing, or sharing, or through the reputation of the writer, but simply by being consumed. TikTok (or Instagram or Shorts or…) is interested in anything that I don’t skip past; watching to the end means my eyes are on the platform and consuming ads. The algorithm can serve me stories written by anonymous sources I’ve never heard of, and it does not need my opinion or my satisfaction when it can measure my attention.

This violates the tacit contract. The game is no longer iterated! My satisfaction with the author’s previous work does not matter, so long as I consumed it. The author’s incentives in this game are to capture and hold attention, regardless of the readers’ satisfaction. And so (after almost 1000 words) we arrive at what I set out to talk about: garden path content. These are story-shaped objects that have no intention to communicate at all. Instead, they are merely story-shaped; they seem at every moment to be leading to an exciting revelation or a grand conclusion but just keep leading you down the garden path until they stop at a dead end. It’s clickbait that doesn’t need you to click, only stare.

Now, this sort of thing has been done on purpose in the past, artistically; see Steve Martin:

“What if there were no punch lines? What if there were no indicators? What if I created tension and never released it? What if I headed for a climax, but all I delivered was an anticlimax? What would the audience do with all that tension? Theoretically, it would have to come out sometime. But if I kept denying them the formality of a punch line, the audience would eventually pick their own place to laugh, essentially out of desperation. This type of laugh seemed stronger to me, as they would be laughing at something they chose, rather than being told exactly when to laugh.”

But Martin doesn’t arrange an entire show that way; for the trick to work the way he describes it, the joke-shaped object has to be embedded in a larger fabric of real jokes, to set the audience’s expectation for tension and release. Garden path content relies on other authors to provide that fabric, and by itself gives no satisfaction at all.

@tkfifd #viralvideo #tkfifd ♬ original sound - TK FIFD

(Note that in addition to the central video, which has a storylike structure but no story, there’s the Mr. Incredible reaction image, which also implies something is going to change, something is going to happen.)

You may find yourself consuming garden path content repeatedly because it leaves you feeling you must have missed something. You’ve been so thoroughly conditioned to expect the intent to communicate that you assume it must be there if you just read more closely.

But since it has no intention, garden path content doesn’t even need to be original. It can be collaged and Frankenstein-ed together from attention-holding moments of other stories:

@respectcali #respect #amazing #foryou #omg ♬ Love You So - The King Khan & BBQ Show

A pastiche of setups with no punchlines.

The final piece of the unhappy puzzle is once I suspect you can guess for yourself: what if these authors had a tool that could generate endless story-shaped objects without intention – without any human intervention at all? Statistical machine learning models are here to provide exactly that service; limitless, practically-free text that is easily human enough to trigger our expectations of intention and satisfaction, and so hold our attention. Whether it fulfills those expectations is left entirely to chance; in this context it does not matter.

In written form, ML-generated garden path content frequently comes in the shape of a FAQ; paragraphs of text that wander from their subject might be slightly easier to catch than a series of related-but-not-quite-sensible questions:

Here’s a relatively obvious example that only takes a few paragraphs to catch on to. But this one archived from the same site is quite a bit trickier.

I’ve started to see these sites in the first SERP more and more often; often two or three different ones in one search. It used to be you’d occasionally see a site that was just a gigantic list of search terms, like a dictionary pasted onto a single page, used to improve the PageRank of other, spammy sites. This new form is different because it intends to fool not only the search engine but the human searcher as well. They are formatted, structured, and mostly grammatically correct.

And while fully ML-generated video that can fool a human may still be in the future, algorithmically collaged video, with computer-generated voiceover, is absolutely among us, and has been for years.

(You may be muttering intently at this point about Grice’s Maxims, or the cooperative principle – and it’s true, this kind of violation of storytelling is also a more fundamental violation of the expectations of communication. But the relevant context for garden path content is the structure of pressures that create it and the system that creates those pressures. The fact that it is maxim-violating is merely what makes it feel uncomfortable, not an explanation of how it comes about.)

And I don’t see any trend in the structure and design of social media that will decrease the rate or effectiveness of garden path content – as long as “any form of attention” is the metric that platforms seek to maximize, attention thieves will thrive.

And so I finally reach my prediction: media literacy in the coming decade will not only require us to identify new varieties of malicious, rage-inducing fakery that has characterized the last ten years online but will also require recognizing as quickly as possible the non-emotional non-communication that is garden path content. Some questions to ask:

  • If it shows a process, do I know the intended result? Especially relevant for videos including cooking, DIY, manufacturing, cleaning, etc
  • Why would someone film this/write this? Why did this person film this/write this?
  • Are there obviously collaged elements suggesting a big reveal is coming? Reactions, duets, react memes, “stitch incoming” text. Who applied those elements?
  • Why is someone reacting to this? Why is this person reacting to this?
  • What would a satisfactory conclusion to this even look like? Could there be one?

Most garden path content is short enough that asking these questions consciously won’t be fast enough: you’ll have consumed it before you’ve answered them. But bby asking, I hope, we can train ourselves to recognise these story-shaped non-stories and dismiss them instincively.

Have you run into a particularly interesting specimen of garden path content? Reach out! I’d love to hear about it.

My books are open! I am currently taking contracts for the second half of 2023. Drop me a note.