Who Chose That Hour Last Night?

"Our attention has fallen shorter than a goldfish's, so we only watch short things." It's the most familiar sentence used to explain the short-form craze. Yet the premise is fake from the start. In the "humans 8 seconds, goldfish 9 seconds" comparison, the 8 seconds is in fact a figure attached to humans, and tracing it back, the statistics-aggregation firm Statistic Brain put out a number with no basis, a Microsoft Canada report copied it down, and the BBC tracked it and refuted it. The goldfish's memory is fine, too. The conclusion is off as well. Ask Korean users and 59% still say they prefer the full version (short-form preference 49%), yet short-form consumption is at an all-time high. If both premise and conclusion are wrong, why can't we stop scrolling?

The Fake Premise of "Because It's Short"

First, let's set down the conventional wisdom that "we watch because it's short" with data. The average single continuous viewing of short-form is 21 minutes. Content being short does not make the usage short. 26% of teens, once they pick it up, go past an hour. Content length and the time you stay held move separately.

The gap between preference and use is sharper still. In Korea, YouTube Shorts holds an oligopoly over the market (Shorts 75% · Reels 43% · TikTok 20%), and Instagram and TikTok rewrote their all-time-high dwell times again.

All-time high (2026-01, Korea · WiseApp) — Instagram MAU 27.97M · 32.6 billion minutes/month, TikTok 9.43M · 9.2 billion minutes. But this is total app dwell time, not a short-form-only tally.

Yet ask the same users and the majority say they prefer the full version (overall 59% versus short-form 49%). The amount watched is at an all-time high, while preference points the other way. The only group where it reverses is teens (short-form 62% versus full 47%).

That said, this gap is circumstantial, not proof. The answer "I prefer the full version" may carry a social-desirability bias toward looking more serious. Still, what's clear is that this picture isn't explained by the single word "short."

Treat teens as an exception and you may miss the point. That preference itself crossed over to short-form in the heaviest group may be not because they're a different breed, but because they're a leading indicator, showing first the direction the rest will head. Of course the opposite is possible too — if it's a life-cycle effect where the younger you are the more short-form you watch, tapering off as you age (the 27% of those 60 and over who watch under 5 minutes fits that side too), then teens are not the future but merely one phase. Cross-sectional data can't separate the two. Either way, the single word "short" misses it.

So What Is Scarce, Then

The answer is on the supply side. In 1971 the economist Herbert Simon wrote that "a wealth of information creates a poverty of attention." Today that wealth is beyond reckoning. About 500 hours go up on YouTube every minute. To watch a single day's worth would take 82 years. Shorts are played more than 200 billion times a day (because the counting basis changed in March 2025 to "count every play," that figure can't be compared on the same yardstick as the prior 70 billion).

Simon was right. The more infinite content becomes, the scarcer the resource that is truly scarce — attention — turns out to be. And if attention is scarce, then the act of choosing what to spend it on itself acquires value. Each second you let slip while deciding what to watch is a second you couldn't spend watching something — the opportunity cost of attention. And the feed handles even this allocation on your behalf.

The Architecture That Takes Over the Choosing

Here the decisions short-form takes on are two. One is "what to watch." The recommender picks the next video to surface (I covered this branch in leaving what to watch to a recommender, and it's outside this piece's range of direct proof). The other is "when to stop." Aza Raskin, who invented the infinite scroll in 2006, publicly regretted it in 2018, revealing that the design's intent was "to remove the decision point where the user would exit." The moment when you used to decide with a click whether to move on or not was erased wholesale.

So a swipe isn't choosing one item off a menu; it's a reaction — pressing approve or reject on what you're given. There's an honest counterargument: isn't the swipe itself an active veto exercised every few dozen seconds, and if you can move on the instant you dislike something, hasn't control actually grown? Half right. But that's the right to refuse what you're given, not the right to choose what gets put up as a candidate. Netflix and the bookstore have someone else assemble the list too, yet there's a decision point there — "you choose, you start, and there's an end." What short-form erased is exactly that point — both the start and the stop. The outsourcing this piece directly proves goes only that far (the removal of the stop decision). The larger claim — that the recommender has taken even "what to watch" — is circumstantial, leaning on leaving what to watch to a recommender.

"I'm just watching because it's fun, though?" True. But why is it fun? Because the algorithm has learned your taste and spoon-feeds you only the high-hit-rate stuff. The fun isn't a rebuttal to outsourcing; it's the result of outsourcing. "Because it's fun" and "because I didn't choose" are the front and back of the same mechanism, and on top of it a variable reward keeps your hand from stopping.

One thing I'll make clear. This is not because willpower is depleted, nor because of decision fatigue. The "choice overload" hypothesis — that too many options paralyze you — showed an average effect of essentially zero in a meta-analysis of 50 experiments, and ego-depletion theory — that repeated decisions wear down self-control — found no effect in a simultaneous 23-lab replication. Short-form is not a device for soothing depleted willpower. The architecture lifts choice out of the default, and on top of that variable reward hardens stop-less viewing into a habit.

This outsourcing has a supply-side mirror. Creators know a video gets buried if it can't catch your finger in the first few seconds, so they reverse-engineer and shoot the hooks the algorithm rewards. The creator's design fills the seat the viewer vacated by giving up the choosing, and the platform turns the two into currency. In 2025 more than half of Instagram's ads ran on Reels, and Reels' annualized revenue was cited at the scale of $50 billion.

Where Dopamine Belongs

Where should we place the commonly invoked "dopamine addiction"? The dopamine that neuroscience speaks of is not a substance that detects pleasure. It's closer to a "reward prediction error" circuit that amplifies the signal when something better than expected happens. What's more, the "wanting" that craves something and the "liking" that actually enjoys it are split apart in the brain. What short-form hooks is not pleasure but the pursuit of "what comes next," and a variable reward whose next you can't know keeps that pursuit switched on.

I read this relationship as the structure being the engine and dopamine the lubricant. A variable reward works only once there's already an architecture for it to roll on. Only after the structure has erased the stop and filled the next screen does dopamine roll that delegation smoothly. This is a design that makes a habit, not an addiction disorder in the clinical sense. Close it off with the single word "addiction" and you miss the real engine — the architecture.

So What Comes Next

Now we look ahead. From here on it's inference, not measurement, so I write with the strength of conviction turned down.

Producing content endlessly is getting steadily cheaper. One analysis tallied that of the first 500 Shorts shown to a brand-new account, 104 — about 21% — were fully AI-generated "slop." When infinite generated output is layered on top of infinite content, the next resource to grow scarce is clear: the editorial judgment that sorts what's trustworthy and what's worth watching — trustworthy curation. The scarce good shifts from content and attention toward "the eye for choosing," and taste and curation become capital in their own right.

Recall the teens, the leading indicator, and the parents' question follows. Is it harmful to development and cognition? The honest answer is "we don't know yet." An association between short-form and attention decline is observed, but causation is not proven, and the reverse explanation — that people whose attention is already scattered watch more short-form — is also possible. I'll neither pronounce harm nor guarantee safety.

This forecast can be tested by a few signals. Whether the share of AI slop in the recommender feed and source labeling (watermarks) get introduced; whether "a human chose this" curation products pull ahead of pure-algorithm feeds in monetization and reach; whether the platforms' promise shifts from "more content" to "a trustworthy filter." There's one more signal: forced-friction legislation aimed at the very design of "removing the decision point." China placed a daily 40-minute cap and a late-night block on under-14s in Douyin's youth mode; the European Commission, viewing TikTok's infinite scroll and autoplay as "addictive design," issued a preliminary finding of breach of the Digital Services Act; and New York and California in the US restricted minors' "addictive feeds" by law. This regulation is a different grain from the signals above. It doesn't test the curation forecast — it tests the diagnosis. If engagement holds even after friction is forcibly inserted, the driver was not the architecture; if engagement buckles under friction, the diagnosis that "the structure that lifted out choice is the engine" was right. If the curation signals above run backward, the forecast is wrong; if this regulatory experiment comes out backward, the diagnosis is wrong.

Short-form took our time. What it took more quietly is the decision of when to stop. Add to that the recommender taking over even what to watch, and the seat from which you could answer "I decided to watch this" goes blurry. Ask who chose that one hour last night, and the people who can answer clearly grow ever fainter. In the age of delegation, what actually grows scarce is the ability to decide not to delegate.