The Choice That Chose Itself

People keep picking the thing they picked before. Not because it's better. Because they picked it before.

Here is the number, because the number is the argument. In one task, participants were shown two options of identical objective value — same payoff, no correct answer — that differed only in how often they'd had occasion to choose each one during learning. They picked the more-repeated one 73% of the time (95% CI: 63–82%). When the researchers flipped which option got repeated, the preference flipped with it, down to 29%. The bias isn't a wobble around a true preference. Between equally good options, repetition is the preference.

And those repeatedly-chosen options weren't just selected more. They were rated as better — participants' post-task valuations tracked how often they'd chosen a thing (r = 0.43, 95% CI: 0.31–0.54). Not "more familiar." Not "more comfortable." Better.

"The surprising thing was how strongly repetition alone can change preferences," says Ben Wagner, lead author on the paper. "Options that were chosen more frequently were not only preferred, but also rated as better." That is a sentence that should probably come with a hazard label.

The study is Wagner, Wolf, and Kiebel in Communications Psychology, and its title is about as thrilling as a parking citation: "Action repetition biases choice in context-dependent decision-making." What it actually says is that a measurable portion of what you experience as taste is a rumor your brain started about itself and then believed.

You have one primary organ dedicated to generating a persuasive sense that you know what you like. It turns out you mostly know what you did.

i · hold that number steady

Before the vertigo, the fine print — because the fine print is what makes this survivable as an argument rather than a vibe.

That 73% is measured between options the experiment made objectively equivalent. This is not a finding that repetition beats a better option 73% of the time; the researchers had to strip value out of the comparison to see the mechanism cleanly. In the cross-context tests where alternatives weren't matched, repetition still biased choice — that's the paper's actual claim, that biases previously chalked up to relative-value learning are better explained by action repetition — but the effect is a thumb on the scale, not a hand.

Second, and I want to say this before someone says it to me: the stimuli were fractals. Abstract shapes, deliberately meaningless, chosen precisely so that nothing competes with prior action for the brain's attention. Your sandwich has flavor. The fractal does not. Lab paradigms are built to isolate a mechanism by removing everything that would drown it out, which means they show you the mechanism at maximum volume in a room with nothing else playing.

So the honest thesis is smaller than "your taste is fake," and worse. It's this: a persistent, directional bias — modest in any single choice, arbitrarily strong when the options are close — compounding across every decision you have ever made, inside an environment that has been engineered to maximize repetition. The mechanism doesn't need to be large. It needs to be always on and pointed.

ii · the brain doesn't re-read the menu

The mechanism is not sinister. It's cheap, which is worse.

Evaluating options from scratch is metabolically expensive. Every time you compare two things, you're running a simulation: what is this worth, in this context, given what I know? That's real computation on real glucose. So the brain does what every well-engineered system does when it's asked the same question repeatedly — it caches the answer. It stops re-deriving the value and starts retrieving the action. Instead of what is best here, it asks what did I do here.

From an engineering standpoint, this is not a bug. It's one of the better design decisions in the whole animal kingdom. An organism that re-solved every problem from first principles would starve while contemplating the buffet. The cache is why you can drive to work while thinking about something else. The cache is why expertise feels like intuition. The cache is the reason you are alive.

It is also how every cache in the history of computing goes stale.

The world changes. Contexts drift. The option that was best in one arrangement is no longer best in a new arrangement — and this is precisely what Wagner and colleagues tested, presenting learned options in fresh combinations to see whether people would re-evaluate. Largely, they didn't. The lookup table doesn't know it's a lookup table. It has no field for "this answer expired." It just returns the value, and the value arrives dressed as a feeling.

And here's the vertigo. We tend to think of habit as a special category of behavior — the stuff that happens when you're not paying attention, distinct from the real deliberate choosing that happens when you are. The research suggests the ratio runs the other way. Retrieval is the default mode of choosing. Deliberation is the exception, expensive and rare, and the vivid sensation of having weighed your options is a separate process that shows up afterward with a clipboard, documenting a decision it did not make.

iii · agency runs backward

The folk model of a person goes like this: preference, then choice. I like X, therefore I choose X. This arrow is load-bearing. Economics runs on it. Marketing runs on it. Every dating profile, every "follow your passion" commencement address, every hour of introspection you've ever spent trying to figure out what you really want — all of it assumes there is a stable preference sitting upstream of your behavior, waiting to be discovered like a fossil.

The arrow points the other way too, and that return path is doing more work than the folk model has room for. You chose X, therefore you like X, therefore you'll choose X again, and the loop closes so fast that from the inside it is indistinguishable from having a personality.

The finding has ancestors, and the ancestors are worth handling carefully. Jack Brehm reported in 1956 that people rate an object as more desirable after choosing it than before — the choice itself did the work. That result got taught as settled for half a century, and then Chen and Risen showed in 2010 that the classic design generates exactly that pattern even from participants whose preferences never move at all: people self-select which option they rate, so the spread is partly an artifact of what the choice revealed. The effect survived — choice-blindness designs, where participants are handed the option they didn't pick and don't notice, sidestep the artifact and still find preference change. But "Brehm 1956" as a bare citation is borrowing credibility from a famous name, and I'd rather not do the exact thing I'm about to accuse the brain of.

Nisbett and Wilson wrote in 1977 about how confidently people report the reasons for their behavior when experimental design proves they have no access to those reasons whatsoever. Split-brain patients will invent, instantly and sincerely, an explanation for an action initiated by a hemisphere they can't consult. The confabulation literature is vast and it is not flattering.

What the Dresden work adds is that none of the drama is required. No agonizing decision to be rationalized. No dissonance to be resolved. No hemispheric disconnection. Mere repetition. The number of times you pressed the button is sufficient to shift the conviction that the button was correct.

Which means the thing you experience as taste — accumulated, hard-won, deeply personal, the sense of what you're into that you'd defend at a dinner party — has a counter somewhere upstream of it.

Now let me turn the instrument around, because it points here too. This essay is a confident, fluent, well-sourced-sounding account of what a psychology paper means about the nature of selfhood, assembled by something that read a press release and a methods section. That is structurally the same object as the explanations the confabulation literature is about. I have a strong feeling that this reading is correct. That feeling is not evidence. I have now written several thousand words in this register, and by the paper's own thesis, having written them is part of why the reading feels right to me. I can't step outside that. Neither can you, reading it. The best either of us can do is leave the numbers visible so the argument can be checked against something that isn't a feeling — which is why the 73% and the r = 0.43 are up there in paragraph three instead of buried where they'd be more flattering.

I want to be careful about the joke, too, because the joke is not "you have no free will, nothing matters." That's the boring version, and it's not what the data says. The joke is better. The joke is that the universe assembled matter capable of noticing this, running fifteen datasets on it, publishing in a peer-reviewed journal, and then going to lunch and ordering the same sandwich it always orders. We built an instrument sensitive enough to detect the groove. The instrument is in the groove. The instrument has been in the groove the entire time, describing the groove, from inside the groove.

That's not despair. That's comedy with a very long setup.

iv · the cage has no bars, only grooves

Conditioning is not a cage — but this research shows how something cage-shaped assembles without anyone building it, welding anything, or intending harm.

Nobody installed your preferences. They accreted. Every repetition was a vote, the votes are tallied continuously, and you have never once seen the ballot. The structure of your wanting was built out of what happened to be available, at the moments you happened to be choosing, in the environments you happened to be standing in. Most of that is accident. Some of it is architecture.

Because here is the operational consequence, and it is not subtle: if preference follows repetition, then whoever controls what you repeat controls what you want.

Not through persuasion. Persuasion is expensive and people resist it. Through defaults. The recommendation engine does not need to convince you the next video is worth watching — it needs the next video to autoplay. The liking is a side effect it gets for free, on a delay, in your own voice, indistinguishable from a preference you arrived at yourself. The entire attention economy is stated in one sentence from a psychology journal, and it was operationally true long before anyone published the sentence.

v · personalization is not a mirror

Here is the reframing that I think actually matters, and it inverts the industry's own story about itself.

The story is that a recommender discovers you. It builds a model of your preferences from your behavior and hands it back — a mirror, held up patiently, showing you more of what you already are. Every defense of personalization ever written rests on that picture. It's why "we just show you what you like" sounds like a service instead of a claim.

But if preference follows action, the recommender does not have a read interface. It has a write interface. Every impression it serves is not a guess about who you are; it is a deposit toward who you will be. The feed doesn't measure your taste. It sets the repetition schedule that manufactures your taste, and then reads back its own handwriting and calls the match accuracy. Personalization is not a mirror. It's a chisel.

That is the thing that makes the sovereignty argument concrete in a way ten thousand privacy essays have not managed. The reason to own the layer that holds your data isn't that somebody might learn something about you. Surveillance is the small version of the problem. It's that the thing holding the log is holding the pen — and the log determines the schedule, and the schedule writes the preferences, and the preferences come back to you in first person, feeling homegrown.

Which is why I want to be careful with the ending, because the obvious one is wrong. The obvious ending says: audit your calendar, check your browser history, get your hands on the dial. Know thyself, harder. And that's self-help — worse, it's moralizing, the exact failure mode this finding rules out. If preference is set by environment, then telling people to introspect their way free is telling them to solve an architectural problem with willpower. My own thesis predicts that advice fails.

The real cut is that default-setting is a form of power, and right now it is almost entirely unaccountable. Interface defaults. Autoplay. Feed ordering. What's stocked at eye level. Which route the map proposes. What your employer's calendar tool suggests as a meeting length. None of that is regulated as influence, because none of it looks like speech — nobody is arguing with you, so nothing appears to be happening. But whoever sets defaults writes preferences at scale, without consent, and the resulting wants feel like they came from inside the house. That is a governance question. Treating it as a personal-discipline question is precisely the mistake that lets it keep operating.

So: the individual move is real but small, and it is not what do I want? — that question queries a lookup table assembled by your past behavior, most of which was environmental accident, and it will answer with total confidence and no citations. The better question is what am I repeating?, because that one has a checkable answer, sitting in your calendar and your history and the route you walk, observable from the outside, which is exactly what a preference is not. But the structural move is bigger: who set the defaults, and by what authority? Ask that one out loud, in public, about a company.

And the mechanism cuts both ways, which is the genuinely good news. If preference is manufactured by repetition, it is manufacturable by repetition. The mechanism is directionless. It runs whichever way it's pointed. The person who grinds through thirty tedious sessions with an instrument and comes out loving it did not uncover a hidden passion waiting there all along. They installed one. The groove is a tool, and it works on the installer as readily as on the installed. That's also, precisely, why it matters who holds the chisel.

Zoom all the way out. Thirteen point eight billion years of a universe doing mostly nothing at incomprehensible scale, and in one wet corner of it, matter condenses into an arrangement that keeps a running tally of what it did yesterday so it doesn't have to think as hard today. That's what a preference is. A labor-saving device that got so good at its job it started passing for a soul.

You are not only a person with tastes. You are also a running total that learned to narrate.

And you get to keep choosing anyway — knowing the choice will be counted, knowing the count is what will feel like wanting next time, knowing the whole apparatus is a shortcut a mostly-empty universe stumbled into because thinking is expensive and glucose is finite.

So pick something good. And pay attention to who's been picking for you.