The Surprise Gym

In 1818, an English engineer named William Cubitt installed a long stepped wheel at Brixton Prison and set convicts climbing it. The treadwheel was designed as punishment — deliberately pointless at first, weighted for resistance, later hooked to millstones so that forty-four English prisons could at least grind corn with the misery. It was considered cruel enough that the Prison Act of 1898 abolished it. Eighty years of penal history, closed with relief.

Two centuries later, the same machine sits in a mirrored room three blocks from your house, and you pay monthly for access to it. The standard telling of this story is irony. The accurate telling is economics.

When machinery removed exertion from work — from farming, from hauling, from walking to the places work happened — exertion did not disappear from human need. It disappeared from the day. And anything the day no longer supplies, but the body still requires, becomes a product. The fitness industry is not a quirk of vanity. It is the market’s answer to a removal: effort, extracted from labor by machines, sold back by the hour, in calibrated doses, to people who chose it. Nobody in 1818 could have predicted that the descendants of factory workers would pay to climb Cubitt’s wheel. But the pattern, once you see it, was inevitable from the moment the removal began.

The same subtraction is now running against a different nutrient.

I have argued that learning is metabolized surprise — that your brain updates on exactly one signal, prediction error, and on nothing else. And I have argued that the personal stack’s whole mandate is absorption: remove the friction, remove the waiting, remove the error. Every surprise the stack absorbs on your behalf is a surprise your model of the world never gets to eat. The draft arrives finished. The route reroutes before you feel the traffic. The answer comes back before the question got hard. The day gets smooth, because smoothness is the product.

No one will vote for the end of surprise, just as no factory owner set out to make humanity sedentary. The removal completes anyway, because every individual choice points the same direction. Nobody buys the rougher assistant. Nobody asks their stack to let a few errors through for flavor. Absorption is not a conspiracy; it is a gradient, and everything rolls down it.

But the nutrient logic still holds. If the brain learns from prediction error and only prediction error, then a day stripped of prediction error is, neurologically, a fast. You can coast on stored competence for a while — the way a body coasts on muscle it built in harder years. Then the muscle goes. What the day no longer supplies, and the mind still requires, becomes a product.

Strip the mirrors and the branding, and a gym is a precise economic instrument: a market where a removed input is repurchased on purpose. Its product has four properties, and they define what any surprise gym will have to sell.

1. Voluntary

   You choose the exposure. The lift is heavy because you loaded the bar. Surprise you did not consent to is just life going badly; surprise you signed up for is training.

2. Calibrated

   The dose matches your current capacity. A rack holds every weight between too-light and injury, and progression is the entire craft. Random shocks are not calibration; a good opponent is.

3. Attributable

   When the prediction fails, you can locate the failure in your own model. You know which assumption broke, the way a lifter knows which muscle gave. Error you cannot trace teaches nothing.

4. Non-delegable

   The stack can book the session, drive you there, and log the results. It cannot metabolize the error for you — the moment it does, the training data lands in its model instead of yours. That is the one absorption you must refuse.

Read the list again and notice how many surprise gyms already exist, unlabeled. A ranked chess ladder. An improv class. A jazz jam. A crossword tuned to Saturday difficulty. A sparring round. A language exchange with a stranger. Travel booked with no itinerary past the first night. Every one of them sells prediction error at a chosen dose, against an opponent or an environment honest enough to break your model in a traceable way. We have never named the category, because surprise outside those rooms was still free. It is about to stop being free.

The market that emerges will immediately be flooded with a counterfeit, so it is worth fixing the assay now.

A slot machine is maximally unpredictable and teaches nothing. The reason is precise: you never had a model of the reels, so no model fails, so nothing updates. The machine delivers variance, not error. Feeds run the same trick at scale — an endless stream of things you didn’t know, none of which you were wrong about. Novelty without prediction is stimulation, not surprise.

Real surprise requires a prediction to break. Junk surprise only requires that you never made one.

The parallel to food is exact, and it is the reason the gym metaphor earns its keep. Sugar is energy stripped of the work of obtaining it, and an environment full of it produced a new kind of malnutrition — abundant calories, starved bodies. The feed is surprise-flavored variance stripped of the model that makes surprise nutritious, and it will produce the same figure: a person consuming novelty all day and learning nothing, stimulated and starved at once. The assay is one question, asked after the session: is your model of anything different now, or were you just emptied?

Here is the forecast, stated so it can fail. Within five years there will be a recognized product category — named, marketed, subscribed to — whose explicit pitch is calibrated prediction error. Difficulty as a service. Its premium tier will be human, for the reason I gave in Proof of Person: a machine can generate challenge, but a person across the table is the one opponent whose moves are guaranteed not to have been tuned to keep you comfortable. Coaches, tutors, and trainers will quietly re-describe their work — not as instruction, which the stack does better, but as calibration: choosing the weight your model should fail under next.

The leading indicator is the original gym. Two centuries after the exertion removal began, the buy-back market is still growing — industry revenue rose about 8 percent and memberships about 6 percent in 2024 alone. Buy-back markets do not saturate, because they grow with the removal, and the removal of surprise has barely started.

Already true

The exertion precedent completed: a punishment device became a hundred-billion-scale industry, still compounding. And the surprise buy-back already runs everywhere unnamed — ranked ladders, escape rooms, hard-mode games, improv, open-water swims. People demonstrably pay for calibrated failure. They just haven’t been told that’s what they’re buying.

What has to happen

Absorption has to finish its work: daily prediction error has to become genuinely scarce for the median person, not just smooth for the early adopter. And at least one mass product has to start advertising difficulty instead of ease — the moment the marketing flips, the category exists.

Where I am probably wrong

Surprise may not survive scheduling. Exertion doesn’t care whether you saw the barbell coming; surprise might — a shock you booked a slot for is half-absorbed already, and the frame may denature the dose. If, ten years on, the surprise market is still just games and gyms — if no one pays to be wrong about anything that matters — then the appetite died with the diet, and absorption removed not only surprise but the wanting of it. That would be worse news than anything I have forecast here.

Cubitt’s wheel needed eighty years of penal service and a full century of mechanization before anyone would climb it voluntarily. Surprise will not wait that long. The removal is faster this time, and so is the market. When the first storefront opens that sells you nothing but the reliable experience of being wrong, do not laugh at it. It is the treadmill, arriving on schedule.