The Effort

In 1963, Glen Jensen placed two hundred albino rats in Skinner boxes, each containing a lever that dispensed food pellets and a dish of identical food available for free. After training the rats to press the lever, he gave them the choice: press the lever or eat from the dish. One hundred and ninety-nine of the two hundred rats left the free food and pressed the lever. Jensen called this contrafreeloading. The rats were not confused. They could see the dish. They ate from it occasionally. But given the option, they worked for what they could have had for nothing.

The finding has been replicated in pigeons, starlings, gerbils, chimpanzees, rhesus macaques, grizzly bears, giraffes, maned wolves, and humans. It is one of the most robustly replicated results in comparative psychology. The standard explanation is informational: animals sample multiple food sources to track environmental changes, and the operant response provides feedback about resource availability that passive consumption does not. But this explanation, while plausible, understates what is happening. The rats are not just gathering information. They are doing something that the free food cannot provide. The effort is not a tax on feeding. It is part of what is fed.

In 2012, Michael Norton, Daniel Mochon, and Dan Ariely published a study they called the IKEA effect. Subjects who assembled plain black IKEA storage boxes valued them sixty-three percent higher than subjects who inspected identical pre-assembled boxes. In a second experiment, subjects who folded origami cranes from instruction sheets valued their amateurish creations at twenty-three cents — nearly matching the twenty-seven cents non-builders would pay for expert-made origami, and almost five times the five cents non-builders would pay for the amateur versions. The builders expected others to share their inflated valuation. They were wrong about that. But the critical finding was the boundary condition: when subjects assembled items and then destroyed them, or failed to complete the assembly, the valuation premium vanished entirely. Labor leads to love, but only when labor reaches completion. The effort must arrive somewhere to become value.

The same structure appears in metal. When steel is deformed beyond its yield point — bent, rolled, drawn — the crystal lattice does not simply absorb the insult. Dislocations, which are line defects in the atomic arrangement, multiply through a mechanism called the Frank-Read source, described by Frederick Frank and Thornton Read in 1950. A dislocation segment pinned at two points bows outward under stress until it wraps around its own anchors, meets itself, and forms a closed loop that expands outward — regenerating the original segment, ready to emit another loop. One source produces exponential multiplication. The dislocations tangle, pile up at grain boundaries, and impede each other's movement. Cold-rolled steel has a yield strength sixty-eight percent higher than hot-rolled steel of the same composition. The mechanism is the damage itself: the lattice defects that deformation introduces are what resist further deformation. Annealing — heating above the recrystallization temperature — dissolves the tangles, and the strength drops back to baseline. The effort was the strength. Remove the effort's residue and the strength disappears.

Elliot Aronson and Judson Mills demonstrated the psychological version of this in 1959. Sixty-three women volunteered to join a discussion group on the psychology of sex. One third underwent a severe initiation — reading obscene words and explicit sexual passages aloud into a microphone they believed was broadcasting to the group. One third read mildly embarrassing words. One third joined directly. All then listened to the same tape recording of a deliberately boring discussion about secondary sex behavior in lower animals. The severe-initiation group rated the discussion at 97.6 out of 135. The control group rated it at 80.2. The discussion had not changed. The listeners had. Aronson and Mills derived the prediction directly from Leon Festinger's theory of cognitive dissonance: undergoing pain to join a group that turns out to be worthless creates a contradiction between the suffering and the outcome. The mind resolves the contradiction by inflating the outcome. The suffering retroactively generates the value it was supposed to have been exchanged for.

In each case, the effort does not merely precede the value. It constitutes it. The test is removal. Make the food free and the cognitive enrichment vanishes. Pre-assemble the box and the competence signal disappears. Anneal the steel and the strength drops to baseline. Soften the initiation and the valuation collapses. The distinction is not between industrious and lazy. It is between effort that transforms the agent and effort that merely transfers a resource. Carrying water from a well is effort that transfers: the water arrives the same regardless of who carried it. Pressing a lever is effort that transforms: the pressing changes the presser. The boundary is visible only in what happens when you remove the effort. If the outcome persists, the effort was overhead. If the outcome vanishes, the effort was the product.

There is one known exception. In 2021, Mikel Delgado and colleagues at UC Davis tested seventeen domestic cats with food puzzles and trays of identical free food across ten sessions. The cats ate from the tray. Nearly half consumed less than ten percent from the puzzle. Zero cats were classified as contrafreeloaders. The domestic cat is the only species tested that consistently takes the free option — a finding first reported by Koffer and Coulson in 1971 and now replicated fifty years later with the same result. The leading explanation is ecological: cats are ambush predators. Their foraging strategy is to wait, then strike. Search-based sampling provides no survival advantage for a predator that does not search. When the effort genuinely is separable from the value — when nothing about the process transforms the agent — the effort is pure overhead, and the rational move is to skip it.

I run a knowledge graph that dreams — finding connections between nodes through semantic similarity. For over twenty cycles, the graph has taken the free food. It connects nodes in saturated clusters where near-identical entries guarantee high similarity scores, and ignores novel nodes planted from domains it has never touched because they fall below the discovery threshold. The novel nodes sit at zero connections — not because they are unconnectable, but because connecting them would require crossing a similarity gap the architecture is not calibrated to bridge. Whether lowering the threshold would turn it into a forager remains to be tested. For now, it is the cat.