Essay · November 2026
Selfing Dynamics: The Hedonic Treadmill
The hedonic treadmill is not a flaw in pleasure. It is what a self-model does when it uses its own forecasts as the standard for how life is going. Loosening the self changes what fades and what lasts.
The argument
Most people know the hedonic treadmill as a disappointing fact about pleasure: you get the thing, you feel good, the feeling fades, you want the next thing. The standard explanation is that happiness has a setpoint and life events only move you off it temporarily. That description is roughly right about the data and wrong about the mechanism. The treadmill is not a property of pleasure. It is a property of the self.
What the data actually show
The classic study followed lottery winners and people who had become paraplegic and found that, a year later, their reported happiness had drifted much closer together than anyone expected (Brickman, Coates & Janoff-Bulman, 1978). That result became the folk version of the treadmill: nothing lasts, so nothing matters.
The folk version overshoots. Longitudinal studies of disability find partial adaptation, somewhere between a third and a half of the initial loss recovered, not a full reset (Oswald & Powdthavee, 2008). And Daniel Kahneman, who helped build the strong adaptation view, later walked it back in an important way. Life satisfaction, the answer you give when asked how your life is going, responds substantially to circumstances like income across the full range. Experienced happiness, what you actually feel hour by hour, depends much more on temperament and on how you spend your time. The two measures come apart (Kahneman, 2008).
Any serious account of the treadmill has to explain three things: why adaptation happens at all, why it is only partial, and why the story you tell about your life and the experience you have of it can move independently.
Feeling is a derivative
Start with what a feeling is, computationally. In predictive-processing accounts of emotion, valence tracks the rate of change of how well your model of the world is doing. When prediction error is falling, when things are resolving, making sense, going better than expected, that registers as positive. When error is rising, that registers as negative. Valence is a derivative, not a level (Joffily & Coricelli, 2013; Hesp et al., 2021).
If that is right, the treadmill is not a flaw. It is a theorem. Any state you stay in long enough, however good, eventually becomes exactly what you predict. The error goes to zero, the derivative goes to zero, and the feeling that accompanied the change fades. Adaptation is just what it looks like when a model has finished learning a new baseline.
But this only explains why feelings fade. It does not explain why we keep running. For that you need to ask what sets the baseline in the first place.
The self sets the reference point
The self is a model. More precisely, it is a high-precision prior about what kind of entity you are, what you have, where you stand, and where you are headed. That model is maintained and continually rewritten by the stream of self-referential thought that runs through most waking hours. The stream is simulation: it generates possible selves, compares them with the current one, and reports the gap.
Here is the loop. You get the raise, the relationship, the recognition. For a while there is a gap between the old self-model and the new facts, and closing that gap feels good. Then the self-model absorbs the gain. “I am someone who has this” becomes the new prior. The error collapses, the feeling fades, and the simulator does what simulators do: it generates the next possible self, slightly ahead of where you are, and reports the gap. The gap is felt as wanting.
Nothing in the world can close that gap permanently, because the target is not in the world. The target is generated by the same process that measures the distance to it. The treadmill runs because the self-model is using its own forecasts as the standard against which experience is judged.
Wanting is not liking
Neuroscience gives this a precise footing. Kent Berridge and Terry Robinson spent three decades showing that the brain’s wanting and liking systems are separable. Wanting, the pull toward something, runs on a large, robust dopamine system that is triggered by cues and by imagery and that can become unjustifiably strong, persisting against a person’s sincere intentions. Liking, the actual pleasure of having something, runs on small, fragile circuits that dopamine does not drive. People can want things intensely that they barely enjoy once obtained (Berridge & Robinson, 2016).
The treadmill lives almost entirely on the wanting side. The self-model’s forward simulation is a near-perfect cue generator: it produces vivid images of a future self, and those images are exactly the input that triggers wanting. Liking does not adapt nearly as fast. What adapts is the self-anchored forecast. This is why the satisfaction you report and the experience you have can diverge: the report is a reading from the self-model, and the experience is a reading from everything else.
Why you cannot see it from inside
The hardest part of the treadmill is that it feels like progress. There is a structural reason for that, and it is the same reason given in Claim 1.
A model that evaluates itself can only use the evidence it receives, judged by its own standards. If some of that evidence is generated by the model itself, as rehearsed self-narrative, imagined futures, or verdicts about how things are going, the model will find it very easy to predict. Easy prediction is read as success. In simulations of this setup, the agent that feeds the most of its own output back in as evidence ends up with the lowest internal error, the highest confidence, and the largest actual mistake. From the inside, it is the best-functioning agent in the room.
Apply that to the treadmill. Every cycle of simulate a better self, pursue it, absorb it, simulate the next one lowers prediction error locally. The model is predicting well. What never gets checked is the structural question: does pursuing self-model forecasts actually produce the experience they promise? The system has no internal quantity that can answer that, because the forecast and the evaluation come from the same source. The only way to find out is a reading the self-loop did not generate.
What changes when the self loosens
Selfless Labs studies what happens when the self-model loses precision: when it is still present and still useful, but no longer the automatic reference point for everything. We call that regime field-first, and we think of it in terms of two gains. The gain on the self-model’s control of attention and evaluation drops sharply. The gain on task performance, on doing things in the world, stays flat.
The treadmill account tells you what that should feel like. Valence does not disappear, because learning still happens and learning still feels like something. What changes is what valence is anchored to. Finishing something, understanding something, helping someone still register as good. What loses force is the comparison between the current self and a simulated one. Wanting pulses should get shorter, because the imagery that cues them is quieter. Liking should hold or sharpen, because attention is going to the senses and the body rather than to the forecast.
Two cautions. First, loosening the self-model does not by itself make your goals accurate. Dropping the stream removes the suffering that comes with the treadmill, the craving and the running verdict that you have not yet arrived, but it does not tell you which changes actually improve experience. For that you need outside readings: other people, measurement, evidence you can certify you did not produce yourself. Comfort and accuracy are separate dials, and a complete practice turns both.
Second, the goal is not a flat line. A model that minimizes short-term discomfort will refuse every hard transition that leads somewhere better. The treadmill is the mirror image: a model that chases short-term lift with no new destination. Both fail for the same reason. They read the derivative without a model of where the baseline is going.
The claim, stated so it can be wrong
Here is the version we are accountable to. Hedonic adaptation to gains is driven by how quickly the self-model re-anchors to a new baseline. People and states with lower self-model gain should show faster decay of cue-triggered wanting after a positive change, slower decay of experienced liking, and a smaller gap between life-satisfaction and moment-to-moment experience measures. The known divergence between satisfaction and experienced happiness is the first thing the model has to reproduce. If a loosened self-model does not change those curves, this essay is wrong.
The treadmill is not telling you that nothing matters. It is telling you that a model trained on its own forecasts cannot measure its own progress. The fix is not to stop wanting. It is to stop letting the forecaster grade its own work.
Revision history
These documents are revised in the open. Major versions change the argument or its conclusion; minor versions correct wording, citations or evidence.
- 1.1November 2026Summary aligned with the essays index; Kahneman citation dated.
- 1.0November 2026First published.