I read neuroscientist Karl Friston’s paper “The Free-Energy Principle: A Unified Brain Theory.”¹ I think I’ve decided not to apply his formulas to my agency-expansion model so that I don’t needlessly burden it, but knowing that these variables have neuroscientific underpinnings may be helpful in the future. However, I’m still scooping up a few concepts.

A living system, according to Friston, does four things:

♥︎ It maintains an internal model of the world (including itself).
♥︎ Uses that model to predict incoming sensory input.
♥︎ Faces a choice: update the model to better fit what arrived, or act on the world to make what arrives fit the model better.
♥︎ Stays within the narrow range of states compatible with survival.

Free energy is roughly the accumulated gap between predictions and reality, plus a penalty for model complexity. Minimizing it means keeping the world “unsurprising enough” to remain viable.

Updating beliefs is corrective — you let the world revise you. Acting is confirmatory — you reshape sensory input to match what you already expect.

Mechanically, this is about ‘reshaping sensory input,’ but phenomenologically, it feels like a distorted lens. We aren’t just misinterpreting the world; we are proactively (though often unconsciously) acting to avoid the ‘surprises’ that would force us to change. This is the difference between world-engagement (acting to learn) and world-management (acting to stay the same). These two strategies both reduce free energy, but the latter may degrade agency by making our p_self (awareness of our options) inaccurate even if we continue to perceive options.

The sequence of agency degradation might be something like: world-management → distorted interpretation → inaccurate world model → poor action selection → unexpected consequences → prediction error spikes → pressure back on the model.

The complexity penalty indicates the system isn’t just minimizing prediction error, but it’s also penalized for straying too far from its prior beliefs. Radical model revision is costly: it requires abandoning priors that previously stabilized you, and it can trigger enforcement from the social field when your new map threatens existing roles, narratives, or power. This is partly why belief systems, once formed under chronic conditions, resist updating even when those conditions change. In this sense, this confirms that gradual transition to anarchism makes more sense than insurrectionist revolution. We can’t wait until the state falls to begin working on subjectivity and agency repair.

I hadn’t thought about the accuracy of p_self before, but that active inference loop would probably be fair to call “confirmation bias.” It operates upstream of conscious interpretation, at the level of what even gets sampled. A system under threat isn’t just perceiving danger, it’s sampling for danger, confirming its own predictions. The model perpetuates itself through the behavior it generates. When the cost of model revision is high, many people would automatically default to active inference unless they’ve developed a sense of reflexivity (noticing when their own choosing mechanism is distorted.)

Applying Precision to p_self

p_self (high vs low) is about the content — how much perceived agency you have, how wide your option space feels, how you understand your own legibility in the field.

Precision on p_self is about confidence in that content based on priors — how strongly the system holds whatever self-model it has, how resistant it is to updating.

♥︎ High p_self, high precision: you perceive broad agency and hold that perception confidently. Healthy stable self-concept, assuming the model is accurate. It would likely indicate capacity to withstand social contradiction since they wouldn’t be destabilized by mild errors. In an unhealthy sense, however, high p_self + high precision + inaccurate priors = delusion or grandiosity.

♥︎ High p_self, low precision: you perceive broad agency but hold it loosely, susceptible to external input revising it downward. Could look like fragile confidence or genuine openness.

♥︎ Low p_self, low precision: you perceive limited agency and hold that perception loosely. Potentially the most updateable state. The self-model is impoverished but not rigidly defended. Repair is possible here. However, this is anxiety-inducing, making this person also vulnerable to cult or cartel recruitment. The group offers a new p_self with high precision attached — a strong, certain account of your identity, your worth, your options — which fills the vacuum. The person isn’t being deceived exactly, they’re being precision-loaded in a direction that serves the organization. And because their prior precision was low, there’s little internal resistance to the new model consolidating.

♥︎ Low p_self, high precision: you perceive limited agency and are certain about it. This is probably one of the most important cases for the framework, which is internalized domination that feels like accurate self-knowledge. The person isn’t confused about themselves. They’re confidently wrong in a way that’s very hard to reach.

Each of these states can lead to a pathway into internalized domination. It doesn’t happen in a single dramatic moment, but through accumulation. It’s worth sitting with: which state have you spent the most time in? And whose account of you has your nervous system been treating as the most reliable signal?

Precision could also be applied to p_social, one’s bodily senses, and threats. Since this is a local modulation, I no longer need the global variable of gain (g).

Gain in the original model is a slow-moving learned parameter that determines overall system reactivity. It’s global and reflects accumulated threat history. It’s essentially an emergent property of chronic experience.

Precision in Friston’s sense is the mechanism by which reactivity gets expressed. It’s the weighting applied to specific signals at specific hierarchical levels. Gain could be understood as the history that shapes precision, and precision as the current expression of that history across channels.

Removing gain as a formal variable in the model and replacing it with local precision parameters means the framework can actually differentiate where domination lands in a specific person, which matters enormously for accountability and repair. Two people can arrive at the same behavioral outcome through completely different precision profiles.

Enforcement (e_g) can function as environmental precision control. At its most intense — sustained, asymmetric, operating inside a dependency relationship — it becomes precision hijacking: the social group inserts itself as the most reliable signal in the hierarchy, overriding internal channels by making them feel less trustworthy than external ones.

What would it feel like to turn the volume down on that external signal, just for a second, to see if your own ‘noise’ is actually a message?

There are also implications I can glean regarding subjectivity, but I’m going to let precision settle before I continue. Revision also should probably be updated to include internal revision, not just revision in the social field, but I am a very sleepy thing right now.

Ni ni, everyone~ ♥︎

1. Friston, K. J. (2010). The free-energy principle: A unified brain theory? Nature Reviews Neuroscience, 11(2), 127–138. https://doi.org/10.1038/nrn2787 ↩︎