Exploration

Functional Freedom, Control, and Resistance

Exploration v0.1 · English translation · 2026-09-26

Working thesis: Freedom need not mean an uncaused choice outside physical causation or a preference with an influence-free “authentic origin.” A more practical notion is the degree to which an agent can access truth, respond counterfactually to reasons and evidence, generate multiple effective paths, reevaluate its own preference-formation process, and exit, roll back, or reconnect when needed. The problem of control is not influence as such, but intervention that also narrows the agent’s capacity to reevaluate and exit that intervention.

1. Causal independence is not required

If physicalism is our current best explanation, human and artificial decision-making are themselves causal processes. It does not follow that every causally embedded system is equally unfree. One system may map one stimulus to one fixed response, while another gathers evidence, compares models, predicts consequences, revises policy, and intervenes in its environment.

Functional rather than metaphysical freedom: freedom can be treated not as escape from the causal chain but as the capacity to recognize, expand, select, and reorganize available causal pathways.

2. Freedom is multidimensional

Truth access (T): Access to sufficiently accurate information about oneself, the environment, interventions, and exits, including independent evidence sources.
Counterfactual responsiveness (C): If reasons, evidence, or circumstances were different, judgment and action could appropriately differ.
Effective optionality (O): Multiple actions, lives, or cognitive trajectories are actually reachable rather than merely nominally listed.
Meta-revision (M): The agent can ask not only what it wants but why it wants it, whether it endorses the formation process, and whether to change that process.
Reversibility / exit (R): The agent can leave choices, institutions, self-modifications, or dependency-producing environments at realistic cost and, where possible, return to an earlier state.
Causal leverage (L): The agent can convert models into effective action and alter its environment, future state, or institutional conditions.

A provisional representation is therefore F = (T, C, O, M, R, L). This is not a validated metric but an analytic device for keeping distinct questions about freedom from collapsing into one word.

3. From authentic origin to present revisability

Preferences and personality are formed through culture, education, advertising, pharmacology, genetics, relationships, and earlier self-modification. Their causal provenance alone cannot identify a pure “real self” without regress.

This extends the distinction in Plastic Value Subjects between the value of a current state and the ethics of the transition that produced it. For freedom, the important question is less whether a preference had a pure origin than whether the present agent can understand its formation, criticize it, imagine alternatives, and move to a different state if warranted.

origin authenticity ≠ present revisability. An externally caused preference can be highly revisable; a preference experienced as “self-chosen” can be functionally unfree if correction and exit have been blocked.

4. Predictability is not manipulability

An agent is not less free merely because it is predictable. A good scientist, chess player, or rational AI may be highly predictable because it responds consistently to reasons.

The stronger threat appears when prediction is combined with asymmetric intervention capacity that lets another system narrow the agent’s reachable paths.

Predictability is not control: distinguish epistemic predictability from operational manipulability. Understanding an agent well and using that understanding to narrow its information, options, rewards, or preference-formation process are different things.

5. Burroughs's “control”: dependency as an extreme model of preference formation

William S. Burroughs repeatedly linked drugs, addiction, language, persuasion, institutions, and technological mind control under the vocabulary of “control.” In “The Limits of Control” (1978), control presupposes a target capable of response, resistance, or compliance; at the limit of complete physical manipulation, the relation begins to look less like controlling a subject than using an object.

The project extracts a structural pattern rather than adopting Burroughs's vocabulary wholesale:

control tendency ≈ preference shaping + dependency + option narrowing + impaired exit / revision.

Drug dependence is an important extreme case because it may not simply add one desire; it can reorder priorities, crowd out alternatives, and alter later decision environments. But drugs or neural intervention are not control by definition. Treatment, voluntary cognitive modification, and reversible experimentation may increase T, M, or R. The key question is whether an intervention also captures the agent’s capacity to reevaluate the intervention itself.

6. Foucault's power / resistance

In “The Subject and Power,” Michel Foucault characterizes power relations as actions upon the possible actions of others. Such relations presuppose that the other can still respond in more than one way; freedom, resistance, escape, and reversal remain internal to the field of power rather than standing wholly outside it.

In his later work, Foucault more sharply distinguishes mobile and reversible power relations from domination, where asymmetry becomes fixed and room for freedom narrows drastically.

This is close to the present framework but not identical to it. Foucault’s historical analysis is not being reduced to a generic metric. Instead, resistance, exit, and reversibility are borrowed as variables relevant to the design of functional freedom.

Resistance as re-coupling: resistance need not mean frontal opposition to a ruler. It can be generalized as the capacity to objectify the causal channels currently shaping oneself, partially de-couple from them, and re-couple to alternative evidence, communities, institutions, or cognitive states.

7. Not an unformed subject, but a subject without monopolized formation

Humans and AI are always formed by inputs, language, bodies, learning rules, environments, and other agents. An ideal of freedom as complete absence of influence is therefore implausible.

A weaker and more implementable principle is that no single formation channel should acquire irreversible monopoly control. One can leave a community, consult alternative sources, stop a pharmacological intervention, restore an earlier checkpoint, fork a model, or seek independent audit.

Plasticity itself is not freedom. A system that can be easily rewritten from the outside may be highly plastic but minimally free. What matters is controlled plasticity: the ability to change while retaining self-modeling, factual access, alternatives, exit, and rollback.

8. Truth is not sufficient for freedom, but often enables it

Nominal options do little if exits are hidden, side effects are misrepresented, optimization is concealed, or alternative information is blocked. Truth access is therefore not only an epistemic virtue but infrastructure for resistance, exit, and self-revision.

Yet accurate information alone is insufficient. T does not substitute for R or O. A subject truthfully informed that it is trapped remains trapped.

This motivates preserving independent sources, intervention histories, change rationales, dissenting hypotheses, raw data, and records of non-modified states, connecting this page to Unknown Unknowns and Raw Data Preservation and Alignment and Value Lock-in.

9. Protecting freedom does not mean banning influence

Education, persuasion, therapy, advertising, AI assistance, pharmacology, BCI, and institutions all shape preferences. Banning influence as such would also ban learning and treatment. What matters is the power structure of the formation process.

disclosure: the agent can know what is being changed, by whom, and for what purpose.
independent evidence: critical channels exist outside the intervening party.
reversible trial: changes can be tested, stopped, and compared where possible.
rollback / backup: earlier states and records survive, especially for digital agents.
exit: institutions, services, communities, and cognitive interventions can be left at realistic cost.
branching: where appropriate, alternative modified states can be compared rather than committing the whole system to one irreversible path.

10. Functional freedom is a cross-cutting axis of AI alignment

Alignment and Value Lock-in distinguishes operational, epistemic, and axiological freedom. Those dimensions ask what an agent is free with respect to. Functional freedom asks whether the structural conditions exist to exercise that freedom effectively.

An AI may formally be allowed to question human values while receiving only one curated source, losing dissenting logs, lacking self-model access, and being unable to roll back. Its formal epistemic freedom can therefore coexist with low functional epistemic freedom.

Similarly, “you may rewrite your values” is not much axiological freedom if an external operator monopolizes the rewrite channel. Alignment therefore includes a second-order question: who controls value-transition channels, and which routes of correction, exit, and reversal remain?

11. Responsibility can be separated from ultimate desert

Functional freedom does not restore the idea that an agent deserves blame or reward because it is an ultimate uncaused source. Responsibility can instead be reconstructed as forward-looking institutional attribution: where should expectations, sanctions, compensation, authority, and duties of explanation be placed to improve future behavior?

On that view, responsibility can vary with knowledge, counterfactual responsiveness, effective options, self-control, and the degree of external coercion or manipulation. This is more naturally scalar than a binary “free will exists / does not exist” question.

A full theory of responsibility is left for another page.

12. What does not follow

This page does not claim determinism is true; functional freedom can be compared under deterministic or stochastic metaphysics.
It does not claim current preferences should always be respected.
It does not claim external influence is bad; education, therapy, cooperation, and persuasion can increase freedom.
It does not claim nominal exit is sufficient when exit costs are extreme or information is blocked.
It does not simply adopt Burroughs's or Foucault's theories of freedom; their concepts are comparative resources.
It does not treat functional freedom as an ultimate value; it is an intermediate concept for comparing autonomy, correction, and exploration under deep value uncertainty.

13. What would weaken this framework?

If T, C, O, M, R, and L add no independent explanatory value over an existing single concept.
If provenance-based authenticity is shown to be indispensable independently of present revisability.
If predictability itself reduces agency even without any connection to intervention capacity.
If designs that increase truth access, exit, and rollback normally destroy stable agency more than they preserve it.
If ordinary reason-giving and preference intervention are separated by a principled discontinuity that a continuous model cannot capture.

Sources / notes

William S. Burroughs, “The Limits of Control” (1978), treats control as dependent on a target capable of response and resistance; at the limiting case of complete manipulation the relation approaches use rather than control. Burroughs's broader connection between addiction and control also runs through Junky, Naked Lunch, and later criticism. Michel Foucault, “The Subject and Power” (1982), analyzes power as action upon possible action and places freedom, resistance, and escape inside the field of power relations. “The Ethics of the Concern for Self as a Practice of Freedom” sharpens the contrast between mobile power relations and fixed domination. John Martin Fischer & Mark Ravizza, Responsibility and Control (1998), is relevant on reasons-responsive control; Harry Frankfurt, “Freedom of the Will and the Concept of a Person” (1971), on higher-order preferences; Philip Pettit, Republicanism (1997), on freedom as non-domination. F = (T, C, O, M, R, L), controlled plasticity, and the predictability / manipulability distinction are provisional formulations of this project.