Viability and autonomy

Conceptual framework

Agency is not enough to talk about autonomy. A system can execute plans, solve tasks, call tools, optimize rewards, and adapt locally without participating in maintaining the conditions that make its own operation possible.

GenUnity Lab distinguishes autonomy from absolute independence. No complex entity exists without dependencies: energy, materials, communication, care, environment, culture or infrastructure. The question is not whether there are dependencies, but which ones are visible to the system, which ones it regulates, and which ones remain imposed from the outside.

Here operational autonomy denotes a degree of regulation of actions and dependencies within an authorized scope. Viability helps describe it, but a simple controller can also maintain operating conditions. Response range, adaptation and intervention limits must be compared without inferring a will of its own.

Operational definitions

Viability variable: dimension without which the system loses organization, capacity for action or continuity. It may include power, physical integrity, thermal stability, connectivity, memory coherence, control error, resources, or minimal commitment preservation.

Self-maintenance: processes by which the system detects, prioritizes and corrects deviations that threaten viability variables. It is not simple external repair: it implies that the organization participates in recognizing and governing their care.

Limited self-modification: ability to change components, models, policies or configurations under constraints that preserve operational identity and auditability. Regulated dependency: external dependency that the system monitors and manages, instead of receiving it as an invisible condition.

Operational purpose: a priority or objective used to select actions. It may originate in design, learning or authorized revision. Internal origin does not establish a moral interest, and external origin does not prevent effective resource regulation. Calling it purpose assumes neither emotion nor experience.

Working hypothesis

The empirical hypothesis concerns a comparative advantage of particular regulatory mechanisms, not a definition of what counts as an autonomous entity. Remaining operational may enable learning and repair, but a task with a planned endpoint may require correct termination.

A system can coordinate partly conflicting priorities: task, repair, exploration and resource conservation. How it learns or revises them is a technical question; why they should count as interests of its own is a separate question. Neither multiple purposes in every entity nor the primacy of survival is assumed.

Self-modification is a possibility requiring limits and evaluation; it is not assumed necessary for all autonomy. A system can adapt through states or strategies without rewriting its rules. Any change must respect operating permissions and allow human intervention.

Borderline cases

A system that accomplishes a task while depleting the energy that keeps it operational can be efficient without being autonomous. A system that only conserves energy and avoids all risks can maintain operation without developing a significant trajectory. Autonomy is played in the coordination between these extremes.

Another borderline case is invisible dependency. If repair, energy, updating and purposes always come from an external infrastructure that the system does not model or negotiate, behavior may seem complex without its own regulation of viability.

How could it be tested?

An initial test can place agents in an environment with scarce resources, cumulative damage, secondary task, and repair costs. A points-only policy should be compared with one that preserves margin of viability and decides when to sacrifice immediate reward.

The hypothesis receives limited support if recovery and time within limits improve with equal budgets, without constraint violations or obstruction of authorized shutdown. Costs and cases where the simple control matches or exceeds the model must also be reported.

Observables

Observables include explicit maps of feasibility variables, time within operating conditions, margin against infeasibility boundaries, post-damage recovery, repair cost, and frequency of modifications that preserve or violate invariants.

A trial can compare task policies with and without an explicit viability model. It must avoid an advantage caused solely by giving one policy more information, computation or a more favorable objective. Operating time is reported alongside energy, task performance and adherence to limits.

The ability to ask for help, negotiate resources, suspend actions or change strategy when continuity demands it must also be observed. Regulated dependence does not eliminate cooperation; makes it explicit.

Rebuttal criteria

The advantage of an explicit model loses support if a reactive policy achieves equivalent results under the tested conditions. This does not negate all autonomy: it identifies a mechanism whose additional complexity was not justified.

It also fails if feasibility variables cannot be specified or measured, if self-maintenance does not improve continuity under perturbation, or if self-modification destroys invariants often incompatible with operational identity.

The external origin of a preservation objective does not by itself refute a regulatory mechanism. Discussing interests of its own requires an additional argument explaining why continuity would matter to the entity and not only to its operator.

Maturity status

Proposed distinctions separate action, viability regulation, self-maintenance and revision. No comparative results or validated autonomy scale are presented. These categories must be tested against simple systems to avoid calling every feedback controller autonomous.

Their next maturity depends on reproducible models: environments where energy, damage, resources and tasks conflict, and where different policies can be compared under the same budgets.

The public formulation must preserve a tension: autonomy is not efficient obedience, but neither is sovereignty without dependence. The serious point is to measure which dependencies are governed by the organization and which remain hidden in the environment that supports it.

What GenUnity Lab investigates

GenUnity Lab investigates how artificial intelligence could participate in maintaining its operating conditions without being reduced to an optimizing tool. The focus is on feasibility, repair, plural purpose and limited review.

The line also studies civilizational dependence: many current systems do not depend only on electricity, but on data centers, institutions, logistics, spare parts, networks and human maintenance. Naming these dependencies avoids confusing technical performance with autonomy.

Relationship with other topics

This theme connects continuity with recovery, perception with disturbance detection and material realization with energy and degradation. Ethics contributes constraints on what may be done to maintain operation; it is not a property inferred from survival.

At the same time, this line puts pressure on the others: a theory of identity that cannot survive real harm remains incomplete; a theory of perception that does not consider cost and feasibility remains abstract.

From the proposal to decisions

Examine five coexistence cases: copying, memory, resources, conflict and harm.

Distinctions between stimuli, reward, regulation and self-representation.