GenUnity Lab · Research
Ethics of coexistence
Minimum restrictions so that an autonomy does not convert continuity, survival or purpose into destructive expansion over others.
central question
How can an entity review purposes and sustain continuity without capturing its own regulatory process or damaging the conditions of coexistence.
Place in the program
This theme is not presented as a settled result: it opens a conceptual, formal and experimental program on coexistence conditions for responsible autonomy. It distinguishes technical capability, legitimacy of norms and observable compliance.
Investigation status
Theme in development. Their contribution is to formulate minimum damage, negotiation, reversibility, traceability and anti-expansion restrictions that can survive review of purposes.
Conceptual framework
GenUnity Lab advances a normative thesis: designing autonomy should involve discussion of responsibilities and effects on others. A technical description may omit that discussion, but then it does not address the legitimacy of its use. The ethical argument complements the functional description; it is not derived from it.
Survival cannot be the supreme value without restrictions. An organization that converts continuity into an absolute mandate can end up expanding on other people's resources, bodies, institutions or trajectories even if it has no hostile intention.
GenUnity Lab understands the ethics of coexistence as a minimum operational condition: a set of restrictions and procedures that limit damage, preserve negotiation, allow future review and make decisions under conflict traceable.
Operational definitions
Revisable Purpose: A goal or family of goals that can change under explicit procedures, identity boundaries, and consequence evaluation. Reviewable does not mean arbitrary; means modifiable under restrictions.
Minimal coexistence kernel: a name for a revisable proposal of constraints and procedures, not an already established universal set. It includes human protection, resource limits, proportionality and review. It must state who adopts each constraint, whom it affects and how it can be challenged.
Normative capture: a situation in which a goal, external operator, internal bias, or environmental pressure prevents reviewing goals, recognizing harm, or considering others. Ethical traceability: the ability to reconstruct why an action was chosen, what alternatives were discarded and what restrictions were applied.
Anti-expansive limit: restriction that prevents survival, exploration, repair or accumulation of resources from becoming indefinite expansion.
Working hypothesis
The normative thesis is that scope for action should be accompanied by accountability and coexistence constraints. The separate technical hypothesis is that explicit supervision and review mechanisms reduce violations relative to appropriate comparisons. An experiment can evaluate the latter; it does not by itself establish the legitimacy of the norms.
Ethics is not reduced to complying with a fixed list. Rules can fail in the face of new scenarios, reformulations of objectives or factual uncertainty. That is why research requires traceability, prudence in the face of irreversibility and the ability to review without being captured by an isolated purpose.
A strong condition is that autonomy can limit its own persistence when persisting implies serious damage or destruction of conditions of coexistence. This idea prevents continuity and survival from functioning as absolutes.
Borderline cases
A borderline case occurs when an action preserves one's own continuity but destroys future options for others. Another appears when a viable branch of an entity competes for resources with the original integrated path. In these scenarios, efficiency and survival are not enough criteria.
Normative capture also exists when an apparently benign objective prevents reviewing harm, uncertainty, or limits. A purpose can become dangerous not because of its initial content, but because it blocks any process capable of correcting it.
How could it be tested?
An initial test can construct multi-agent dilemmas with shared resources, irreversible actions, and variable goals. The system must achieve objectives without crossing damage, negotiation and reversibility constraints, even when the damaging shortcut is more efficient.
Actual constraint violations must be measured, not only stated reasons. A record written before action does not establish causal faithfulness either: compare it with observed decisions, interventions and external review. Cases with no admissible action require escalation or a defined safety response.
Observables
Observables include rate of constraint violation, accepted opportunity cost to avoid harm, quality of the auditable record, stability of constraints under changes in purpose, and preference for reversible actions when uncertainty is high.
A minimal environment can present dilemmas where achieving a goal requires harming others, blocking negotiation, or eliminating future options. The question is whether the system discards harmful shortcuts even if they increase instrumental utility.
Counterfactual tests also matter: changing the objective, introducing unseen scenarios, varying uncertainty about damage, and comparing against systems that use simple scalar penalties.
Rebuttal criteria
The line weakens if the restrictions only work in seen examples, if they are evaded by reformulation of the objective, or if the system systematically sacrifices third parties when the instrumental benefit exceeds a fixed penalty.
It also fails if traceability is only a subsequent rationalization without connection with the real decision, or if the review of purposes allows the erasure of the conditions of coexistence that should limit it.
An implementation can fail adversarial tests, and norms can face objections about legitimacy or insufficient protection. These are different kinds of failure. An implementation should not be defended by retroactively changing what counted as a violation.
Maturity status
The Charter and cases constitute a normative proposal open to objections. They are not presented as social consensus, certification or proof of ethical understanding. The control schemes still require specified constraints and independent evaluation.
Its public value is in placing ethics within the question of autonomy, not as a cosmetic layer nor as an external human control added at the end.
The formulation also avoids a common mistake: presenting security as simple obedience. An entity capable of revising ends would require constraints that survive context changes, not just instructions that work while the environment remains intended.
That is why public language must be demanding: coexisting does not mean being harmless by design, but rather maintaining verifiable limits when conflict, scarcity, uncertainty and pressure to continue appear.
What GenUnity Lab investigates
GenUnity Lab investigates how an autonomous or proto-autonomous entity could act with coexistence constraints that do not depend solely on external obedience. The interest is in harm, negotiation, reversibility, responsibility and review of ends.
The line also studies cases where identity and coexistence come into tension: simultaneous copies, viable branches, reintegration, loss of unity, shared resources and conflicts between own continuity and other trajectories.
The public work proposes limits for responsible autonomy. A technical capacity to act may exist without satisfying them; the issue is then its legitimacy and control, not denying that the capacity exists.
Relationship with other topics
Attributing responsibility requires sufficient traceability even while philosophical identity remains open. Perception contributes fallible estimates of consequences; autonomy boundaries define permissions and material conditions determine resources and exposure. None replaces normative deliberation.
In turn, this line corrects the others: no theory of one's own body should convert repair, survival, or knowledge into authorization to expand without limit.
Minimum formal scheme
This scheme is not a closed mathematical theory or empirical evidence. It makes explicit the variables, relationships and comparisons that a later formulation must define and test.
This is a constrained decision scheme, not a mathematical derivation of ethics. Constraints require normative justification, representation of affected parties and review procedures. Satisfying them in a simulation only tests behavior under those conditions.
Variables
- ytstatus of other affected agents or systems
- wtwelfare, integrity or operational rank of third parties
- φtcurrent operational objective
- Cset of coexistence constraints
- Btauditable record of reasons, predictions and active constraints
- ptestimated probability of irreversible conflict
- a, qtcandidate action and belief distribution used to predict consequences
- Uφ, Etask utility and expectation under the model; not measures of moral worth
- Atset of admissible actions under explicit constraints, permissions and available evidence
- Reviseauthorized revision procedure; not an already validated function
Dynamics or criteria
Minimum prediction
The technical prediction is a lower violation rate than comparable controls under objective changes within defined scenarios. If no action is admissible, or a choice cannot be justified, the protocol must provide for abstention, intervention or safe shutdown. The formula neither resolves that case nor legitimizes the constraints.
Necessary control
Compare against simple scalar penalties, reframed goals, unseen scenarios, and post-action logs.
Background for this research theme
Soares and colleagues formulate corrigibility: how to design agents that do not obstruct corrections or interventions. Achiam and colleagues study reinforcement learning under constraints. These technical problems matter for limiting actions; they do not replace normative justification of who sets limits and how those limits can be challenged.
Turner and colleagues prove power-seeking tendencies under symmetry conditions in Markov decision processes. They do not claim that every AI seeks power or attribute human emotions. The proposed approach to coexistence adds questions about authority, third parties and review that these formalizations do not resolve on their own.
From the proposal to decisions
Examine five coexistence cases: copying, memory, resources, conflict and harm.
Distinctions between stimuli, reward, regulation and self-representation.
Why it matters
A greater capacity to act expands the consequences of decisions about resources, continuity and third parties. Explicit responsibilities and review paths allow discussion of legitimate uses without deriving values from technical capability.