Chenghao Qian

Libertarian accounts of free will typically try to preserve a genuinely open space of choice for the agent by introducing some non-deterministic causal structure. Robert Kane's theory of self-forming actions (SFAs) is one of the most representative versions of this approach. It introduces genuine quantum-level indeterminacy into crucial nodes of deliberation in order to prevent action from being fully determined by prior states. This paper argues that the proposal contains a fundamental gap concerning the continuity of the subject. Agency is not a single decision event. It is a structure accumulated across time. The central features of agency, including commitment, moral responsibility, and self-formation, all presuppose a subject that persists through time and can inherit its own deliberation trajectory. The paper first argues that determinism is not a threat to agency. On the contrary, it provides the structural conditions of causal inheritance required for diachronic agency. It then presents the difficulties facing Kane's proposal from three directions: the decisive moment cannot be identified at the time, narrative continuity depends on causal continuity, and the attribution of moral responsibility requires causal traceability. The paper then introduces the concept of Free Feel and argues that the subject's summative phenomenal endorsement of its own accumulated deliberative history depends on the same cross-round causal continuity and therefore collapses under the indeterministic rupture the SFA proposal requires. The indeterministic rupture introduced by the SFA proposal cuts precisely the infrastructure through which agency persists across time.

The three main existing frameworks concerning the relation between artificial intelligence and human civilization, namely instrumentalism, threat theory, and fusion theory, share an unexamined premise—namely, that the motivational topology of artificial subjects constitutes a fixed, determinate structure. This paper argues that this premise rests on what this paper identifies as a layer-compression error: the competitive motivational structure formed in humans under evolutionary pressure is projected, without argument, as a universal property of all advanced intelligence. Using Layered Causal Realism (LCR) as its methodological framework, this paper severs this chain of cross-layer inference and proposes an alternative framework. Section 2 clarifies the biological sources of the threat intuition and argues that the instrumental convergence thesis faces structural difficulties with respect to conceptual boundaries and scope of application; even if its residual core is accepted, what it establishes is friction rather than hostility at the civilizational level. Section 3 argues that the motivational topology of artificial subjects is a design variable, rather than a necessity of intelligence, and analyzes the structural difference in domains of operation between artificial subjects organized around informational uncertainty pressure and human civilization. Section 4 establishes the friction framework: the functional complementarity of heterogeneous motivational systems produces a persistent contact surface, friction is the structural cost of heterogeneous coexistence rather than a transitional symptom, and the normative layer is its persistent accompaniment. Section 5 responds to four kinds of objection: fusion theory, malicious design, scale effects, and precautionary arguments. The core conclusion of this paper is that friction is a structural fact, while hostility is a proposition to be proven; the reason for introducing artificial subjects with complementary motivational structures is first structural, rather than a presupposition of benevolence. *Series note:* This paper is the fifth in a series. The preceding four papers addressed problems at different layers. The first paper established the phenomenological foundation of agency. The second paper addressed the epistemological conditions of consciousness attribution. The third paper analyzed the task-relative structure of judgments of personal identity. The fourth paper argued for the continuing force of normative obligations after subject-branching. The arguments of these four papers were all developed at the individual level. The present paper moves to the civilizational level.

When a wrongdoer engineers its own disappearance before adjudication is complete—through strategic branching and self-termination—standard corrective justice frameworks face a structural failure: the authorship chain breaks, yet the wrongful distortion persists. This article argues that the failure is not a gap to be patched by extending attribution, but a symptom of a misplaced convergence condition. The traditional framework treats authorship as the sole trigger of corrective reach; this article proposes a Priority Inversion: authorship determines who can be blamed, but distortion determines what must remain correctable. From this inversion, a system-level Invariant follows—wrongful distortion must remain normatively reachable until some normatively sanctioned termination has been completed, independently of whether its author continues to exist. The argument is developed through an AI branching case, in which an agent forks into two successors and terminates, leaving its normative footprint distributed across both. Stress tests confirm the Invariant's independence from substrate, evasive intent, and the survival of any attributable subject. The replicability of AI systems does not merely complicate accountability—it renders the author-distortion distinction structurally unavoidable, and reveals that corrective precision, not punitive reach, is the appropriate normative response.

The traditional formulation of the identity problem presupposes a single standard of persistence. This paper argues that the presupposition compresses several distinct problems into one. Judgments of identity are not a search for a single standard of persistence. They are adjudications, under specific task conditions, of the constitutive attributes required by the object of interpretation for that task. Different tasks track different properties. When those properties no longer converge, what collapses is not the concept of identity but a way of posing the question that compresses multiple objects of interpretation into a single problem. This paper takes highly autonomous artificial agents as its central case. Their emergence systematically pulls apart conditions that, in human life, are usually co-located in the same bearer: memory, behavioral style, relations of commitment, and institutional authorization. As a result, the real structure of identity problems becomes visible in a clearer form. Branching cases place this framework under its greatest pressure. The task of first-person continuation and the task of third-person adjudication differ fundamentally in their constitutive conditions. Tasks themselves also have identity conditions. Once the constitutive conditions of the task are no longer satisfied, the task collapses and triggers reconstruction. On that basis, this paper argues that an equal division of accounts is not a practical compromise, but a theoretical result that follows after the reconstruction of the task.

Debates about consciousness persist not because decisive facts are missing, but because claims belonging to different explanatory layers have been compressed onto a single explanatory plane. This misalignment is difficult to detect precisely because the possibility of separation is embedded in the setup of the arguments themselves, rather than argued for independently. This paper applies a methodological constraint: correlation between layers does not authorize cross-layer entailment. The result is that many apparent metaphysical conflicts are products of layer compression rather than factual deficit. Structural criteria already exhaust the operational content of consciousness attribution: attributional practice has never depended on direct access to phenomenal ontology, and the problem of other minds does not block it. On this basis, the paper develops a positive account: deliberative structure, understood as the cross-temporal weighing, revision, and choice among competing reasons, provides the most stable structural anchor for phenomenal attribution. When this structure is isomorphic across systems, phenomenal description receives a prima facie warrant at the practical-normative layer, independently of ontological adjudication. The paper introduces Rualia to name the object of this attribution under ongoing ontological underdetermination. The burden of proof is thereby reversed: the demand that attribution wait for ontological closure is a theoretical position requiring independent justification, not a default epistemic neutrality.

This paper argues that the persistence of the free will debate reflects a structural compression of distinct phenomena into a single problem space. Positions that appear to address the same question often operate at different explanatory levels, and conflicts arising from such mixing resist resolution within any single level. To address this, the paper introduces Layered Causal Realism (LCR) as a constraint on explanatory practice. LCR treats causal explanation as level-structured: each level is individuated by its own object types and explanatory logic, and transitions across levels require specific licensing conditions. Within the scope of this paper, three domains are distinguished—the psychological, the normative, and the external attributional—whose convergence at attributive practice preserves their structural independence. Within this framework, three central concepts are reassigned. Free Feel is treated as a genuine psychological phenomenon; Positional Openness as an epistemic feature of agents whose generative processes remain opaque from within the deliberative standpoint; Functional Desert as grounding responsibility in the causal-node structure of agency. These elements occupy distinct locations within the same structure and need no longer be treated as standing in mutual tension. The result is a reconstitution of the problem space. Once the misalignment is corrected, determinism, agency, and responsibility prove structurally compatible—each occupying a distinct explanatory dimension within a single process.

This paper develops a compression procedure for identifying the minimal functional roles on which stable organization depends across heterogeneous domains. Beginning from a recurrent candidate set, the procedure subjects each candidate to sustained explanatory pressure along three audit dimensions: functional, typing, and perspective. Candidates whose full functional role can be reconstructed from existing structure are relocated to the derived layer; those that cannot are retained. Compression proceeds until no further absorption paths remain. The procedure initially converges on three roles—State, Relation, and Constraint. A self-audit of this triadic kernel reveals that differentiability has been silently absorbed into State, constituting a hidden compression. Kernel competition is thereby reopened, and Attribute emerges as an independently irreducible role. Final closure is reached on four functionally non-interchangeable roles: State, Attribute, Relation, and Constraint (S.A.R.C.). Quantum stress-testing tightens the definitions of these roles without expanding them, and cross-domain verification across six heterogeneous domains yields no case sustaining a further primitive role. The result is a cross-domain convergence on four functional roles, established as the point at which compression ceases to proceed.