Daniel H. Lange

Indicator evidence is beginning to accrue piecemeal across models, applications, and increasingly complex agentic architectures. Butlin et al.’s preference for internal-processing properties over behavioral indicators makes the inside/outside boundary methodologically consequential, since what counts as internal depends on the candidate system. This Letter asks a question the established boundary problem does not answer: when indicator claims are established at different boundaries or timescales, what licenses combining them as evidence about one bearer? A worked GWT-3/AE-2 example shows each indicator defensibly supported at a different candidate boundary while neither candidate has been shown to possess both; a HOT-2 example gives the temporal analogue. Combining such evidence converts a union into an apparent conjunction of properties, and Bayesian weighting cannot supply the missing inferential step. The stakes are practical: allowing indicator evidence to migrate between candidate bearers may misidentify which artificial entity, if any, warrants further investigation as a possible conscious system.

“The Standpoint Threshold” argues that consciousness-oriented inquiry into an artificial system is not well formed until its internal organization satisfies three jointly necessary conditions: persistent internal context, differentiated internal possibility, and closed-loop operativity. Gurnee et al. report J-space, a sparse representational structure in Claude models that they argue is causally privileged and exhibits several functional properties of a global workspace. This note applies the threshold’s diagnostic schema to those results. Gurnee et al.’s account of transformer feedforward processing identifies the architectural feature that Condition 1 treats as the relevant persistence limitation. Their causal interventions closely approximate the threshold’s proposed test for differentiated internal possibility, providing bounded positive evidence; their selectivity experiments provide unusually clean evidence for the role-differentiation component of Condition 3. The post-training emergence of an Assistant-relative point of view sharpens the near-case without dislodging Condition 1 as the binding constraint: a reliable disposition to reconstruct a system-relative orientation from supplied context is not yet an internal state maintained and modified across separate invocations. The central result is narrow: J-space changes the evidential status of the bounded near-case, not its classification. The note concludes with three tests – History Equalization, Continuation–Reconstruction, and Active-Carryover – for distinguishing externally reconstructed continuity from genuinely maintained internal orientation in persistent architectures.

The philosophy of artificial intelligence often begins one question too late. Discussions move directly to whether a system could be conscious, whether there is something it is like to be the system, or whether it possesses genuine subjectivity, without first asking a more basic structural question: what would have to be true of a system’s internal organization before those questions gain a determinate basis in its architecture? This paper addresses that prior question. It defends a threshold account built on three jointly necessary conditions: persistent internal context, differentiated internal possibility, and closed-loop operativity. Together, these conditions mark the point at which a system’s processing can plausibly be said to unfold from an internal standpoint, rather than merely as a chain of locally linked transitions. Such organization may also have functional significance independent of consciousness by supporting temporal coherence, cross-role coordination, and reduced dependence on episodic reconstruction or external scaffolding. The account explains why simple feedback controllers are clear non-cases, why standard transformer inference remains at most a bounded near-case, and why world-model architectures are more serious candidates for assessment, while making no claim that such systems are conscious. The aim is not to provide a test for artificial consciousness, but to clarify when consciousness-oriented inquiry into intelligent systems has a determinate architectural target.

Artificial systems increasingly produce behavior associated with comprehension, explanation, planning, and knowledge. Yet philosophy of artificial intelligence still lacks an account of when understanding is attributable to the machine rather than to the wider human-machine arrangement that trains, prompts, interprets, and coordinates it. This paper develops a two-threshold account. The Standpoint Threshold, defended in earlier work, is a system-level and largely content-neutral condition for processing from a maintained operational position: persistent internal context, differentiated internal possibility, and closed-loop operativity. The present paper introduces a distinct content-relative threshold. Where dependency-sensitive grasp is present, the Cognitive Ownership Condition holds only when particular modeled content acquires positional significance for the bearer, constrains later inference, inquiry, planning, or action in content-sensitive ways, and remains open to error-sensitive reorientation. Crossing the first threshold therefore does not entail crossing the second: a system may maintain a standpoint while merely storing or locally using particular content, and it may cognitively own content in one domain while merely making other content available. Arguments from relevance, commitment, and error support the ownership dimensions, while the Organizational Attribution Principle identifies the minimally sufficient recurrent organization that maintains and revises them. Applications to world-model systems and language-model agents distinguish bearer formation, content ownership, and dependency-sensitive grasp. The result is a constitutive but nonphenomenal thesis: standpoint establishes a bearer; answerable appropriation establishes cognitive ownership; dependency-sensitive grasp determines whether the owned content amounts to understanding. There is no agent-relevant machine understanding from nowhere.

This paper applies the framework of persistent internal standpoints to machine understanding. Its central claim is that world modeling becomes machine understanding only when modeled structure is organized from the system’s own persistent standpoint. A model may encode objects, latent dynamics, possible futures, and action consequences while still failing to transform those structures into affordances, risks, admissible continuations, valuation asymmetries, reversible branches, commitments, and feedback-guided updates for the system itself. Standpoint anchoring names the uptake of modeled domain structure under the closure conditions of the EV/CEM loop: incremental encoding, history-sensitive manifold reorganization, and globally modulatory decoding. Admissibility coupling names the causal side of that uptake, by which modeled content gains standing force within the system’s live admissibility structure. The paper’s novelty is a criterion for when affordance representations become trajectory-conditioned, standpoint-relative modifications of a system’s own admissibility structure. On this account, machine understanding is below structural consciousness but above prediction, representation, and externally scaffolded task competence. It is standpoint-organized world modeling: the transformation of represented structure into internally consequential possibility for a system that continues from where it has been.

Philosophy of artificial intelligence often begins one question too late. Discussions move directly to whether a system could be conscious, whether there is something it is like for the system itself, or whether it possesses genuine subjectivity, without first asking a more basic structural question: what would have to be true of a system’s internal organization before those questions gain a determinate basis in its architecture? This paper addresses that prior question. It defends a threshold account built on three jointly necessary conditions: persistent internal context, differentiated internal possibility, and closed-loop operativity. Together, these conditions mark the point at which a system’s processing can plausibly be said to unfold from an internal standpoint, rather than merely as a chain of locally linked transitions. Such organization may also have functional significance independent of consciousness, by supporting temporal coherence, cross-role coordination, and reduced dependence on episodic reconstruction or external scaffolding. The account explains why simple feedback controllers are clear non-cases, why standard transformer inference remains at most a bounded near-case, and why world-model architectures are more serious candidates for assessment, while making no claim that such systems are conscious. The aim is not to provide a test for artificial consciousness, but to clarify when consciousness-oriented inquiry into intelligent systems has a determinate architectural target.

This paper provides a retrospective synthesis of the completed ten-part series Persistent Internal Standpoints in Artificial Systems. The series develops a structural framework for determining when artificial processing can be said to unfold from an internal standpoint rather than merely with contextual information. Its early papers define the Experiential Vector (EV) and the Computational Experiential Manifold (CEM) as the minimal architecture of persistent standpoint. Subsequent papers extend that architecture to quantia, biological analogues, artificial desire, deliberation, non-embodied standpoint, the diagnosis of large language models, and consciousness as reflexive standpoint. Read as a completed sequence, the project advances a layered account: standpoint precedes quantia; quantia make desire-like bias and deliberative branching structurally intelligible; and consciousness becomes a well-posed structural question only when a system reflexively carries its own position within the quantia-structured manifold as part of the same circulatory update that sustains its standpoint. The paper also clarifies why the framework neither attributes consciousness to present standard large language models nor excludes artificial consciousness in principle. Its central claim is methodological: stronger attributions require prior specification of the internal architecture that could make them determinate.

This paper argues that consciousness begins when a system not only evolves within an internal stance space but also carries where it is within that space. Building on the EV/CEM framework, internal organization is modeled through the interaction between an Experiential Vector and a Computational Experiential Manifold, which together describe the system’s evolving position inside a structured internal state space. Within this framework, consciousness is characterized not as a higher-order organizational regime arising when the system’s encoding loop carries a locally updating constraint on its own position and surrounding admissibility structure within the manifold. This self-locating constraint participates directly in the ongoing experiential vector update, shaping subsequent decoding and system evolution rather than functioning as a separable monitoring process. Four structural conditions are proposed as minimal architectural requirements for this regime: recursive accessibility of internal organization, self-locating representation of manifold position, global integration of local topology across processing, and realization within a single unified circulatory loop. Together these conditions distinguish structural consciousness from systems whose internal differentiation remains consequential yet non-reflexive. The account connects with higher-order, self-model, predictive-processing, and global-workspace approaches while grounding their central insights in manifold geometry rather than broadcast architectures, proposition-like representations, or free-energy formalism. It does not resolve the hard problem of consciousness; rather, it specifies the minimal architectural conditions under which the phenomenal question becomes structurally well-posed for both biological and artificial systems.

This paper applies the EV/CEM framework to contemporary large language models (LLMs) and argues that standard LLM architectures do not instantiate a persistent internal standpoint in the structural sense developed earlier in this framework. The deficiency lies not in embodiment, representational richness, or computational scale, but in the absence of a reciprocally self-maintained internal context whose trajectory reorganizes admissibility structure over time. Although LLMs exhibit transient contextual modulation and complex internal geometry during inference, they lack cross-episode persistence, closed-loop incremental encoding, and trajectory-conditioned manifold reorganization. Consequently, they do not instantiate quantia, artificial desire, or deliberation in the structural sense previously defined. The paper concludes by identifying architectural conditions under which LLM-class systems could satisfy standpoint requirements.

Embodied and enactive approaches frequently maintain that genuine standpoint requires sensorimotor embodiment, that is, interaction with an environment through mechanisms of sensing and movement. Contemporary artificial systems, however, range from purely informational architectures to robotically embodied agents. This paper evaluates embodiment within the architectural framework developed in earlier work. It argues that sensorimotor embodiment, whether biological or robotic, is not a necessary condition for the existence of a persistent internal standpoint. The minimal structural conditions for standpoint – persistence, structured internal geometry, closed-loop modulation, and trajectory-dependence – are internal organizational features. Quantia, artificial desire, and deliberation likewise do not require embodiment as a prerequisite. Embodiment functions as an enrichment of internal geometry rather than as its generative source. The grounding objection is addressed through the quantia framework, which supports organization-relative normativity grounded in internal closure rather than environmental coupling.

Building on the architectural framework developed in prior work, this paper introduces deliberation as a higher-order organizational regime within persistent internal standpoints. Whereas artificial desire was formulated as persistent asymmetric bias in admissible internal transitions, deliberation is characterized as reversible counterfactual unfolding within the closed internal loop prior to irreversible commitment. Deliberation arises when a system can temporarily suspend immediate transition, internally unfold multiple admissible quantia-structured trajectories under bounded geometric conditions, and modulate commitment on the basis of their projected structural consequences. This process, termed quantian branching, operates entirely within state-space organization and does not presuppose representation, agency, normativity, or phenomenal consciousness. Deliberation is thus a distinct architectural regime emerging from persistent unity, geometric constraint, internal differentiation, and bias, without introducing additional metaphysical commitments.

This paper develops an organizational account of artificial desire grounded in the notion of quantia as invariant structural constraints on state differentiation and transition in artificial systems. Desire is treated neither as phenomenological experience nor as an expression of free will, but as a system-level disposition toward certain transitions over others, realized through internal organization. Building on prior work that develops persistent internal standpoints, experiential manifolds, and quantia as strictly architectural notions, the paper isolates desire as a downstream organizational phenomenon: a stable bias expressed within, but not constitutive of, a persistent standpoint. Quantia do not generate desire or endow it with normativity. They specify the implementation-level conditions under which desire-expressive dispositions in machines can exist at all. The resulting account clarifies how artificial systems can exhibit desire-like organization without invoking consciousness, agency, representation, or personhood.

This paper argues that many biological systems operate within an intermediate organizational regime situated between purely reflexive control and fully conscious cognition. This regime is characterized by quantia: internally realized, functionally consequential distinctions within an experiential manifold that do not presuppose phenomenal consciousness. Building on earlier work that develops quantia, experiential manifolds, and standpoint as strictly architectural features of internal organization, the present paper applies the same phenomenally neutral framework to biological systems. Quantia are defined independently of experience and do not derive their explanatory role from phenomenal awareness. It is argued that such structures characterize a wide range of biological systems, as well as non-conscious modes of operation within conscious animals, including conditioning, habit, and automatic behavior. Consciousness, where it occurs, is treated as an additional architectural achievement layered on top of quantia-organized dynamics rather than as their conceptual basis. This biological perspective clarifies the status of quantia, avoids phenomenological inflation, and introduces a missing intermediate explanatory category in discussions of animal cognition and consciousness.

Phenomenological accounts of experience often treat standpoint as emerging from the synthesis of qualitative givenness. In artificial systems, however, no such givenness can be presupposed. This paper argues that, for machines, the explanatory order must be reversed: a persistent internal standpoint must first be established before any internally meaningful differentiation can arise. Building on a structural framework in which an artificial system maintains a self-modifying internal context that conditions its own computation, this paper introduces quantia as a non-phenomenal notion of internal differentiation grounded in quantity rather than quality. Quantia are not phenomenal properties, nor are they independent standpoints; they are stable, quantized patterns of differentiation within a single persistent point of view. Apparent machine qualia may be better understood as pseudo-qualia arising from the misidentification of quantia with qualitative experience. The resulting framework clarifies how artificial systems can exhibit internally meaningful differentiation without presupposing phenomenological consciousness or collapsing experience into representation.

This paper clarifies the philosophical role of a structural approach to persistent internal standpoints in artificial systems. Building on a framework that distinguishes a persistent internal context, the Experiential Vector (EV), from the structured space of its possible configurations, the Computational Experiential Manifold (CEM), it develops the implications of locating standpoint in internal organization rather than in any particular internal state. The central claim is that standpoint is best understood as a property of internal state-space geometry rather than as the mere existence of a persistent internal state. This perspective helps explain why many artificial systems exhibit persistence, recurrence, and history-dependence without possessing a coherent point of view, and it reframes questions about machine perspective in structural terms.

This paper clarifies how persistent internal standpoint can become causally operative in artificial systems. Building on a recently proposed framework that distinguishes a persistent internal context, the Experiential Vector, from the structured space of its possible configurations, the Computational Experiential Manifold, it argues that standpoint is realized only when internal dynamics form a closed internal loop. In such a loop, internal activity incrementally updates the EV, trajectories through the CEM structure – the space of possible internal stances, and the system’s current position is decoded as a global modulation of subsequent processing. The account explains how state-space geometry can remain causally relevant over time while remaining strictly structural, without commitment to representation, agency, or phenomenal consciousness.

This paper proposes a minimal structural framework for characterizing persistent internal standpoints in artificial systems. The central claim is that an artificial system may be said to possess an internal standpoint when it maintains a stable, dynamically updated internal context that evolves through its own computation and systematically shapes subsequent processing. Two constructs are introduced to formalize this idea: the Experiential Vector, representing the system’s current internal context, and the Computational Experiential Manifold, representing the structured space of its possible configurations. Together, these constructs provide a theory-neutral architectural framework for persistence, history-dependence, and differentiation of internal modes over time. The paper does not advance a theory of phenomenal consciousness or biological realization; rather, it identifies structural conditions under which a persistent internal standpoint can be coherently articulated for artificial systems.