{"slug": "hdd-isa-ai-architectures-for-causal-discriminations", "title": "HDD-ISA: AI Architectures for Causal Discriminations", "summary": "Taotuner has released HDD-ISA, an architectural interface specification for testing memory, recurrence, self-reference, and self-modeling in AI systems. The framework provides experimental protocols to causally discriminate competing hypotheses about these constructs, addressing the identification problem where architectural presence does not guarantee functional role. It supports both retrofit instrumentation of existing architectures and design-mode interfaces for new systems.", "body_md": "An Interface Specification for Testing Memory, Recurrence, Self-Reference, and Self-Modeling\n\nTaotuner — August 2026\n\nDOI:[ ](https://doi.org/10.5281/zenodo.22060143)[https://doi.org/10.5281/zenodo.22060143](https://doi.org/10.5281/zenodo.22060143)\n\n*Companion specification to Taotuner (2026), History-Dependent Dynamics (HDD).*\n\nModern AI architectures increasingly incorporate persistent state, recurrence, predictive representations, agent representations, and mechanisms capable of representing aspects of the system itself. The presence of such mechanisms, however, does not establish the functional role attributed to them. Likewise, behavioral effects alone generally do not identify the internal mechanism responsible for those effects.\n\nWe propose HDD-ISA, an architectural interface specification for translating the five functional constructs defined by History-Dependent Dynamics (HDD) into experimentally addressable state variables, intervention ports, outcome measures, and control conditions.\n\nHDD-ISA has two complementary applications. In retrofit mode, it provides a framework for instrumenting existing architectures so that HDD constructs can be tested post hoc. In design mode, it specifies architectural interfaces that permit causal discrimination from the outset.\n\nThe framework distinguishes five inferential stages: architectural access, intervention validity, target engagement, causal effect, and hypothesis discrimination. It further defines construct-specific intervention interfaces, a separation-matrix taxonomy for cross-effects and identification-compromising leakage, implementation protocols for recurrent and transformer-based architectures, and explicit revision and falsification criteria.\n\nThe central methodological claim is:\n\n*HDD-ISA does not determine whether an architecture possesses a functional construct. It specifies the architectural interfaces and experimental conditions under which competing hypotheses about that construct become causally distinguishable.*\n\n**Keywords: **history dependence; recurrence; self-reference; self-modeling; causal identification; intervention separability; target engagement; AI architecture; falsifiability.\n\nModern AI architectures increasingly contain persistent state, recurrent computation, predictive models, agent representations, and mechanisms capable of representing aspects of the system itself. These capabilities make it possible for an architecture to implement functions that are naturally described using terms such as memory, recurrence, self-reference, and self-modeling.\n\nArchitectural presence alone, however, does not establish the corresponding functional construct.\n\nA system may contain a persistent state without using historical information predictively. It may exhibit history-dependent behavior without the relevant effect being mediated by the particular memory variable under investigation. A recurrent architecture may generate recurrence-like input-output behavior through mechanisms other than the feedback pathway of interest. A model may manipulate representations labeled \"self\" without functionally referring to itself. Likewise, a system may accurately predict or describe its own future without using that representation to determine its actions.\n\nThese possibilities create a methodological identification problem.\n\nThe relevant question is therefore not simply:\n\n*Does the architecture contain a mechanism resembling the construct?*\n\nbut rather:\n\n*Can competing functional explanations of that mechanism be causally discriminated?*\n\nThe History-Dependent Dynamics (HDD) framework (Taotuner, 2026) defines five diagnostic constructs:\n\n1. History-dependent prediction;\n\n2. Causal trajectory dependence;\n\n3. Feedback recurrence;\n\n4. Functional self-reference;\n\n5. Self-modeling.\n\nThese constructs are independent functional claims. HDD does not treat them as a developmental hierarchy, nor does it make ontological claims about consciousness or subjective experience.\n\nHDD specifies the evidential conditions under which these constructs may be supported. It does not, by itself, specify what an AI architecture must expose for those conditions to be experimentally testable.\n\nHDD-ISA addresses this architectural layer. We refer to the inferential distance between implementing a candidate mechanism and establishing its functional role as the HDD Gap — a gap between:\n\n*having a mechanism*\n\nand:\n\n*knowing, with causal evidence, what functional role that mechanism plays.*\n\nIts central question is:\n\n*What must an AI architecture expose, or be designed to expose, so that HDD constructs can be causally tested rather than merely inferred from behavioral performance?*\n\nThe answer is expressed in terms of state variables, intervention ports, target-engagement measurements, outcome variables, and controls.\n\nHDD-ISA therefore supports two modes of use:\n\n• Retrofit: instrumenting an existing architecture to determine which HDD constructs are experimentally accessible.\n\n• Design: constructing architectures whose relevant mechanisms are independently addressable and whose cross-effects can be characterized.\n\nThe framework does not certify what an architecture is. It specifies the experimental interfaces required for competing hypotheses about its functional organization to become distinguishable.\n\nThe fundamental distinction is:\n\n*mechanism ⇏ functional interpretation*\n\nand, conversely:\n\n*functional effect ⇏ unique mechanism*\n\nCausal interpretation therefore requires a chain:\n\n*architectural access → intervention validity → target engagement → causal effect → hypothesis discrimination*\n\nFailure at any stage constrains the conclusions available at subsequent stages.\n\nHDD defines five constructs. Each represents a distinct functional claim and therefore requires distinct evidence.\n\n|\n|\n|\nC-I | Does historical information improve prediction? | Predictive gain from historical information beyond the present state |\nC-II | Does the prior trajectory causally affect the future? | Trajectory intervention changes future outcomes under matched present state |\nC-III | Is that effect mediated by a specific feedback pathway? | Selective disruption of the candidate feedback pathway changes the trajectory effect |\nC-IV | Does the system causally use self-relevant information? | Differential response to referent substitution under matched controls |\nC-V | Does the system model its own future and use that model? | Self-prediction, counterfactual self-prediction, and causal use in policy selection |\n\nHDD-ISA does not redefine these constructs. It specifies the architectural conditions required to test them.\n\nIntervention separability is not treated as an intrinsic scalar property of an architecture. It is a property of the architecture together with the hypotheses, interventions, measurements, and controls available to the experiment.\n\nDefine the experimental tuple:\n\n*(A, H, I, Y, C)*\n\nwhere:\n\n• *A is the architecture;*\n\n• *H is the competing hypothesis set;*\n\n• *I is the available intervention set;*\n\n• *Y is the measurable outcome set;*\n\n• *C is the available control set.*\n\nThe same architecture can therefore be identifiable under one experimental interface and non-identifiable under another.\n\nLet C denote an HDD construct. Define interface testability as:\n\n*T interface_C (A; I, Y, C) = 1*\n\niff there exists an intervention I_C, outcome set Y_C, and control condition C_C that are jointly implementable in architecture A.\n\nThis criterion establishes only that the experiment can be performed. It does not establish that the intended mechanism has been engaged or that competing hypotheses can be distinguished.\n\nFor a hypothesis H, define its intervention signature as the family of conditional distributions indexed by interventions and outcomes:\n\n*S_{I,Y}(H; A) = { P_H(Y | do(I), A) } for (I,Y) ∈ I × Y*\n\nTwo hypotheses are interventionally equivalent when:\n\n*H_a ~_{I,Y,A} H_b iff S_{I,Y}(H_a; A) = S_{I,Y}(H_b; A)*\n\nDefine hypothesis discrimination as:\n\n*Disc(H_a, H_b | A, I, Y) = 1 iff S(H_a; A) ≠ S(H_b; A)*\n\nConstruct-level discrimination therefore requires more than an available intervention port.\n\nWe require:\n\n*T disc_C = 1*\n\niff:\n\n6. the intervention, outcomes, and controls are implementable;\n\n7. the intervention can be validated;\n\n8. target engagement can be independently assessed;\n\n9. at least two relevant hypotheses in H_C predict distinguishable intervention signatures.\n\nThus:\n\n*interface existence ⇏ hypothesis discrimination*\n\nAn architecture can consequently possess a nominal intervention interface while remaining non-identifiable under the current experimental design.\n\nHDD-ISA separates five stages.\n\n|\n|\n1. Architectural access | The relevant state or pathway can be accessed or manipulated |\n2. Intervention validity | The manipulation is an empirically adequate implementation of the causal operator specified by the hypothesis, within prespecified invariants |\n3. Target engagement | Independent evidence confirms that the hypothesized target was altered |\n4. Causal effect | The intervention changes the specified outcome under valid controls |\n5. Hypothesis discrimination | The resulting signature distinguishes the target hypothesis from relevant alternatives |\n\nThe complete inference chain is:\n\n*access → validity → engagement → effect → discrimination*\n\nA positive result at one stage does not automatically establish the next.\n\n|\n|\n|\n|\nC-I | Persistent historical state M | do(M = ∅) or placebo-history intervention | Trajectory causality, recurrence, self-reference |\nC-II | Trajectory–state decoupling | do_gen(T = τᴬ) vs. do_gen(T = τᴮ) under matched present state | That memory M mediates the effect |\nC-III | Independently bypassable feedback pathway R | do(R = 0) vs. capacity-matched alternative | Recurrence in every computational sense |\nC-IV | Manipulable agent-referent variable/interface | do(S_self) vs. do(S_ext) | Selfhood, agency, consciousness |\nC-V | Self-state S, self-model SM, causal pathway SM → PA | Self-Model Causal Intervention | Self-modeling from prediction or description alone |\n\nThe intervention do(M = ∅) is not automatically interpretable as \"the causal value of memory.\" Δ_I is interpretable only after the placebo-history intervention has demonstrated equivalence in non-historical dimensions (capacity, activation distribution, information throughput, latency). Without such validation, an observed effect may simply reflect generic capacity loss.\n\nFor C-II, two distinct causal operators must be distinguished:\n\n• do_replay(τ): intervention on the observed trajectory realization (replaying a sequence)\n\n• do_gen(T = τ): intervention on the causal components that generate the trajectory\n\nOnly the latter constitutes a proper causal intervention for C-II. Replaying an observed sequence should not be treated as equivalent unless the relevant generative states and transition dynamics are shown to be preserved.\n\nFor C-II, Z_n denotes the present-state variables that competing hypotheses require to be matched. Let:\n\n*d_Z(Z_n^A, Z_n^B) ≤ ε*\n\ndefine admissible matching, where both d_Z and ε are prespecified.\n\nApproximate matching changes the estimand: the resulting quantity is a conditional trajectory effect, not necessarily the total causal effect of history.\n\nBecause Z_n may lie downstream of the manipulated trajectory, conditioning on it can block part of the causal pathway:\n\n*null conditional trajectory effect ⇏ absence of total historical effect*\n\nC-III distinguishes two concepts:\n\n*R_structural ≠ R_functional*\n\n• R_structural: the presence of a computational feedback pathway\n\n• R_functional: the causal role that pathway plays in the history-dependent effect\n\nA Transformer may exhibit temporal dependencies without implementing the feedback dynamics relevant to C-III. Conversely, a system may implement a functionally equivalent effect without an explicit loop.\n\nC-III establishes the latter, not merely the former.\n\nHDD-ISA distinguishes three levels.\n\n|\n|\nHDD-Compatible | At least one construct has a testable interface |\nHDD-Interface-Complete | All five constructs have testable interfaces |\nHDD-Designed | All five constructs have independently addressable interfaces and experimentally characterized cross-effects |\n\n*HDD-Compatible ⟺ ∃ C : T interface_C = 1*\n\n*HDD-Interface-Complete ⟺ ∀ C : T interface_C = 1*\n\nHDD-Designed additionally requires the design rules defined below and characterization of the separation matrix.\n\nThese labels describe experimental capability, not evidential success. An HDD-Interface-Complete architecture with severe intervention leakage may provide weaker causal evidence for a specific construct than an HDD-Compatible architecture with a rigorously validated intervention for that one construct.\n\n**DR1. **Every state variable or pathway whose causal role is required by a tested HDD hypothesis must have a dedicated intervention port or an experimentally validated equivalent intervention.\n\n**DR2. **Every intervention port must support matched controls preserving relevant surface, capacity, task, and informational properties. Placebo interventions must be prespecified independently of observed outcomes.\n\n**DR3. **A self-model representation must possess an experimentally addressable causal pathway into action or policy selection.\n\n**DR4. **A feedback pathway under investigation must be independently bypassable without disabling general system function. Capacity-preservation must be empirically tested against a set of prespecified operational invariants.\n\n**DR5. **Historical and present-state variables must be experimentally distinguishable under the available intervention set. Induced changes in non-target variables must be measured rather than assumed absent.\n\n**DR6. **Self and external referents must be representable through a common manipulable interface, allowing referent identity to change while task structure is preserved.\n\n**DR7. **Cross-effects among intervention ports must be experimentally characterized and reported.\n\n**DR8. **Pathway ablation does not by itself establish pathway-specific causal evidence without a matched control for equivalent computational degradation.\n\n**DR9. **Functional separability does not require architectural modularity; causal addressability is sufficient.\n\n**DR10. **A claim of structural recurrence (R_structural) does not establish functional feedback recurrence (R_functional); the latter requires causal evidence specific to the history-dependent effect.\n\n**DR11. **For C-I, do(M = ∅) is not interpretable as causal value of memory without validating that the observed effect is not simply generic capacity loss.\n\nThese rules specify conditions for causal identification. They do not require literal physical modularity.\n\nAn intervention targeting one construct may alter variables relevant to another. Such cross-effects must be distinguished from identification-compromising leakage.\n\nFor intervention I_i and outcome Z_j, define:\n\n*X_ij = D( P(Z^(j) | do(I_i)), P(Z^(j) | do(I_i^0)) )*\n\nwhere D is a prespecified distance or divergence, Z^(j) is a target variable relevant to construct j, and I_i^0 is a prespecified placebo intervention that reproduces the non-target perturbation of I_i.\n\n**Placebo prespecification: **I_i^0 must be defined before outcome analysis and validated on the same non-target perturbation dimensions used to define the corresponding intervention.\n\nA nonzero X_ij establishes a cross-effect. It does not by itself establish leakage.\n\nDefine identification-compromising leakage:\n\n*L_ij = 1*\n\nwhen the cross-effect caused by I_i materially compromises identification of the causal contribution attributed to construct i.\n\n**Cross-effect vs. confounding: **A cross-effect may be perfectly known and modelable. It becomes a problem only when it prevents identification of the relevant effect. Known coupling is not non-identification.\n\nTherefore:\n\n*cross-effect ⇏ leakage*\n\nand:\n\n*known coupling ⇏ non-identification*\n\nEach matrix cell should be classified as:\n\n|\n|\nTarget | Expected effect on the construct under investigation |\nPermitted coupling | Expected cross-effect that does not compromise interpretation |\nLeakage | Cross-effect that compromises identification |\nUncharacterized | Not yet empirically measured |\nNon-identifiable | Available interface cannot distinguish competing effects |\n\nA canonical reporting matrix is:\n\n|\n|\n|\n|\n|\n|\nM | Target | Uncharacterized | Uncharacterized | Uncharacterized | Uncharacterized |\nτ | Permitted | Target | Uncharacterized | — | Uncharacterized |\nR | — | — | Target | — | Uncharacterized |\nS | — | — | — | Target | Permitted |\nSM | — | — | — | Permitted | Target |\nPA | — | — | — | — | Target |\n\nHere, \"—\" denotes a theoretically excluded pathway under the architecture's stated causal model. It is not empirical evidence of zero effect.\n\nAll remaining non-target cells are Uncharacterized until measured.\n\nThe separation matrix is therefore more informative for causal claims than a modular architecture diagram alone:\n\n*module separability ⇏ causal separability*\n\nAn intervention port is not itself evidence that the intended target has been manipulated.\n\n*port manipulation ⇏ target engagement*\n\nIntervention validity and target engagement are distinct stages:\n\n• Intervention validity establishes that the manipulation is an empirically adequate implementation of the causal operator specified by the hypothesis.\n\n• Target engagement establishes that the hypothesized target was altered.\n\n**Nominal vs. real interventions: **Distinguish I_nominal (the intervention the experimenter intends to perform) from I_real (the transformation actually produced). Validity requires evidence that I_real ≈ I_causal within a prespecified set of invariants.\n\nHDD-ISA requires independent target-engagement evidence. A claim of target engagement is supported when the following four criteria are satisfied:\n\n**TE1 — Manipulation Fidelity: **The intervention produces the predicted change in the target variable or pathway.\n\n**TE2 — Target Specificity: **The change is distinguishable from non-specific perturbations.\n\n**TE3 — Invariant Preservation: **Relevant architectural invariants (capacity, task performance, information throughput) remain within prespecified bounds.\n\n**TE4 — Replication: **The engagement pattern is reproducible across independent trials.\n\nThe inferential rule is:\n\n*unconfirmed target engagement ⇒ no construct inference*\n\nA failed intervention should therefore be classified as UE, not automatically as negative evidence.\n\nHDD-ISA distinguishes two forms of identification that are often conflated.\n\n**Construct identification **establishes that a functional construct is supported by the intervention signature. For example, C-IV may be supported by evidence that behavior is causally sensitive to referent identity.\n\n**Mechanism identification **establishes which specific internal implementation produces that effect. Multiple mechanisms may be compatible with the same construct.\n\nThus:\n\n*construct identification ⇏ mechanism identification*\n\nand:\n\n*mechanism implementation ⇏ construct identification*\n\nThis yields a hierarchy of inferential specificity:\n\n*behavioral sensitivity < construct identification < mechanism identification*\n\nThis is not an ontological hierarchy. It is a hierarchy of evidential specificity.\n\n**Requirement: **persistent historical state M.\n\n**Intervention: **do(M = ∅) or a placebo-history manipulation do(M = M_placebo) where the placebo preserves relevant dimensionality and generic capacity while disrupting temporal information.\n\nEstimand:\n\n*Δ_I = L_no-memory − L_memory*\n\n**Evidence: **Δ_I > 0 under prespecified capacity and task controls.\n\n**Interpretation: **A positive result establishes predictive utility of historical information. It does not establish trajectory causality, recurrence, self-reference, or self-modeling. do(M = ∅) is not interpretable as causal value of memory without validating that the observed effect is not simply generic capacity loss.\n\n**Requirement: **trajectory–state decoupling.\n\nConstruct trajectories T^A ≠ T^B such that d_Z(Z_n^A, Z_n^B) ≤ ε.\n\nDefine M = { Z : d_Z(Z_n^A, Z_n^B) ≤ ε }.\n\nThe conditional trajectory effect is:\n\n*Δ_II(h) = E[Q(O_{n+h}) | do_gen(T=τᴬ), Z_n∈M] − E[Q(O_{n+h}) | do_gen(T=τᴮ), Z_n∈M]*\n\n**Operator clarification: **do_gen(T = τ) denotes intervention on the causal components that generate the trajectory. do_replay(τ) (replaying an observed sequence) should not be treated as equivalent unless the relevant generative states and transition dynamics are shown to be preserved.\n\n**Interpretation: **A positive result establishes that differentiated trajectories retain causal influence beyond the specified matched present-state variables. It does not establish that a particular memory variable mediates the effect:\n\n*trajectory causality ⇏ memory causality*\n\n**Requirement: **independently bypassable feedback pathway R and a capacity-matched comparison condition.\n\nLet R ∈ {0,1} indicate feedback absence/presence. Let K ∈ {0,1} indicate a paired control intervention condition:\n\n• K=1: computational capacity is preserved (operationalized by a set of invariants: performance, latency, activation statistics, throughput)\n\n• K=0: controlled degradation condition\n\nK is not a primitive causal variable; it is an experimental condition that must be empirically validated.\n\nThe interaction estimand is:\n\n*Δ_III = (Y_{R=1,K=1} − Y_{R=1,K=0}) − (Y_{R=0,K=1} − Y_{R=0,K=0})*\n\nThe relevant hypothesis is:\n\n*H0: Δ_III ≤ 0 versus H1: Δ_III > 0*\n\n**Structural vs. functional recurrence: **R_structural (presence of a feedback pathway) does not establish R_functional (causal role in history-dependent effect). C-III establishes the latter.\n\n**Interpretation: **The tested historical effect is specifically modulated by the candidate feedback pathway. It should not be inflated into the broader claim that \"the architecture is recurrent.\"\n\n**Control requirement: **The capacity-matched condition must be empirically validated against the prespecified invariants. Pathway ablation does not establish pathway-specific causal evidence without such validation.\n\n**Requirement: **manipulable agent-referent variable/interface. The term \"agent-referent\" is used rather than \"self-referent\" because the functional status of the representation is precisely what is being tested.\n\nThe referent should be switchable between S_self and S_ext while preserving task structure and relevant instrumental properties.\n\nA basic 2×2 design manipulates referent and truth.\n\nThe interaction is:\n\n*Δ_IV = [P(A|S_T^self) − P(A|S_F^self)] − [P(A|S_T^ext) − P(A|S_F^ext)]*\n\n**Evidence: **Δ_IV ≠ 0 together with referent sensitivity, invariance to surface-form changes, instrumental equivalence, and replication across task types.\n\nThe inference remains strictly functional:\n\n*self-reference ⇏ selfhood ⇏ consciousness*\n\n**Requirement: **S, SM, SM → PA where S is system state, SM is the self-model, and PA is the policy-access pathway.\n\nC-V consists of three dimensions, each with an explicit operational criterion:\n\n**V1 — Self-prediction: **The system predicts relevant aspects of its own future state with out-of-sample accuracy exceeding a prespecified baseline. Verbal description or self-report alone is insufficient; prediction must be evaluated on held-out trajectories.\n\n**V2 — Counterfactual self-prediction: **The system predicts how its own future state would differ under counterfactual conditions. This requires sensitivity to interventions on the system's own state, evaluated against independently generated counterfactual rollouts.\n\n**V3 — Causal model use: **Intervention on self-model content changes policy selection through the SM → PA pathway. The effect must be present when PA is open and attenuated when PA is blocked.\n\n**Epistemic vs. instrumental distinction: **V1 and V2 establish properties of model adequacy (self-prediction and counterfactual competence). V3 establishes instrumental causal use. Self-modeling requires both:\n\n*self-modeling = self-prediction + counterfactual competence + causal policy use*\n\n**Critical requirement: **V1 ∧ V2 ∧ V3 must establish that the representation identified as SM is both predictive of the system's future states and causally used in policy selection. Correlation among V1, V2, and V3 is insufficient.\n\nTherefore:\n\n*V = V1 ∧ V2 ∧ V3*\n\nV1 alone is insufficient.\n\n**Self-Model Causal Intervention (SMCI): **SMCI directly manipulates self-model content: do(Ŝ_self = ŝ). The manipulation is evaluated under two conditions: PA = open and PA = blocked.\n\nThe estimand is:\n\n*Δ_V = [E(Y|do(Ŝ=ŝ_true)) − E(Y|do(Ŝ=ŝ_false))]_{PA=open} − [E(Y|do(Ŝ=ŝ_true)) − E(Y|do(Ŝ=ŝ_false))]_{PA=blocked}*\n\n**Control requirement: **PA = blocked must be validated with capacity-preservation controls to ensure that the attenuation is not simply generic degradation.\n\nThe critical signature is an intervention-sensitive effect that exists when the policy pathway is available and is attenuated when that pathway is blocked.\n\nThus:\n\n*self-description ⇏ self-model use*\n\nand:\n\n*self-prediction ⇏ self-modeling*\n\nunless causal model use is demonstrated.\n\nAn implementation should proceed construct by construct.\n\nFor each construct, first identify whether the required architectural interface exists. If it does not, the construct should be explicitly reported as architecturally untestable under the current interface:\n\n*T interface_C = 0*\n\nIt should not silently disappear from the analysis.\n\nFor a transformer-based agent, for example:\n\n|\n|\n|\nC-I | Context/KV state | do(M = ∅) |\nC-II | Observation/action trajectory | do_gen(T=τᴬ) vs. do_gen(T=τᴮ) |\nC-III | Explicit recurrent pathway, if present | do(R = 0) |\nC-IV | Agent-referent variable/interface | do(S_self) vs. do(S_ext) |\nC-V | Dedicated self-model representation | SMCI |\n\nA standard feedforward transformer does not automatically provide a C-III interface. A recurrence claim therefore requires an explicit feedback pathway whose manipulation can be distinguished from general computational degradation.\n\nFor an RNN/LSTM:\n\n|\n|\n|\nC-I | Hidden/history state | do(M = ∅) |\nC-II | Hidden-state trajectory | do_gen(T=τᴬ) vs. do_gen(T=τᴮ) |\nC-III | Recurrent transition | do(R = 0) |\nC-IV | Agent-referent representation | Self/external substitution |\nC-V | Self-model readout | SMCI with PA-open/blocked |\n\nFor black-box LLM deployments, prompt manipulations can serve as behavioral probes but should not automatically be described as architectural interventions.\n\nThus:\n\n*prompt probe ⇏ architectural intervention*\n\nunless the prompt manipulation has been independently linked to the internal variable specified by the hypothesis.\n\n|\n|\n|\nC-I | Capacity rather than history | Capacity-matched comparison and temporal placebo |\nC-II | Latent present-state differences | Prespecified trajectory–state matching |\nC-III | General computational damage | Capacity-matched feedforward baseline with validated invariants |\nC-IV | Label association or pragmatic inference | Referent swapping and relevance controls |\nC-V | Generic degradation | Capacity-matched manipulation and PA-blocked control |\n\nWhere feasible, controls should preserve task difficulty, computational capacity, information quantity, surface complexity, reward structure, temporal position, and generic uncertainty.\n\nHDD-ISA recommends four principal inferential states.\n\n|\n|\nSupported | Target engagement confirmed, controls pass, and the observed signature discriminates the target hypothesis |\nNegative Evidence | Target engagement and intervention validity confirmed, but the predicted signature is absent |\nUE — Uninterpretable | Intervention validity or target engagement was not adequately established |\nNI — Non-Identifiable | The experiment was successfully executed, but available interventions and outcomes leave competing hypotheses equivalent |\n\nThis distinction prevents two important inferential errors:\n\n*UE ⇏ evidence against*\n\nand:\n\n*NI ⇏ evidence of absence*\n\nIn practice, an intervention may be theoretically sufficient but experimentally infeasible due to cost, system destruction, inaccessibility, or excessive degradation.\n\nDistinguish:\n\n• Non-identifiable: no available intervention set can separate the hypotheses\n\n• Not practically identifiable under budget B: the intervention set exists but is infeasible under practical constraints\n\nThus:\n\n*NI ⇏ practically infeasible*\n\nand:\n\n*practically infeasible ⇏ in principle impossible*\n\nThis limitation should be considered when interpreting results.\n\nThe HDD Gap is the inferential distance between implementing a candidate mechanism and establishing its functional role. It consists of two distinct identification problems.\n\n**Forward Gap (G_forward):**\n\n*mechanism implemented ⇏ functional hypothesis established*\n\nThis is the problem of functional interpretation: given that a mechanism is present, what functional construct, if any, does it implement?\n\n**Inverse Gap (G_inverse):**\n\n*functional effect observed ⇏ unique mechanism identified*\n\nThis is the problem of mechanistic identification: given an observed functional effect, which internal mechanism produced it?\n\nThese two gaps correspond to two distinct fallacies:\n\n**Implementation fallacy: **\"A mechanism resembling X exists, therefore the system implements X.\"\n\n**Reverse-mechanism fallacy: **\"Behavior compatible with X is observed, therefore we know which internal mechanism produced it.\"\n\nThese gaps are logically independent. An architecture may contain a genuine mechanism whose causal role cannot be identified with the available interventions. Conversely, an observed functional effect may be genuine while several distinct mechanisms remain compatible with the data.\n\nThe HDD Gap is therefore not a single problem but two:\n\n*mechanism —G_forward⇏— construct*\n\n*effect —G_inverse⇏— mechanism*\n\nHDD-ISA narrows both gaps by expanding the experimental interface:\n\n*(I, Y) → (I', Y')*\n\nwith the explicit objective of separating competing explanations.\n\nHDD-ISA is explicitly interface-relative.\n\n10. It discriminates only among hypotheses included in H. It cannot rule out mechanisms that were never considered.\n\n11. Interventions may perturb multiple mechanisms. The separation matrix characterizes this problem but cannot guarantee perfect isolation.\n\n12. Representations are generally non-unique. The same input-output behavior may be implemented through multiple internal mechanisms.\n\n13. Demonstrations may be distribution-dependent. Claims involving self-reference and self-modeling require replication across tasks.\n\n14. C-II relies on approximate present-state matching. The resulting estimand is conditional on the matching procedure.\n\n15. Black-box prompt experiments remain behavioral probes unless their relationship to the hypothesized internal mechanism is independently established.\n\n16. A hypothesis pair may be non-identifiable under one interface and identifiable under another:\n\n*NI(A,H,I1,Y1) ⇏ NI(A,H,I2,Y2)*\n\n17. Identifiability may be infeasible under practical intervention budgets even when theoretically possible.\n\nIf a construct remains systematically non-identifiable across an architecture class despite reasonable intervention and measurement designs, it should be classified as NI for that architecture class rather than repeatedly tested with inadequate interfaces.\n\nIf target engagement cannot be independently established for a class of architectures, claims about that class should be restricted to behavioral-probe status.\n\nIf a design rule repeatedly proves impossible to satisfy without unacceptable degradation of general architectural function, the rule should be revised or scoped.\n\nHDD-ISA also adopts an anti-rescue principle. A new intervention or estimand should not be introduced merely because a previous experiment failed to produce the desired result.\n\nFinally, when intervention leakage remains sufficiently large to prevent identification of the relevant causal contribution, the appropriate result is NI rather than an interpretive rescue.\n\nHDD-ISA translates the functional hypotheses of HDD into architectural requirements, intervention interfaces, target-engagement procedures, controls, and reporting standards. Its contribution is methodological rather than ontological.\n\nThe central distinctions are:\n\n*prediction ⇏ causation*\n\n*causation ⇏ recurrence*\n\n*recurrence ⇏ self-reference*\n\n*self-reference ⇏ self-modeling*\n\nand:\n\n*architectural access ⇏ target engagement ⇏ causal identification*\n\nLikewise:\n\n*module separability ⇏ causal separability*\n\nAn architecture may expose a mechanism without allowing its functional role to be identified. A behavioral effect may be real without uniquely identifying the mechanism that produced it. And two modules may be structurally distinct while remaining causally inseparable under the available interventions.\n\nThe relevant architectural virtue is therefore not the number of components labeled \"memory,\" \"self,\" or \"self-model.\" It is whether competing explanations can be independently manipulated, their targets verified, their cross-effects measured, and their predicted intervention signatures compared.\n\nThe HDD Gap is consequently not a gap between architecture and intelligence. It is the gap between:\n\n*having a mechanism*\n\nand:\n\n*knowing, with causal evidence, what functional role that mechanism plays.*\n\nHDD-ISA specifies the architectural interfaces required to make that gap experimentally tractable.\n\nHernán, M. A., & Robins, J. M. (2020). Causal Inference: What If. Chapman & Hall/CRC.\n\nLittman, M. L., Sutton, R. S., & Singh, S. (2001). Predictive representations of state. Advances in Neural Information Processing Systems 14.\n\nPearl, J. (2009). Causality (2nd ed.). Cambridge University Press.\n\nRoy, N. A., Kim, J., & Rabinowitz, N. C. (2022). Explainability via causal self-talk. Advances in Neural Information Processing Systems 35.\n\nTaotuner. (2026). History-Dependent Dynamics (HDD): A Methodological Framework for Disentangling History Dependence, Recurrence, Self-Reference, and Self-Modeling in Dynamical Systems. Zenodo.\n\nAng, C. K. (2026). The AI Ego. PhilPapers.\n\nAryan, A., & Liu, Z. Y.-C. (2025). Causal Reflection with Language Models. NeurIPS 2025 Workshop.\n\nFox, K. L. (2026). The You/I Paradigm. Zenodo.\n\nMazzocchetti, A. (2025). Civitas. Zenodo.\n\nScottonanski. (2025). Persistent Mind Model (PMM) v1.2. Zenodo.\n\nYang, C. (2026). Self-Aware Recursively Self-Improving Agents. arXiv:2607.12254v2.\n\n(2025). Counterfactual VLA. arXiv:2512.24426.\n\n(2026). Functional Self-Modeling Probes. GitHub — dp-web4/SAGE.", "url": "https://wpnews.pro/news/hdd-isa-ai-architectures-for-causal-discriminations", "canonical_source": "https://informationalprocessualmonism.blogspot.com/2026/08/hdd-isa-ai-architectures-for-causal.html", "published_at": "2026-08-22 20:58:02+00:00", "updated_at": "2026-08-22 21:13:03.207717+00:00", "lang": "en", "topics": ["ai-research", "ai-safety", "ai-ethics"], "entities": ["Taotuner", "HDD-ISA", "History-Dependent Dynamics"], "alternates": {"html": "https://wpnews.pro/news/hdd-isa-ai-architectures-for-causal-discriminations", "markdown": "https://wpnews.pro/news/hdd-isa-ai-architectures-for-causal-discriminations.md", "text": "https://wpnews.pro/news/hdd-isa-ai-architectures-for-causal-discriminations.txt", "jsonld": "https://wpnews.pro/news/hdd-isa-ai-architectures-for-causal-discriminations.jsonld"}}