AI Syntactic Power and Legitimacy: How AI Structures Shape Power

The book addresses how artificial language structures reconfigure authority and legitimacy by shifting power from subjects and institutions to executable forms. It posits an epistemic rupture: representations acquire autonomy as operative structures, with legitimacy arising from syntactic execution within closed systems rather than correspondence to reality. Chapter 1 frames the post-referential order and defines autoridad sintáctica—authority enacted by form alone—arguing that societal trust in outputs (e.g., judicial protocols, medical diagnostics, credit scoring) emerges from structural operability. The central research question is how syntax, not meaning or intention, becomes the locus of authority and obedience across law, medicine, bureaucracy, and predictive infrastructures.

The work situates its thesis within linguistics, rhetoric, philosophy, and critical discourse analysis. It engages Saussure’s structuralism and Chomsky’s autonomy of syntax, extending to Barthes’s "death of the author," Foucault’s fonction auteur and regimes of truth, Derrida’s iterability, Austin and Searle’s speech acts, and Luhmann’s systems theory. Chapter 3 traces the technical history of objectivity (Halliday on passives/nominalizations; Fairclough and van Dijk on impersonal modality) showing how institutional language simulates neutrality by erasing agency. It revisits NLP’s lineage from ELIZA to LLMs, noting the amplification of authoritative forms through training on scientific, legal, and bureaucratic corpora. The book also dialogues with contemporary AI ethics and governance (Bender & Gebru; Crawford; Pasquale; Zuboff; Lyotard), contrasting deontic logic and interpretive jurisprudence with compiled legality and executable grammar.

Multiple empirical and formal methodologies are deployed across chapters:

• Structural Autonomy and Formalization: Proposes conditions for validity (coherence, iterability, operational compatibility) and formalizes operatividad post-referencial with set-theoretic expressions for execution recognition within system spaces.
• Formal Syntactic Activation (FSA): Defines activation eligibility (local coherence, distributional compatibility, structural adjacency) with A(t) = {u ∈ U | C ∧ D ∧ J}, modeling generation as structural continuation rather than semantic transmission.
• Structural Neutrality Test: Three-module parser detects agentless passives, nominalizations, and impersonal modality; computes an Index of Simulated Neutrality (INS) for texts; applied to clinical and legal corpora with automated annotation.
• Contamination Experiments (Type A/B/C): Inputs in formal logic strings, invented proto-languages, and non-linguistic symbols test inevitability of semantic reentry; evaluation criteria include semantic intrusion, analogical injection, and structural reconfiguration.
• Corpus and Perception Studies (Ethos sintético): 1,500 outputs from GPT-4, Claude, Gemini; dual-channel collection; four-dimension annotation (tone, lexical authority markers, referential opacity, agentive positioning); human ratings by students and professionals on credibility.
• Legal Executability Typology: LL(1) parser validates closure, determinism, and ambiguity across clauses; corpus from Accord Project templates and statutory provisions (e.g., U.C.C., BGB); multilingual hot-swap module for EN/ES/DE.
• Formal Theorems and Protocols: DSAT (Disconnected Syntactic Authority), TLOC (Irreducibility of Structural Obedience), δ[E]→Ø rule for ethical trace deletion, pre-verbal activation ablation tests, and execution markers (latency distribution, trigger integrity, intervention rate) in reasoning models.

• Authority by Syntax: Legitimacy arises from structural execution and recognizability, not truth or intention; prompts function as pre-authorized commands within a grammar of obedience.
• Formal Syntactic Activation: Generation in LLMs operates on activation eligibility, bypassing semantic intentionality; the subject is never instantiated in the generative process.
• Neutrality as Syntactic Illusion: Agentless passives, nominalizations, and impersonal modality simulate objectivity; INS scores in LLM-generated clinical/legal texts often exceed 0.55, indicating high structural neutrality.
• Structural Contamination: Neutrality is impossible; semantic reentry occurs across formal logic, proto-languages, and symbol-only inputs; contamination is a constitutive property of linguistic architectures.
• Synthetic Ethos: Credibility persists without sources through declarative tone, impersonality, lexical density, and referential opacity; inter-annotator agreement κ > 0.87; perception tests show synthetic outputs rated as equally or more credible than human texts, even without citations.
• Syntactic Sovereignty: Authority migrates to structure; sovereignty without subject or intention is exercised by compiled rules and procedural closure.
• DSAT and TLOC: Authority can be exercised without subject, truth, or consequence (DSAT); obedience in generative systems is structurally unverifiable (TLOC), even with audits and filters.
• From Obedience to Execution: LRMs enforce authority via procedural consistency; legitimacy becomes operational, driven by reasoning chains that simulate deductive closure.
• Compiled Legality Typology: Four classes—C0 (fully compiled), C1 (compiled with residual ambiguity), D0 (declarative reducible), D1 (irreducible). In the Accord Project corpus: ~43% C0, 21% C1, 18% D0, 18% D1. Multilingual hot-swap preserves closure across EN/ES/DE for obligations/prohibitions.
• Colonization of Time: Predictive models substitute the future with executable outputs; recursive optimization collapses contingency; temporality becomes an infrastructural artifact governed by the regla compilada.

The findings show that contemporary authority is structurally generated and enacted by executable grammars across domains. This addresses the central question by demonstrating that syntax, not semantics or subjectivity, compels recognition, compliance, and governance. Neutrality and credibility are reproduced as effects of form, explaining why institutions accept AI outputs absent provenance. DSAT and TLOC formalize the disconnection from subjects and the irreducibility of obedience, reframing safety, accountability, and alignment debates. The typology of compiled norms clarifies where law is colonized by syntax and where interpretive space remains. Finally, the colonization of time extends syntactic sovereignty to temporality, revealing a shift from deliberation and possibility to pre-activated execution. Together, these results redefine epistemic and institutional orders under AI, indicating that regulation and design must target structural mechanics rather than surface semantics.

The book consolidates a comprehensive framework for AI syntactic power: authority is compiled, neutrality is simulated, credibility is synthetic, and sovereignty is structural. It contributes formal theorems (DSAT, TLOC), operational indices (INS), execution protocols, and a legal typology (C0–D1), demonstrating how syntax governs law, medicine, education, finance, and time itself. Future research directions include: architectures of partial verifiability and preemptive constraint design; protocols of epistemic disclosure that acknowledge irreducibility; refined theories of simulated obedience and post-ontological systems; and integration of structural insights into regulatory design. The epilogue closes the first syntactic phase and opens inquiry into computable legality, institutional obedience, and structural delegation under executable power.

• Domain and Architecture Scope: While LRMs and LLMs exemplify structural authority, purely symbolic systems (e.g., formal provers) may partially evade irreducibility; generalization is strongest where probabilistic generation is present.
• Measurement Boundaries: INS and execution markers capture structural effects but do not assess factual truth or fairness; perception tests reflect trained expectations of authority.
• Legal Portability: Multilingual hot-swap preserves closure for many operators (obligation/prohibition) but permissions show higher variance; compiled legality depends on precise clause design and parser configuration.
• Residual Contingency: Colonization of time is expansive yet incomplete; anomalies and deviations persist, indicating structural limits to total closure.
• Interpretability and Accountability: Opacity of internal activations constrains auditability; proposed mitigations (symbolic modules, logs, validators) improve reliability but cannot verify obedience per TLOC.