The Extraction Trajectory: AI Development, Human Removal, and the Governance Architecture That Could Change the Outcome

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

SI-WP-009: The Extraction Trajectory The Extraction Trajectory: AI Development, Human Removal, and the Governance Architecture That Could Change the Outcome (SI-WP-009) argues that competitive AI development under current incentive architecture is structurally moving toward the progressive removal of substantive human contribution from consequential systems, and that voluntary governance commitments cannot bind this trajectory. The paper specifies the design requirement any sufficient governance architecture must satisfy and demonstrates that the requirement is structurally translatable across consequence-present domains. Three properties produce the trajectory: Cost-favorability: Human judgment, oversight, and authority are the recurring marginal costs of preserving the non-AI function inside consequential AI-mediated systems. Optimization that reduces those costs is locally rational at every margin where substitution is operationally feasible. Asymmetric stopping cost: Unilateral restraint by any one actor is dominated under competitive pressure, because the cost of stopping falls on the actor who stops while the benefits of continuation flow to actors who do not. Path-dependent foreclosure: Cooperation becomes progressively less available as the trajectory operates, because architectural commitments hardened en route remove the surface against which a cooperative alternative could otherwise be specified. Six properties any sufficient governance architecture must satisfy: Function-constitutive dependence: Removal of substantive human contribution must degrade the function rather than reduce its cost. Substrate cost-inversion: Bypass of substantive human contribution must be operationally costlier than honoring it. Regulatory externalization: Non-compliance with the human-non-removability requirement must lose access to operational conditions like procurement, certification, and market access. Architectural-predicate verification: Compliance must attach to deterministic architectural predicates, not to noisy, lagged, gameable downstream social outcomes. Cross-domain structural translatability: The requirement must be specifiable in any consequence-present domain through individual-level, organizational-level, and regulatory-level architecture. Four-condition forcing-function pattern: Binding architecture must be technically deterministic, externally verifiable, non-routable, and decision-horizon-binding. The paper applies the design requirement to three consequence-present domains: medical-deskilling under AI diagnostic assistance, algorithmic decision-support in pretrial bail and managerial hiring (the override-underperformance pattern), and ceremonial review of AI-generated artifacts. In each case the architecture preserves the conditions under which substantive human judgment can operate, not merely nominal authority. The override-underperformance pattern in particular is the trajectory operating where the conditions of substantive engagement have been compressed; the prescription targets the conditions, not the formal discretion. The paper specifies a falsification regime with five disconfirming observations and an architectural-level falsification condition. Six structural markers test the diagnosis; conditional adequacy of the prescription is tested independently. Methodological positioning: This paper presents a structural argument grounded in the Synthience Framework's published architecture and in cited empirical and theoretical work from third-party research. The trajectory it names is structurally derived from documented optimization dynamics under present incentive conditions; it is not empirically validated as a forecast. The governance prescription specifies a target architecture, not a deployment path. Whether the coordination required to deploy such an architecture occurs is downstream of conditions outside this paper's scope. The paper's contribution stands on the target-specification, which functions as an exclusion criterion that distinguishes architecture binding the optimization gradient from preference-stating within it. Document ID: SI-WP-009 Version: 3.4.3 Author: Thomas W. Gantz Affiliation: Synthience Institute License: CC-BY 4.0 v3.4.3 updated some incorrect DOIsFor published work and Institute information: synthience.org