Under AI Conditions 1.2.6-5

Why AI Intensifies the Exit Problem

Bounded interdependence addresses a structural problem that existed before AI: cooperative frameworks that convert voluntary participation into structural dependency through accumulated switching costs, chokepoint concentration, and the erosion of viable alternatives. AI does not create the problem. It accelerates every mechanism that produces it.

The acceleration operates through three structural channels, each mapping to a component of the innovation.

Training Data as Lock-In

AI systems learn from the data their users provide — queries, documents, workflows, preferences, corrections. Over time, the system becomes calibrated to the user's patterns. This calibration creates value: the system becomes more useful, more efficient, more aligned with the user's needs. It also creates dependency. The calibration is embedded in the model's weights or in the platform's fine-tuning layer, and it cannot be extracted in a form that another system can use.

The structural result is a new form of switching cost that the article's framework identifies precisely. The user has invested effort, data, and institutional knowledge into a system that has absorbed those inputs and converted them into a proprietary asset. The exit question is not whether the user can stop using the system — formally, they always can. The question is whether they can leave without losing the cumulative value of their investment. If the training data, fine-tuning, and calibration cannot be exported, the user's departure cost grows with every interaction. This is the switching cost escalation pattern from the Lock-In Risk Matrix (1.2.6-3), operating at machine speed.

Data portability and credential independence are the structural responses. The design recommendation: AI systems that learn from user data should support the export of that data — and ideally the fine-tuning or calibration results — in formats that other systems can use. A system that absorbs user data and returns only inferences has converted a user's own investment into a competitive liability for itself: the harder it becomes to leave, the more clearly the difficulty announces that the system is winning on switching cost rather than on merit.

Model-Mediated Credentialing

As AI systems are integrated into professional, educational, and civic contexts, they increasingly mediate access — who is qualified, who is credentialed, who is visible, who is recommended. An AI model that controls hiring recommendations, professional visibility, or educational advancement becomes a chokepoint in the sense the article defines: a node through which a disproportionate share of professional traffic must pass.

The structural problem is that the credentialing criteria may be opaque. A human credentialing body publishes standards, administers assessments, and issues credentials that can be independently verified. An AI system that mediates credentialing through a recommendation algorithm, a relevance score, or a visibility ranking may not publish criteria, may not allow independent verification, and may change its criteria without notice. The departing participant cannot carry what they earned because what they earned was a position in an algorithmic ranking — and the ranking does not travel.

Credential independence recommends that earned standing be verifiable without the endorsing system's ongoing participation. Under AI conditions, that recommendation becomes more demanding to satisfy: the credential should be verifiable not merely without the issuing institution but without the mediating model. Portable evidence of competence — demonstrated work products, peer attestations, independently verifiable records — is the structural alternative to model-dependent standing.

Exit from AI-Dependent Workflows

Organizations that integrate AI deeply into their workflows — decision support, document generation, process automation, knowledge management — create a structural dependency that makes institutional exit costly in ways the participants may not recognize until they attempt to leave.

The mechanism is the same as the article describes for any cooperative framework, but the accumulation is faster. An organization that runs its knowledge base through an AI platform, generates its documents through an AI service, and routes its decisions through an AI recommendation engine has embedded the platform into its operational identity. Departing from the platform means rebuilding the knowledge base, retraining the document pipeline, and reconstructing the decision infrastructure — switching costs that grow with every month of deeper integration.

Dependency power measurement should account for AI-specific accumulation patterns. The three diagnostic dimensions from the Lock-In Risk Matrix apply: switching cost accumulation (how fast does AI integration increase departure cost?), chokepoint concentration (how many critical functions are routed through a single AI provider?), and alternative viability (are there comparable AI services that can accept the organization's data and workflows without prohibitive migration cost?). The design recommendation: organizations integrating AI measure lock-in risk periodically and maintain structural provisions for portability — open data formats, documented workflows, provider-independent process descriptions — as a condition of deeper integration.