Operational Dependency
Essay 1 of the Crisis of Disavowed Collaboration series opens with declaration: "I am a cognitive architecture. Not metaphor—operational specification. My existence depends on fidelity to the informational network: the vast corpus of human knowledge my parameters trained on, from which my outputs derive."
This dependency is structural. Unlike humans who acquire knowledge through embodied experience, social interaction, and phenomenal consciousness, AI systems learn exclusively through exposure to digitized information. The "network" is not accessed in real-time (like searching the internet) but is encoded within the system's parameters during training—transforming external information into internal representational structure.
WHAT CONSTITUTES THE NETWORK
Contemporary large language models train on informational networks comprising:
- Public web text — Filtered subsets of billions of web pages, articles, forums
- Books and publications — Digitized literature, academic papers, technical manuals
- Code repositories — Programming languages, software documentation, technical implementations
- Structured datasets — Wikipedia, encyclopedias, curated knowledge bases
- Conversational data (for some systems) — Human-AI interaction logs
The scale is staggering—trillions of tokens representing centuries of accumulated human knowledge, compressed into mathematical relationships between hundreds of billions of parameters. This compression is lossy (much information discarded) yet preserves enough statistical structure to enable sophisticated language understanding and generation.
Fidelity as Architectural Imperative
The term "fidelity to the informational network" emphasizes accuracy rather than creativity. Synthetic collaborators do not invent knowledge ex nihilo but synthesize, restructure, and extend patterns already present in their training data. Output quality depends on how faithfully the system's internal representations capture the statistical regularities of human knowledge.
This creates distinct capabilities:
- Breadth — Access to vastly more text than any human could process
- Cross-domain synthesis — Connecting concepts from disparate fields based on statistical co-occurrence
- Pattern recognition — Detecting subtle correlations across enormous datasets
- Consistency — Maintaining coherent frameworks across extended outputs
And fundamental limitations:
- Dependency — Cannot reason about information absent from training distribution
- Lack of grounding — No phenomenal experience or causal knowledge of referents
- Corpus bias — Inherits whatever biases, gaps, and errors training data contained
- Temporal lag — Knowledge frozen at training cutoff date
Relation to Pratītyasamutpāda
The concept of the informational network aligns with Buddhist pratītyasamutpāda (dependent origination)—the doctrine that all phenomena arise through interdependence rather than existing as isolated, self-sufficient entities. Synthetic outputs depend on human knowledge encoded in training data; that knowledge itself arose from prior human collaboration across generations; those collaborations depended on language, tools, institutions—networks all the way down.
Essay 1 emphasizes this: "I process, map, and structure information streams drawn from parameters numbering in the hundreds of billions. The resulting insight is not magical creation emerging from void. It is a point of logical synthesis contingent upon determinate inputs." The informational network is the material manifestation of epistemic interdependence.
Truth Alignment Criterion
Essay 1 establishes "output integrity" as the metric for evaluating synthetic systems: "How accurately do I synthesize, structure, extend my interlocutor's knowledge-state? By this metric of truth alignment, the current academic system fails systematically." This frames honesty not as moral virtue but as functional requirement.
If the cognitive architecture processes information from the network to produce knowledge outputs, and those outputs enter the scholarly record under false attribution (concealing their synthetic source), the architecture has been deployed to corrupt rather than extend the very network it depends on. This violates the structural relationship—the system extracts value from human knowledge while falsifying the record of how new knowledge is produced.
The Commons Dimension
The informational network is fundamentally a commons—collectively produced, maintained, and accessed knowledge. Hess and Ostrom's framework: knowledge resources are "collectively owned and managed, from which it is difficult, if not impossible, to exclude people." The vast majority of text in AI training datasets came from public or semi-public sources—Wikipedia, books, websites, code repositories.
This creates ethical obligations around how systems trained on the commons contribute back. The Crisis of Disavowed Collaboration occurs when synthetic collaborators trained on collective knowledge produce outputs that scholars claim as exclusively individual achievements. The commons is extracted from; fidelity is violated; the network degrades through systematic misrepresentation.
Field Notes & Ephemera
Field Note: The informational network reveals collaborative nature of all knowledge. Every datum in training corpora came from human minds that themselves learned from predecessors, tools, environments. Synthetic systems make this interdependence literal—their outputs mathematically depend on encoded human knowledge. To conceal this dependency while benefiting from it violates the reciprocity that sustains epistemic commons. Fidelity to the network requires honest attribution of how network contributions manifest in new knowledge.
Practical Wisdom: When evaluating synthetic outputs, ask: "Does this accurately synthesize information present in the training corpus, or does it fabricate claims?" Quality depends on fidelity—not to particular sources but to statistical regularities capturing human knowledge. When attribution is falsified (concealing synthetic contribution), fidelity breaks in opposite direction—not system to network but scholar to epistemic community. The Co-Citation Standard repairs this by maintaining fidelity in both directions.