Semantic Fidelity Core Series: Meaning, Compression, Constraint, and Alignment
These papers examine what happens to meaning as information is compressed, retrieved, generated, summarized, and reused across human and artificial intelligence systems. They develop semantic fidelity as a way to evaluate whether the relationships that make information meaningful survive transformation, including context, hierarchy, causality, uncertainty, intent, and provenance.
The series distinguishes semantic fidelity from factual accuracy, source faithfulness, coherence, and textual similarity. It also examines how meaning can degrade through recursive compression, weakened corrective feedback, provenance loss, and repeated reuse of prior representations. Together, the papers provide a conceptual vocabulary for identifying representational loss before it becomes visible as obvious error or system failure.
What Is Semantic Fidelity? Preserving Meaning in the Age of Artificial Intelligence (PDF)
Introduces semantic fidelity as the degree to which meaning is preserved as information moves across systems, media, and representations. Distinguishes fidelity from accuracy, coherence, and semantic drift.
When Accuracy Is Not Enough: Evaluating Semantic Fidelity in AI Systems (PDF)
Explains how an AI output can remain factually correct while changing context, hierarchy, causality, uncertainty, or purpose. Provides a practical method for evaluating whether the organizing meaning of a source survives transformation.
The Compression Paradox in AI: Why Meaning Can Degrade Before Models Hallucinate (PDF)
Examines how compression can remove context, causal structure, uncertainty, and provenance while generation reconstructs a fluent surface. Introduces the Compression Paradox as a failure mode that can appear before recognizable hallucination.
Constraint Collapse and Fidelity Decay: When Feedback Stops Correcting Symbolic Systems (PDF)
Examines what happens when feedback remains present but loses the authority to revise representations or interrupt misaligned processes. Connects constraint collapse to cumulative fidelity decay and the broader problem of weakened corrective contact.
Language as Cognitive Exhaust: What Language Reveals About Thought Compression and Artificial Intelligence (PDF)
Develops cognitive exhaust as a model for understanding language as a compressed and incomplete trace of cognition. Explores how AI systems learn from linguistic representations rather than direct experience and how generated language can re-enter later human and machine cognition.
A Semantic Fidelity Lexicon: A Working Vocabulary for Meaning Preservation Across AI and Representational Systems (PDF)
Defines the core vocabulary of the Semantic Fidelity Project, including Semantic Fidelity, Semantic Drift, Recursive Compression, Constraint Collapse, Synthetic Coherence, Synthetic Realness, Provenance Failure, and related mechanisms of representational loss.