Tallennettuna:
| Päätekijät: | , |
|---|---|
| Aineistotyyppi: | Recurso digital |
| Kieli: | englanti |
| Julkaistu: |
Zenodo
2026
|
| Aiheet: | |
| Linkit: | https://doi.org/10.5281/zenodo.18308554 |
| Tagit: |
Lisää tagi
Ei tageja, Lisää ensimmäinen tagi!
|
Sisällysluettelo:
- <p>Abstract<br>This paper examines semantic degradation as a systemic failure mode in knowledge<br>systems, with particular attention to recursive contamination in artificial intelligence.<br>The paper introduces the concept of knowledge prion risk to describe how meaning<br>can degrade even when syntax remains intact. Drawing carefully scoped analogies<br>from biological prion diseases, it describes how systems trained on their own<br>outputs—or on outputs from similar systems—propagate structural corruption over<br>time. The paper argues that downstream interventions such as filtering, alignment<br>training, or improved prompting cannot cure substrate contamination. The only<br>effective response is strict control of what enters the epistemic substrate—what the<br>paper terms epistemic hygiene. This analysis extends beyond AI systems to any<br>recursive knowledge environment where outputs become inputs. The paper does not<br>name specific vendors or propose specific architectures; it offers a diagnostic frame<br>applicable across contexts.<br>Keywords: semantic degradation; prion risk; recursive training; AI safety; meaning collapse;<br>contamination; epistemic hygiene; model collapse; recursive contamination; substrate<br>corruption</p>