Differences
This shows you the differences between two versions of the page.
| dido:01-fdis-ra:02-background-and-motivation:02-5-why-a-reference-architecture-is-needed:02-5-3-ai-opacity:start [2026/07/12 13:01] – created nick_dido | dido:01-fdis-ra:02-background-and-motivation:02-5-why-a-reference-architecture-is-needed:02-5-3-ai-opacity:start [2026/07/18 12:33] (current) – external edit 127.0.0.1 | ||
|---|---|---|---|
| Line 3: | Line 3: | ||
| [[dido: | [[dido: | ||
| - | Artificial intelligence-assisted techniques are increasingly used in reporting systems to extract, classify, enrich, and analyse | + | Artificial intelligence-assisted techniques are increasingly used in reporting systems to extract, classify, enrich, and analyze |
| Artificial intelligence-assisted processing also introduces forms of opacity, nondeterminism, | Artificial intelligence-assisted processing also introduces forms of opacity, nondeterminism, | ||
| Line 15: | Line 15: | ||
| * What a validation or conformance process assessed | * What a validation or conformance process assessed | ||
| - | Model behaviour, confidence thresholds, training data, model versions, prompt logic, retrieval context, and configuration may remain implicit or tool-specific. As a result, | + | Model behavior, confidence thresholds, training data, model versions, prompt logic, retrieval context, and configuration may remain implicit or tool-specific. As a result, |
| When systems introduce artificial intelligence-assisted processing without architectural constraints, | When systems introduce artificial intelligence-assisted processing without architectural constraints, | ||
| Line 22: | Line 22: | ||
| * Probabilistic outputs being treated as asserted facts | * Probabilistic outputs being treated as asserted facts | ||
| * Confidence measures being omitted or interpreted inconsistently | * Confidence measures being omitted or interpreted inconsistently | ||
| - | * Model or configuration changes altering | + | * Model or configuration changes altering |
| * Inference and interpretation are being collapsed into a single opaque result | * Inference and interpretation are being collapsed into a single opaque result | ||
| * Historical outputs are becoming difficult to reproduce or reassess | * Historical outputs are becoming difficult to reproduce or reassess | ||
| - | * Tool-specific | + | * Tool-specific |
| - | These conditions undermine traceability, | + | These conditions undermine traceability, |
| AI opacity is therefore not an unavoidable property of artificial intelligence techniques. It is an architectural failure mode that arises when systems integrate artificial intelligence capabilities without explicit governance of their semantic roles, authority, inputs, outputs, versions, uncertainty, | AI opacity is therefore not an unavoidable property of artificial intelligence techniques. It is an architectural failure mode that arises when systems integrate artificial intelligence capabilities without explicit governance of their semantic roles, authority, inputs, outputs, versions, uncertainty, | ||