dido:01-fdis-ra:02-background-and-motivation:02-5-why-a-reference-architecture-is-needed:02-5-3-ai-opacity:start

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

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_didodido: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:01-fdis-ra:02-background-and-motivation:02-5-why-a-reference-architecture-is-needed:start|Return to 2.5 Why a Reference Architecture Is Needed]] [[dido:01-fdis-ra:02-background-and-motivation:02-5-why-a-reference-architecture-is-needed:start|Return to 2.5 Why a Reference Architecture Is Needed]]
  
-Artificial intelligence-assisted techniques are increasingly used in reporting systems to extract, classify, enrich, and analyse reported information. These techniques can improve scalability, processing speed, and operational efficiency, particularly when systems process large volumes of unstructured or semi-structured information.+Artificial intelligence-assisted techniques are increasingly used in reporting systems to extract, classify, enrich, and analyze reported information. These techniques can improve scalability, processing speed, and operational efficiency, particularly when systems process large volumes of unstructured or semi-structured information.
  
 Artificial intelligence-assisted processing also introduces forms of opacity, nondeterminism, variability, and uncertainty that differ from traditional rule-based processing. Artificial intelligence-assisted processing also introduces forms of opacity, nondeterminism, variability, and uncertainty that differ from traditional rule-based processing.
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, organisations may find it difficult to explain how a particular output was produced, determine which assumptions influenced it, or reassess the output under revised [[dido:99_annexes:annex-b-terms-and-definitions:s:semantic|semantic]] definitions or governance rules.+Model behavior, confidence thresholds, training data, model versions, prompt logic, retrieval context, and configuration may remain implicit or tool-specific. As a result, organizations may find it difficult to explain how a particular output was produced, determine which assumptions influenced it, or reassess the output under revised [[dido:99_annexes:annex-b-terms-and-definitions:s:semantic|semantic]] definitions or governance rules.
  
 When systems introduce artificial intelligence-assisted processing without architectural constraints, they may amplify ambiguity rather than reduce it. Common failure modes include: When systems introduce artificial intelligence-assisted processing without architectural constraints, they may amplify ambiguity rather than reduce it. Common failure modes include:
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 behaviour without visible governance+  * Model or configuration changes altering behavior without visible governance
   * 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 behaviour becoming embedded as de facto semantic authority+  * Tool-specific behavior becoming embedded as de facto semantic authority
  
-These conditions undermine traceability, auditability, comparability, reproducibility, and evidentiary continuity, particularly in regulated reporting environments where organisations must explain outcomes and reconstruct historical interpretations.+These conditions undermine traceability, auditability, comparability, reproducibility, and evidentiary continuity, particularly in regulated reporting environments where organizations must explain outcomes and reconstruct historical interpretations.
  
 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, and permissible uses. 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, and permissible uses.
  • dido/01-fdis-ra/02-background-and-motivation/02-5-why-a-reference-architecture-is-needed/02-5-3-ai-opacity/start.1783886500.txt.gz
  • Last modified: 2026/07/12 13:01
  • by nick_dido