AI system classes mapped to DIKW layers and decision phases, with typical autonomy risks and safeguards
| AI class | Primary DIKW contribution | Decision phases most affected (frame/evaluate/commit/ enact) | Typical autonomy risks | Autonomy-preserving safeguards |
|---|---|---|---|---|
| Predictive/descriptive (ML and data mining) | Data → information (aggregation, patterning) | Frame, evaluate | Over-weighting historical patterns; metric myopia; false objectivity | Data governance and provenance; counter-metric reviews; alternative framing prompts; sandbox scenario tests |
| Prescriptive/optimization and recommenders | Information → knowledge (action suggestions) | Evaluate, commit | Optimization without values; goal mis-specification; automation bias | Value/constraint reviews; multi-objective criteria; opt-out and override protocols; red-team “what-ifs” |
| Generative AI assistants (LLMs) | Information ↔ knowledge (summarize, ideate, articulate) | Frame, evaluate, enact | Fluent but unfounded rationales; leakage of internal knowledge; prompt steering | Source citation/RAG; confidentiality controls; traceable prompts; justification checklists |
| Perception and extraction (NLP/CV/OCR) | Data formation (from unstructured inputs) | Frame, enact | Garbage-in; biased extraction; loss of context | Human validation on samples; bias audits; retention of raw context; escalation rules |
| Primary | Decision phases most affected (frame/evaluate/commit/ enact) | Typical autonomy risks | Autonomy-preserving safeguards | |
|---|---|---|---|---|
| Predictive/descriptive ( | Data → information (aggregation, patterning) | Frame, evaluate | Over-weighting historical patterns; metric myopia; false objectivity | Data governance and provenance; counter-metric reviews; alternative framing prompts; sandbox scenario tests |
| Prescriptive/optimization and recommenders | Information → knowledge (action suggestions) | Evaluate, commit | Optimization without values; goal mis-specification; automation bias | Value/constraint reviews; multi-objective criteria; opt-out and override protocols; red-team “what-ifs” |
| Generative | Information ↔ knowledge (summarize, ideate, articulate) | Frame, evaluate, enact | Fluent but unfounded rationales; leakage of internal knowledge; prompt steering | Source citation/RAG; confidentiality controls; traceable prompts; justification checklists |
| Perception and extraction (NLP/CV/ | Data formation (from unstructured inputs) | Frame, enact | Garbage-in; biased extraction; loss of context | Human validation on samples; bias audits; retention of raw context; escalation rules |
Sharing content requires targeting cookies to be enabled. Please update your cookie preferences to use this feature.