Data & Information

Data and information exist in layers of varying certainty and permanence, from deterministic source-of-truth records to contextual AI-generated inferences that users can configure based on their needs.

Two-dimensional framework

Data can be categorized along two independent dimensions that determine trustworthiness and appropriate treatment:

Certainty

Deterministic

• Same input always produces same output
• No probability or confidence scores needed
• Fully explainable through rules or computation
• Examples: RDF inferences, aggregations, rule-based derivations

Probabilistic

• Same input may produce different outputs
• Confidence scores communicate uncertainty
• Explanations are approximations
• Examples: AI pattern detection, ML predictions, semantic similarity

Provenance

Explicit

• Created directly by users or source systems
• Manually entered or imported
• Examples: Form submissions, uploaded files, user selections

Derived

• Computed from explicit data using rules or logic
• Deterministic transformations
• Examples: Calculated totals, RDF inferences, taxonomy classifications

Inferred

• Discovered through pattern analysis
• Typically probabilistic
• Examples: Semantic similarity, AI-detected relationships, clustering

Deduced

• Suggested possibilities not yet confirmed
• Speculative, lowest certainty
• Examples: AI recommendations, predicted missing data, suggested next steps

Layered data architecture

1. Foundation: Explicit deterministic data—user input, source records, canonical relationships
2. Computed: Derived deterministic data—aggregations, rule-based inferences, RDF reasoning
3. Inferenced: Inferred probabilistic data—AI-discovered patterns, semantic connections, learned models
4. Speculation layer: Deduced probabilistic data—AI suggestions, domain knowledge applications, predicted needs

Configurability principle

Different users and contexts benefit from different ratios of hard to soft data. The design may support adjustable layering:

• Intensity: How much soft data to show. Minimal (hard data only) → Moderate → Maximum (all inferences)
• Specificity: Which domain knowledge to use. Generic patterns → Domain-specific → User-provided
• Interactivity: How soft data behaves. Passive (view only) → Suggestive (accept/reject) → Active

Resources

• Hard and Soft — Amelia Wattenberger on interface material properties
• Data and Reality — William Kent on the nature of information systems
• Living in Information — Jorge Arango on information environments