Skip to main content

Intelligence Capability (IC)

Every AI Product must declare the Intelligence Capabilities (IC) through which it provides its intelligence proposition.

IC answers:

What intelligence does this AI Product provide?

This page supersedes the earlier overloaded Capability Type taxonomy. AIPS now keeps Product Type, intelligence capability, agency, embodiment, execution mode, and risk as separate ontology dimensions.


Why Intelligence Capability Matters​

  • Discoverability → consumers can search for the intelligence they need.
  • Comparability → Products can be compared by actual cognitive/intelligence capability rather than implementation labels.
  • Governance → capability-specific obligations can be applied without conflating them with Product Type.
  • Composability → Product designers can understand which intelligence capabilities can participate in higher-order propositions.
  • Evolution → an AI Product can add or remove capabilities without necessarily changing its APT.

Core Intelligence Capabilities​

An AI Product may declare one or more of the following.

Perception​

Interprets sensory or environmental input.

Examples: visual perception, speech perception, sensor interpretation.

Classification​

Assigns an input to one or more categories or classes.

Examples: document classification, transaction categorization.

Detection​

Identifies a condition, anomaly, event, or pattern of interest.

Examples: fraud detection, anomaly detection, object detection.

Prediction​

Estimates an unknown or future outcome.

Examples: default probability, demand forecast, failure prediction.

Generation​

Creates new content or artifacts.

Examples: text, image, audio, video, code, structured content.

Recommendation / Ranking​

Selects, prioritizes, or recommends options.

Examples: next-best-action, ranked search results, product recommendations.

Reasoning​

Derives conclusions from evidence, knowledge, rules, or context.

Examples: symbolic reasoning, evidence synthesis, constraint-based reasoning.

Planning​

Determines a sequence of actions toward a goal.

Examples: task planning, route planning, workflow planning.

Optimization​

Determines preferable configurations, allocations, or actions under objectives and constraints.

Examples: resource optimization, scheduling, portfolio optimization where the optimizer constitutes the intelligence proposition.

Control​

Directs a digital or physical system dynamically.

Examples: autonomous navigation control, robotic manipulation control.


Multi-capability Products​

IC is multi-valued. A Product does not need a single dominant capability.

For example:

Autonomous Navigation AI Product

IC = Perception + Reasoning + Planning + Control

and:

Fraud Detection AI Product

IC = Detection + Classification + Prediction

AIPS therefore does not require a single primaryCapabilityType merely for taxonomy convenience.


Separation from Other Dimensions​

Several common AI labels belong elsewhere in the ontology:

Earlier / common labelAIPS interpretation
PredictiveIC = Prediction
DescriptiveUsually decomposed into capabilities such as Classification, Generation, or Reasoning depending on the actual proposition
PrescriptiveUsually Recommendation, Planning, Optimization, and/or Decision effects
GenerativeIC = Generation
AgenticPrimarily Agency / Autonomy (AR) plus Interaction / Execution Mode (MO); may also involve Planning and Reasoning IC
SymbolicIntelligence mechanism / AI Primitive paradigm; the actual IC may be Reasoning, Planning, Classification, etc.
HybridConstitution / mechanism composition characteristic, not an IC itself
ModelAI Primitive or Model Product, not an IC
AgentProduct/system form characterized by IC + AR + MO rather than a peer IC

Relationship to APT​

APT and IC answer different questions:

APT → Where does the intelligence originate?
IC → What intelligence does it provide?

A Primary and a Derived AI Product can expose exactly the same Intelligence Capabilities.

For example:

Product A: APT = Primary, IC = Detection
Product B: APT = Derived, IC = Detection

Metadata Requirements​

AIPROD should declare:

  • one or more Intelligence Capabilities;
  • a short semantic description of each capability in the Product context;
  • supported tasks/use cases where needed;
  • material capability limitations;
  • version-to-version capability changes.

Capability changes that alter the Product's essential intelligence proposition may require a new major Product version and AI qualification re-evaluation.


Example​

AI Product: Enterprise Research Assistant

  • APT: Primary.
  • IC: Reasoning + Generation + Recommendation + Planning.
  • AR: Assist.
  • MO: Interactive / Agentic goal-driven.
  • IE: Generated report + evidence + recommended next steps.

This is more precise than labelling the Product simply “Agentic”.


Summary​

  • IC defines what intelligence the AI Product provides.
  • IC is multi-valued and orthogonal to APT, Agency, Embodiment, Mode, Effect, and Risk.
  • The previous single “Capability Type” field is deprecated as an umbrella classification.
  • Products should be described through explicit ontology dimensions rather than one overloaded label.

Principle: APT tells us where intelligence comes from; IC tells us what that intelligence does.