Skip to main content

AI Product Classification Dimensions

AIPS does not use one overloaded "capability type" field to represent fundamentally different AI characteristics.

Terms such as Model, Agent, Generative, Symbolic, Agentic, AGI, and Hybrid answer different questions and should not be treated as peers in a single taxonomy.

AIPS therefore separates AI Product classification into orthogonal dimensions.


1. AI Product Type (APT)​

APT answers:

Where does the productized intelligence originate?

  • Primary — intelligence is constituted without a material upstream AI Product dependency, directly from AI primitives together with data, knowledge, rules, tools, and controls.
  • Derived — the distinct intelligence proposition materially depends upon one or more upstream AI Products.

A Derived AI Product may have a Composite derivation topology where multiple upstream AI Products materially contribute.

See AI Product Ontology.


2. Intelligence Capability (IC)​

IC answers:

What intelligence does the Product provide?

An AI Product may declare one or more capabilities.

CapabilityMeaning
PerceptionInterpret sensory or input environments
ClassificationAssign categories or classes
DetectionIdentify conditions, anomalies, or events
PredictionEstimate future or unknown outcomes
GenerationProduce new content or artifacts
Recommendation / RankingSelect, prioritize, or rank options
ReasoningDerive conclusions from evidence, rules, or context
PlanningDetermine actions or sequences toward a goal
OptimizationDetermine preferable configurations or actions
ControlDirect a system or environment dynamically

IC is intentionally multi-valued. A physical autonomous AI Product may legitimately declare:

Perception + Reasoning + Planning + Control


3. Consumption Scope (CS)​

CS answers:

Who or where consumes the AI Product?

Typical values include:

  • Local
  • Cross-Division
  • Global / Federated
  • External

4. Purpose (PU)​

PU answers:

Why is the intelligence consumed?

Typical values include:

  • Decision Support
  • Operational Automation
  • Compliance & Risk
  • Client Value Creation
  • Revenue Enablement
  • Innovation
  • Research / Exploration

Purpose does not determine APT.


5. Interaction / Execution Mode (MO)​

MO answers:

How is the intelligence invoked or operated?

Typical values include:

  • Batch
  • Real-time / Event-driven
  • Synchronous API
  • Interactive
  • Embedded
  • Continuous
  • Agentic / Goal-driven

This dimension is broader than inference serving and accommodates symbolic, agentic, and physical AI execution patterns.


6. Intelligence Effect (IE)​

IE answers:

What does the intelligence produce or cause?

Typical effects include:

  • Score
  • Classification
  • Prediction
  • Detection / Alert
  • Recommendation
  • Generated Content
  • Explanation / Evidence
  • Decision
  • Plan
  • Digital Action
  • Physical Action

The term Effect is deliberately broader than "Output" because embodied AI may materially act on the physical world.


7. Agency / Autonomy (AR)​

AR answers:

What causal or decision authority is delegated to the AI Product?

A conceptual scale is:

  1. Inform — provides intelligence or evidence;
  2. Recommend — proposes an outcome;
  3. Assist — participates in a human-controlled workflow;
  4. Act with Approval — prepares an action requiring authorization;
  5. Autonomous Action — acts within delegated constraints without per-action human approval.

Agency is independent from APT. A Primary AI Product and a Derived AI Product can both be highly autonomous or purely advisory.


8. Embodiment (EM)​

EM answers:

How does the AI Product relate to the physical world?

Typical values include:

  • Non-embodied — operates in the digital domain;
  • Physical-world Perceptive — senses or interprets the physical world;
  • Embodied / Actuating — controls physical systems or devices;
  • Closed-loop Autonomous — continuously perceives, reasons, and acts in the physical world.

Embodiment is an orthogonal AI characteristic, not an APT.


9. Risk Tier (RT)​

AIPS retains R0–R4 as the AI governance risk classification.

Risk Tier is independent from Product Type. Risk is computed from evidence-backed ontology/context drivers and produces a Governance Obligation Profile; no single IC label acts as the risk type. For example:

APT = Primary
IC = Detection + Classification
AR = Recommend
EM = Non-embodied
RT = R3

is a coherent classification.


10. Where Existing Terms Belong​

The following commonly used AI labels should not be collapsed into a single Product Type field.

TermBetter interpretation in AIPS
ModelAI Primitive and potentially a Model Product
AgentProduct/system form with characteristic Agency, Mode, and Capability dimensions
Multi-Agent SystemTopology / composition characteristic
GenerativeIntelligence Capability
PredictiveIntelligence Capability
SymbolicIntelligence mechanism / primitive paradigm
HybridConstitution / mechanism composition characteristic
AgenticAgency and Interaction/Execution characteristics
AGI / ASIBreadth/generality of intelligence; future taxonomy dimension, not current APT

11. Why the Separation Matters​

Orthogonal dimensions improve:

  • Discoverability — consumers can search by the characteristic they actually care about;
  • Governance — obligations can be computed from risk, agency, embodiment, effect, and capability without overloading one field;
  • Composability — intelligence provenance remains distinct from workflow topology;
  • Versioning — a Product can change mode, agency, risk, or implementation without necessarily changing APT;
  • Future-proofing — new mechanisms can be introduced without redefining the fundamental ontology.

Principle: APT describes where the intelligence comes from; IC describes what it does; AR describes what authority it has; EM describes how it relates to the physical world; RT describes what governance applies.