AI Product Ontology
The AI Product Ontology defines what constitutes an AI Product, how AI Products are intrinsically classified, and how their intelligence, consumption, agency, embodiment, risk, and dependency characteristics are described.
AIPS specializes the general Product ontology defined by BPS. It therefore separates:
- Productness — whether something qualifies as a Product at all;
- AI Product Family qualification — whether the Product's intrinsic nature is genuinely artificial intelligence;
- AI Product Type (APT) — how the productized intelligence is constituted;
- orthogonal characteristics — what the intelligence does, how it is consumed, what authority it exercises, how it interacts with the physical world, and what governance applies.
Core principle: Uses AI does not imply is an AI Product.
1. Canonical AI Product Definition
An AI Product is a governed Product whose value-bearing essence is a distinct artificial-intelligence capability and whose current constitution directly or transitively contains at least one qualifying Artificial Intelligence Primitive.
An AI Product may provide one or more intelligence capabilities including inference, perception, classification, detection, prediction, generation, recommendation, reasoning, planning, optimization, decision formation, or intelligent action.
A modern machine-learning model is one possible intelligence primitive, but AIPS does not require every AI Product to depend on a trained ML model. Symbolic reasoners, planners, knowledge reasoners, intelligent search, and hybrid intelligence mechanisms can also qualify where they materially constitute artificial intelligence.
2. AI Product Qualification
A candidate Product qualifies for the AI Product Family only when all of the following are true.
AQ1 — Product qualification
The candidate first qualifies as a Product under BPS. A script, notebook, model checkpoint, prompt, dataset, feature store, transient workflow, or experiment does not become an AI Product merely because it is AI-related.
AQ2 — Intelligence is the value-bearing essence
Artificial intelligence itself is the primary value being productized.
A Product may contain or use AI while having another intrinsic nature. For example, a smartphone can be AI-enabled while remaining intrinsically a physical/engineered Product.
AQ3 — Distinct intelligence proposition
The Product exposes a distinct, governed intelligence proposition such as:
- detect payment fraud;
- generate product imagery;
- recommend suitable actions;
- reason over governed knowledge;
- plan autonomous navigation.
Simple routing or orchestration of existing AI Products does not automatically create another AI Product.
AQ4 — Qualifying AI Primitive dependency
The Product's direct or transitive constitution contains at least one qualifying Artificial Intelligence Primitive.
The presence of an AI Primitive is necessary but not sufficient. AQ1–AQ3 must also hold.
AQ5 — Governed Product boundary
The intelligence capability has a defined Product boundary with appropriate ownership, lifecycle, contracts, controls, versioning, observability, evaluation, and governance.
3. Artificial Intelligence Primitive
An Artificial Intelligence Primitive is a computational capability capable of materially contributing artificial inference, learning, perception, reasoning, generation, planning, optimization, decision formation, or intelligent action.
Typical primitive classes include:
| Primitive class | Examples |
|---|---|
| Learned Model | LLM, classifier, vision model, embedding model |
| Symbolic Reasoner | Expert-system inference engine |
| Knowledge Reasoner | Ontology or knowledge-graph reasoning engine |
| Planner | Goal/action planner |
| Intelligent Search | Heuristic or state-space search capability |
| Optimizer | Optimization engine where it constitutes intelligent behaviour |
| Hybrid Intelligence Engine | Learned + symbolic/reasoning mechanisms |
An AI Primitive may itself be productized, for example as a Model Product, without automatically becoming an AI Product.
4. Model Product vs AI Product
A Model Product is a Product whose value-bearing essence is a reusable learned computational capability embodied in a governed model artifact or package.
A downloadable LLM sold by a model vendor can therefore be:
- intrinsically a Model Product;
- contextually the vendor's Business Product;
- an upstream AI Primitive dependency of a downstream AI Product.
A Productized model is therefore not automatically an AI Product merely because it has metadata, governance, packaging, or an API.
5. AI Product PIN Specialization
AIPS specializes BPS Product Intrinsic Nature (PIN) as follows:
| PIN dimension | AI Product interpretation |
|---|---|
| VBE — Value-Bearing Essence | Governed artificial intelligence |
| EVC — Essential Value Capability | The essential intelligence proposition provided |
| CON — Constitution | AI primitives + data + knowledge + rules + tools + controls + potentially upstream AI Products |
| INV — Identity Invariants | Conditions that must remain true for the Product to remain the same intrinsic AI Product kind |
6. AI Product Type (APT)
AI Product Type (APT) is the family-specific intrinsic classification of an AI Product based on the provenance and construction structure of its productized intelligence.
APT answers:
Where does the productized intelligence originate?
APT01 — Primary AI Product
A Primary AI Product is an AI Product whose productized intelligence is constituted without a material upstream AI Product dependency, directly from one or more AI primitives together with governed data, knowledge, rules, tools, and controls.
Using one or many Model Products does not by itself make the AI Product Derived.
APT02 — Derived AI Product
A Derived AI Product is an AI Product whose distinct intelligence proposition materially depends upon the governed intelligence capability or output of one or more upstream AI Products.
Composite derivation
Where multiple upstream AI Products materially contribute to a new intelligence proposition, the Product remains Derived and its derivation topology may be classified as Composite.
Composite is a derivation topology, not a peer APT.
7. Orthogonal AI Product Classification Dimensions
APT must not be used to encode unrelated characteristics. AIPS therefore models the following dimensions separately.
| Dimension | Question answered | Example values |
|---|---|---|
| APT — AI Product Type | Where does the intelligence originate? | Primary, Derived |
| IC — Intelligence Capability | What intelligence does it provide? | Perception, Classification, Detection, Prediction, Generation, Recommendation, Reasoning, Planning, Optimization, Control |
| CS — Consumption Scope | Who / where consumes it? | Local, Cross-Division, Global/Federated, External |
| PU — Purpose | Why is it consumed? | Decision Support, Operational Automation, Compliance & Risk, Client Value, Revenue Enablement, Innovation |
| MO — Interaction / Execution Mode | How is it invoked or operated? | Batch, Real-time/Event-driven, Synchronous API, Interactive, Embedded, Continuous, Agentic/Goal-driven |
| IE — Intelligence Effect | What does it produce or cause? | Score, Classification, Prediction, Alert, Recommendation, Generated Content, Decision, Plan, Digital Action, Physical Action |
| AR — Agency / Autonomy | What authority does it exercise? | Inform, Recommend, Assist, Act with Approval, Autonomous Action |
| EM — Embodiment | How does it relate to the physical world? | Non-embodied, Physical-world Perceptive, Embodied/Actuating, Closed-loop Autonomous |
| RT — Risk Tier | What AIPS governance regime applies? | R0–R4 |
Risk Tier is computed from evidence-backed drivers and is paired with a version-specific Governance Obligation Profile. Governance obligations are derived from AR, EM, IE, topology, impact/currentness evidence and applicable policy; PrimaryArchetype is not a normative risk input.
These dimensions are deliberately orthogonal. For example, an AI Product can be Primary + Detection + Real-time + Recommend + Non-embodied + R3.
8. AI-enabled and Compound Products
A Product does not inherit AI Product Family membership merely because AI contributes to its construction or operation.
AIPS distinguishes the role AI plays in the broader Product constitution:
- Constructional — AI was used to create the Product, such as an AI-generated movie;
- Enabling — AI materially improves the Product but is not required to preserve its Product kind, such as a smartphone with an AI assistant;
- Essential — AI is co-essential to Product identity, such as the intelligence contribution within a fully autonomous humanoid Product.
A broader BPS Compound Product may therefore have AI as one of multiple co-essential intrinsic family contributions without the entire Product being classified exclusively as an AI Product.
9. Version-specific AI Qualification
AI Product Family membership and APT are version-specific, evidence-backed classifications.
A Product identity may persist while a later version changes its intrinsic constitution enough to alter AI qualification.
AIPS therefore expects AI Product versions to maintain AI Qualification Evidence describing why the current version qualifies as an AI Product.
Typical qualification states are:
- CURRENT
- STALE
- REVALIDATION_REQUIRED
- INVALID
A change to an upstream dependency must trigger impact analysis for downstream Product versions that actually adopt the changed dependency.
10. Core Ontology Principles
- Intelligence Primacy — intelligence itself must be the Product's value-bearing essence.
- Primitive Dependency — every AI Product has at least one qualifying AI Primitive in its direct or transitive constitution.
- Necessary but Not Sufficient — using AI technology does not by itself establish AI Product status.
- Model/Product Separation — a Model Product is not synonymous with an AI Product.
- Type/Capability Separation — APT describes intelligence provenance; IC describes what the intelligence does.
- Derivation Semantics — Derived requires a material upstream AI Product dependency, not merely a Model Product or Data Product dependency.
- Composite Topology — multiple upstream AI Products may create a Composite Derived topology without creating another peer APT.
- Contextual Independence — purpose, mode, risk, agency, embodiment, and Business Product role do not determine APT.
- Qualification Currentness — AI Product qualification must remain supported by current, version-specific evidence.
- Non-Inheritance — a Product does not become an AI Product merely because it contains, consumes, or was constructed using AI.
- Compound Compatibility — AI may be one of multiple co-essential family contributions in a Compound Product.
- Semantic Continuity — a Product's business meaning may persist even if a later version changes Product Family qualification.
11. Canonical Ontology View
12. Example — Fraud Detection AI Product
| Dimension | Example classification |
|---|---|
| Product Family | AI Product |
| APT | Primary |
| IC | Detection + Classification |
| CS | Cross-Division |
| PU | Compliance & Risk + Operational Automation |
| MO | Real-time / Event-driven |
| IE | Fraud score + alert + explanation |
| AR | Recommend |
| EM | Non-embodied |
| RT | R-tier as assessed under AIPS |
| Data dependencies | Governed payment/customer/fraud-history Data Products |
| AI Primitive | Registered fraud model |
| AI Qualification | Current |
13. Relationship to AIPROD and AIPDS
The ontology defines what the AI Product is. It does not replace the semantic or deployment specifications.
- AIPROD records the Product's governed semantic description, including AI qualification, APT, intelligence capabilities, dependencies, governance, and lifecycle metadata.
- AIPDS describes how the AI Product is operationalized, executed, served, secured, observed, scaled, and evolved.
AIPS implementations should preserve this separation so that Product semantics remain stable even when deployment technologies change.