There Is No Single Viewpoint of Ontology

When people speak of ontology in computing, they often assume that it refers to one identifiable method: defining classes, specifying relations, writing axioms, and enabling machines to reason over a domain.

That description is not wrong. It is simply incomplete.

One of the central conclusions of my PhD dissertation, The Ontology of Structured Knowledge in the AI Era: The Imperative of Semantic Foundations, is that there is no single viewpoint from which ontology can be adequately understood or practised.

My own work has taken me across domains that appear, at first, to have little in common: music theory and computational creativity, library science, digital humanities, chemical engineering process safety, and robotic manipulation. Yet each of these domains encounters the same foundational problem. The meanings on which its computational systems depend cannot be reduced to unexamined data structures, isolated labels, or statistical associations.

What differs is the kind of semantic work that ontology is expected to perform.

In library science, ontology is closely connected to identity, authority, provenance, and continuity across records. A person may appear under different names. A work may exist through multiple editions, translations, performances, and manifestations. Records created by different institutions may overlap without being equivalent. The task is not simply to place items into categories, but to preserve distinctions among persons, works, versions, sources, and institutional assertions while making them discoverable across systems.

Ontology here supports integration, but integration does not mean erasing difference. It requires an explicit account of what is being identified, what is merely being associated, and on whose authority a particular assertion is made.

Digital humanities extends this problem into interpretive and historical territory.

Cultural knowledge is often distributed across incomplete archives, inconsistent metadata, inherited classifications, variant terminology, and sources produced under different historical conditions. The categories used to describe a theatrical form, musical practice, historical actor, or cultural artefact may themselves be objects of research rather than neutral containers for information.

An ontology in this context may help reconstruct connections among people, texts, performances, institutions, places, and historical concepts. But it must also preserve the provenance and conditions under which those connections are asserted. A computationally convenient unification can become historically misleading if it collapses distinct terminologies, retrospective classifications, or competing interpretations into a single supposedly authoritative structure.

Ontology therefore becomes part of the research infrastructure. It does not merely organise established facts; it makes the construction, limitation, and contestability of those facts inspectable.

In chemical engineering process safety, the semantic demand changes sharply.

Here, ontology may serve as a mechanism of disciplined constraint. Concepts must be defined precisely enough to support validation, detect missing information, preserve provenance, and prevent unsafe inferences. Ambiguity is not merely an intellectual inconvenience. It may conceal an incomplete process model, an unidentified hazard, a missing connection, or an absent safety-critical component.

An ontology may need to distinguish a process unit from the material stream connected to it, a design structure from an observed operating condition, and a possible hazard from an adequately supported hazard claim. It may also need to work together with validation rules that identify incomplete or structurally inconsistent descriptions before they influence engineering analysis.

Here, ontology participates in a regime of accountability. Its value lies in making assumptions explicit, enforcing justified distinctions, and enabling claims to be inspected before they affect consequential decisions.

In music theory and computational creativity, however, excessive semantic closure can produce a different kind of failure.

Musical concepts do not always benefit from being forced into a single final definition. Tonality, style, consonance, structural function, formal coherence, and aesthetic value may be understood differently across historical periods, theoretical traditions, compositional practices, and analytical purposes.

A rigidly unified ontology could erase precisely the plurality that makes interpretation and creation possible.

Ontology must therefore do more than eliminate ambiguity. It may need to preserve meaningful difference: to represent several definitions of a concept, identify their sources and conditions of use, and allow semantic structures to develop as new analytical or creative practices emerge.

The objective is not unrestricted vagueness. It is structured plurality: an explicit representation of differences that matter, without mistaking uniformity for understanding.

This distinction is particularly important in computational creativity. A knowledge representation that merely freezes an established theory may reproduce a recognised style, but it may also prevent the system from reformulating its own conceptual space. If creativity involves the development, displacement, or recombination of semantic structures, then an ontology may need to support change rather than merely preserve a finished classification.

Robotics and Physical AI introduce a further configuration because several semantic regimes must operate together.

Robotic and Physical AI systems act through the interaction of embodied capabilities, learned models, task specifications, sensor observations, object affordances, environmental conditions, safety constraints, and human purposes. Some of these elements require precise and stable specification. Others remain uncertain, probabilistic, context-dependent, or subject to revision during execution.

An ontology in this context cannot be treated merely as a static catalogue of objects and actions. It must mediate among symbolic specifications, learned behaviours, embodied observations, physical constraints, task intentions, and evaluative criteria, while preserving the different epistemic status of each.

To act safely, a system may need to commit to an actionable interpretation of the current situation. In an assembly task, for example, it must identify the component specified by the work order, distinguish it from visually similar components, and suspend manipulation if a worker enters the shared workspace. Yet that interpretation must remain revisable when the task changes, perception is uncertain, or new instructions alter the intended action.

Ontology in robotics and Physical AI must therefore support both semantic commitment and semantic revision without confusing one with the other.

Taken together, these domains reveal that ontology is not defined by a single attitude toward meaning.

In one context, the primary concern is identity across heterogeneous records. In another, it is the preservation of historical provenance and interpretive plurality. In process safety, semantic incompleteness may conceal physical danger. In computational creativity, excessive closure may suppress the conceptual variation from which novelty emerges. In robotics, stable commitments and dynamic reinterpretation must coexist within the same operational system.

These are not merely different applications of an otherwise uniform technique. They expose different orientations within ontology itself.

Some ontologies operate as formal specifications. Some support authority control and information integration. Some function as research infrastructures through which evidence and interpretation can be traced. Some define constraints whose violation must be detected. Others preserve multiple conceptualisations or provide a semantic space capable of further development. The mistake is to assume that one of these orientations represents “real ontology” while the others are secondary, impure, or defective.

Ontology is better understood as the disciplined construction and governance of semantic commitments under different epistemic, operational, historical, and ethical conditions.

The central responsibility of the ontologist is therefore not merely to construct taxonomies, encode domain terminology in OWL, or maintain logically consistent knowledge graphs. It is to determine which semantic commitments a computational system should make, how strongly it may make them, under what conditions they remain valid, and how their consequences can be inspected.

This responsibility requires judgement. The ontologist must decide when interpretation should be constrained and when plurality should be preserved; when two records denote the same entity and when they should remain only qualifiedly associated; when missing information constitutes a validation failure; when competing definitions represent inconsistency and when they express legitimate conceptual difference; and which claims may be inferred, which may only be checked, and which must remain attached to their evidence, provenance, and conditions of use.

The relevant expertise is consequently integrative rather than tool-specific. Formal logic, conceptual analysis, domain investigation, provenance modelling, validation, and computational implementation are required because semantic commitments do not arise from syntax alone. They must be justified in relation to the purposes, evidence, risks, and historical conditions of the system in which they operate. Every ontology is selective: it makes some distinctions computationally visible while excluding, postponing, or simplifying others. The ontologist must therefore be accountable not only for what a system can represent and infer, but also for what its semantic structure renders invisible or treats as settled.

There is no single viewpoint of ontology because different domains require different regimes of semantic commitment. This plurality does not weaken ontology as a discipline; it defines its responsibility. In the AI era, ontology matters because increasingly consequential systems do not merely store or retrieve information. They interpret, infer, recommend, generate, and act. Ontology provides the discipline required to determine how meaning may become computationally actionable without being historically distorted, operationally unsafe, conceptually impoverished, or represented with a certainty that the available evidence cannot support.