This post will focus on semantic interoperability when an open architecture contains or interacts with multiple ontologies and conceptual data models. Before proceeding, a few definitions are in order.
Semantics is the branch of linguistics concerned with the meaning of different speech forms including their nature, structure, content, and context. In open architecture the term may be applied to any method of conveying information including data, messages and models.
Epistemology is the branch of metaphysics concerned with the nature, sources, and limits of knowledge.
Ontology is the branch of metaphysics that is traditionally concerned with the nature of reality or existence. In modern philosophy it is sometimes defined as a ‘modality of existence’, assuming we can reason about more than one modality. In open architecture, the term is applied to the structural models of concepts.
The Ontology and Epistemology of Conceptual Models
In philosophy we generally accept two ontological realms: physical reality and institutional reality. Physical reality is epistemically objective in that its existence is independent of viewpoint. Its epistemic source is the ‘world’, which is mapped into the ‘mind’. Physical reality is expressed through statements that have the property of being true or false. Institutional reality is, in contrast, epistemically subjective in that it is created by institutions and belongs to institutions. The epistemic source is therefore the ‘mind’, which is mapped into the ‘world’. Institutional reality is expressed through declarations that have the property of authority.
Concepts within models of physical reality include observables and causation. As engineers we may assume that a concise yet adequate conceptual model of physical reality can be developed for a message set or data set of interest.
The challenge is striking the balance between what is concise yet adequate, as opposed to universal but unnecessarily complex. The larger the scope of the open architecture, the more unwieldy and difficult an attempt at a universal conceptual model becomes. As an example it might be adequate in an open architecture to conceptualize time-space as Aristotelian (bodies exist in absolute time and absolute space). But if the requirement is to rigorously model the relationship between different phenomenology of position it would more useful to model time-space as Newtonian (bodies exist in relative space) where the reference point for position is at the same conceptual level and is not a later refinement. The correct fidelity of the model is therefore driven by system requirements. In some systems, such as satellite-based navigation, it might be necessary to conceptualize time-space as Einsteinian (bodies existing in relative time and relative space).
With conceptual models that attempt universality there has been a tendency throughout history to fall into ontological arguments. That is where the a priori reasoning behind the model is ‘proved’ by the model itself. David Hume demonstrated that valid ontological arguments are not possible. A typical example seen in open architecture is an appeal to the internal rigor of the model (mind) to ‘prove’ a claim about the external truth of the model (world). As stated earlier, the epistemic direction for physical reality is the other way.
Emmanuel Kant distinguished between analytic judgments and synthetic judgments. An analytic judgment is an assertion about the a priori reasoning within to the model. A synthetic judgment is an assertion about something external to the model. It is critical that conceptual models are evaluated by synthetic judgments through the process of conjecture and refutation.
Institutional reality is different from physical reality and requires different epistemic and ontological treatment. Concepts in this realm are created by declarations that have the attribute of authority. Examples of institutional reality in a military aircraft system are ‘target’, ‘air-tasking order’, and ‘weapon system’. These are human inventions that are regulated by institutions. To express institutional reality, John R. Searle introduced the construct of ‘counts-as’ to symbolic logic. The general form is: A counts-as B in context C. Institutional examples of the counts-as relationship are:
- This river counts as the boundary in the context of this tribe.
- A raised hand counts as a vote to approve in the context of this election.
These examples show that there is not necessarily an intrinsic conceptual relationship between A and B (even in context C) until declared by an authority or authoritative process within the institution. Targets are declared and prosecuted within the Joint Targeting System using weapon systems that are designed and regulated.
Whereas the challenge with modeling physical reality is balancing conciseness with universality, the challenge with modeling institutional reality is recognizing the pace of change within society and the different institutional realities that exist in different domains. Open business models require that open architecture encourage and foster innovation. This requires ‘semantic openness’ in which component developers and integrators can innovate by declaring new domains (realities/ontologies) and establish counterpart relationships between them.
There have been recent attempts to achieve system-of-systems interoperability through the imposition of a universal conceptual data model to which all governed systems are a logical refinement. This strategy is breaking down for these reasons:
- A universal model of physical reality is theoretically possible but too unwieldy to be useful in a single system. Models of physical reality have always been a balance between conciseness/cost and adequacy/precision.
- A universal model of institutional reality is constraining to growth and innovation. An open architecture should allow third parties to provide content to create or connect to other conceptual models in different systems.
In practice the universal conceptual model is realized in the diagram below where a ‘system of systems’ conceptual model continues to grow with each system that is added. Recent experience has shown that such a strategy does not converge to a steady-state model. The initial model may not indulge physical reality sufficiently for a new system, or new institutional concepts will have to be reconciled with the existing concepts in the universal model. As the model grows into many tens of thousands of hours of effort there is a temptation to defend the model through analytic judgments: the model itself proves the a priori reasoning behind the model and thus the error of any externally sourced reality that contradicts it. We must guard against recreating the Middle Ages in open architecture.
The translator concept also assumes that System 1 and System 2 are interoperable entirely through the exchange of data and are not behaviorally coupled.
The alternative strategy is shown below, whereby the attempt at universality is abandoned in favor of a series of conceptual models that are concise (and thus affordable) yet adequate for their stakeholder needs. Each model efficiently expresses it own conceptual world (or domain) without being polluted by the concerns of other conceptual worlds. Relationships between conceptual worlds are established where counterparts exist through counterpart relationships, but not all concepts in model must have counterparts in another.
There are advantages in domain based partitioning of a system or system-of-systems architecture. First, it bounds the scope of individual modeling efforts to a single problem space, which can focus on a limited set of institutional concepts and tolerable approximations of physical reality to meet stakeholder needs. Second, it allows effective extension of the system-of-systems architecture indefinitely in a manner that is consistent with open business models. Components can be added to architecture repositories that create new models to solve new problems without centralized management of a universal model. Business innovation is fundamentally about seeing and creating new counterpart relationships to address market needs. Third, the behavioral relationship between systems can be managed locally within the subject matter of the component that ties the systems.
What is important therefore in semantic interoperability is not conceptual universality but the ability to create ad hoc relationships when components or architectures are added. It is advantageous if common rules of construction pervade architectures, but not totally necessary. The bridges between conceptual worlds are established by domains that assert counterpart relationships between their models. These relationships may not be foreseeable in the original scope of the open architecture.
Open architectures themselves must be ‘open source’.