Interoperability problems are often described as technical incompatibilities. One system uses JSON while another expects XML. Field names differ. APIs mismatch. Data types conflict. Authentication schemes fail to align. Under this view, interoperability appears to be primarily an engineering problem: standardize formats, normalize schemas, build translation layers, and the systems will communicate successfully.
But beneath most interoperability failures lies something deeper.
Different systems often partition reality differently.
A hospital, insurance provider, pharmacy network, and government agency may all appear to describe the same domain. They may even use many of the same words:
- patient,
- diagnosis,
- provider,
- treatment,
- eligibility,
- encounter.
Yet beneath those shared terms often lie different operational realities, different assumptions, different priorities, and different ways of organizing the terrain itself.
What one system treats as a single entity, another divides into multiple categories. What one system considers essential, another omits entirely. One system models uncertainty explicitly while another collapses it into fixed classifications. One organizes around care, another around billing, another around legal accountability, another around statistical reporting.
The systems are not merely storing different data.
They are stabilizing different interpretations of the same domain.
The triadic structure helps clarify why this happens. Every software system mediates relationships between observers and domains through representational structures:
- schemas,
- ontologies,
- workflows,
- interfaces,
- abstractions,
- categories,
- identifiers,
- and operational assumptions.
These mediating structures emerge from particular goals, constraints, incentives, histories, and organizational contexts. The resulting models are not neutral mirrors of reality. They are selective navigational structures optimized for specific forms of interaction with the terrain.
This is why interoperability becomes difficult even when formats align perfectly.
Two systems can exchange data flawlessly while still misunderstanding each other conceptually. The syntax transfers successfully while the semantics drift underneath. Identical words conceal incompatible partitions of reality.
This problem becomes especially visible at scale.
Large organizations accumulate layers of systems built at different times, by different teams, for different purposes, under different assumptions. Categories harden over years of operational use. Workflows become institutionalized. Interfaces conceal interpretive choices beneath familiar forms. Eventually the models become so embedded in daily practice that they no longer appear as models at all. They simply feel like “the business domain.”
The friction emerges when systems are forced to coordinate across boundaries.
A merger occurs. A government reporting requirement changes. A machine learning system requires unified training data. A healthcare network attempts integrated patient care. A multinational organization standardizes operations across cultures and legal environments. Suddenly it becomes obvious that the systems were never organizing the same reality to begin with.
Artificial intelligence amplifies these tensions further. AI systems increasingly depend on large-scale integration across heterogeneous symbolic environments. But training data assembled from incompatible interpretive frameworks often carries hidden contradictions in:
- labeling assumptions,
- category boundaries,
- relevance structures,
- and operational definitions.
The resulting systems may appear coherent while silently inheriting conceptual instability underneath.
This does not mean interoperability is impossible. Human beings coordinate across differing perspectives constantly. Science itself progresses partly through translation between models operating at different scales and under different assumptions. But successful interoperability usually requires more than technical conversion. It requires interpretive negotiation.
The systems must become aware, at least partially, of the assumptions shaping their representations.
This is why ontology, semantic modeling, metadata, provenance, context-awareness, and explicit representational governance become increasingly important in large-scale systems. The goal is not eliminating differences altogether. Different models often reveal different structures of the same terrain. The challenge is creating frameworks capable of relating these partial perspectives without collapsing them prematurely into false uniformity.
Interoperability is hard because reality does not arrive pre-divided into universally agreed categories.
Every system participates in the division.