Notes on how interoperability emerges
Short essays on semantic interoperability, ontologies, agents and Europe's digital public services, by Claudio Baldassarre.
The semantic cobra effect, part 3: counting every adopter
An adoption count measures reach. A governed ecosystem needs to know whether the parties it exchanges data with have converged, and the count cannot tell it.
Read the postThe semantic cobra effect, part 2: alignments that multiply
A mapping repairs one exchange and leaves the divergence that caused it in place. When programmes count connections as progress, the ecosystem learns to breed mappings.
The semantic cobra effect, part 1: mandates buy compliance
A mandate produces adoption of what it names. Semantic interoperability grows from the decisions taken below that line, after the audit.
Agentic AI and interoperability: a prediction
AI agents will soon take many of the modelling decisions humans take today, at machine speed. Whether that multiplies divergence or accelerates convergence depends on what they can see.
The economics of divergence
Every modelling decision is paid for once. A diverging choice moves the bill from the person who made it to the ecosystem around them, where it grows.
Authority and authoritativeness
A standard's authority comes from who publishes it. Its authoritativeness comes from who builds on it. Programmes that confuse the two choose by letterhead.
When does a vocabulary become an asset?
Publishing a vocabulary creates an artifact and a maintenance commitment. It becomes an asset when the parties around it build on it, and the same vocabulary can be an asset in one ecosystem and a liability in another.
Interoperable with whom, and for what?
Interoperability is always between particular parties, for a particular purpose. A programme that names both can measure progress; one that names neither can only report activity.
What is a governed ecosystem, and are you in one?
A governed ecosystem is any circle of parties bound by shared rules, with defined roles in their common models, a shared purpose and a real cost when those models diverge. Most organisations are in several without naming them.
If the models converge, are we converging too?
Research suggests AI models are converging on a shared representation of reality. As we think through them, our own understanding may follow.
Semantic models that last encode decisions
If an LLM can describe your domain, what is a semantic model for? The ones that earn their keep record what an organisation has decided.
Interoperability is an emergent property
Semantic interoperability appears when many parties, over years, can see each other's models and move towards them. The work is to create those conditions.
Cross-border interoperability is an effect
Services that work across borders follow from even digital maturity between countries. The fastest way to level it is to transplant what already works.
Semantic interoperability needs a metadata contract
Digital services run on contracts. Exchanging data by its meaning needs a second contract, on the data itself, written in a formalism that is not open to interpretation.
What an AI agent saw in a national catalogue
An experiment: letting an AI agent explore a national catalogue of public-sector semantic assets, and asking how interoperable its regions are.
From interoperability to negotiability
Pre-agreed semantics will always leave gaps. Agents that reason over formal models can negotiate meaning at the moment of exchange.