Most approaches to semantic interoperability assume the semantics are agreed in advance. Two parties align their models, and exchange works as long as reality stays within what they agreed.
In practice, mismatches persist. At best we reach shared semantics that are good enough for the cases we anticipated.
Semantic negotiation
A different approach is becoming possible: semantic negotiation. It combines agents, language models and the formal knowledge representations of the semantic web.
Agents equipped with language models can do more than read expressive representations such as OWL and RDF. They can reason with them. That lets them close semantic gaps at the moment of exchange: aligning models, resolving inconsistencies and working around flaws in imperfect models, instead of failing on them.
Picture services that negotiate meaning themselves, with agents mediating intent, context and logic between systems.
A shift in the paradigm
This moves digital services from interoperability to negotiability. Today’s web interfaces are likely to be wrapped by agents able to negotiate every aspect of an interaction: identity, semantics, authorisation, request and response.
Two years ago, many people asked whether semantic interoperability would still be needed once language models arrived. My intuition then was that negotiation was well within the power of these models. As they embed semantics more deeply in their representations, that intuition becomes more practical.
Formal models still matter
Negotiation does not make formal models obsolete. It makes them more useful. An agent can only negotiate well if it has something precise to reason with. Decades of static ontologies were built for a world without such agents; now the agents can put them to work.
The goal changes from perfect shared semantics to semantics that can be negotiated reliably. That is a goal we can reach.