From an article to a white paper: my collaboration with Michael Iantosca

Welcome to the first post on AI TechDoc Knowledge. It seems fitting to begin with a project that shows exactly what this publication is about: the point where structured technical content, semantic systems and artificial intelligence meet.

How it started

Some time ago I wrote an article titled “EvoOntology vs. iiRDS: understanding two divergent ontology paradigms.” I was interested in the tension between two ways of thinking about meaning: carefully governed ontologies on one side, and models that can evolve from experience on the other.

That article sparked an idea for a project Michael Iantosca was working on. Michael is Senior Director of Content Platforms and Knowledge Engineering. He spearheaded DITA XML and is one of the best-known information architects in our field — someone whose work has shaped how an entire industry structures technical content.

His enthusiasm was so immediate and genuine that he spontaneously sent me his revised white paper. And, as communication specialists and technical writers tend to do, I shaped it into a clear, structured and visually coherent form.

What the white paper is about

From Knowledge Graphs to Adaptive Semantics — Evolving DITA and iiRDS Digital Twins for AI is a plain-language introduction to a simple but far-reaching idea:

Inference extends what the knowledge graph knows. Ontology evolution extends how the knowledge graph can represent knowledge.

DITA, iiRDS and product ontologies already give organisations a strong semantic foundation. Connected through a knowledge graph, they form a digital twin of the technical information environment that AI agents can navigate. An inference engine can derive new facts inside that model — but it cannot notice when the model itself is missing something.

That is where the white paper places EvoOntology. It isn't presented as a competing ontology but as an adaptive semantic feedback loop around the governed core, with five steps:

1. Learn how AI agents actually use the model
2. Detect gaps the ontology cannot express
3. Propose a new concept, relationship or rule
4. Evaluate whether agent performance improves
5. Govern by promoting validated changes into the enterprise model

A maintenance example makes this concrete. Technicians keep asking which procedure should follow a component replacement, but the ontology has no relationship for “recommended after replacement.” An evolution process can spot that gap, propose the relationship, test it and — once approved — make it available to the whole system.

Why governance still matters

What I appreciate most about Michael’s thinking is that it never trades trust for speed. Enterprise ontologies stay the governed foundation, the knowledge graph stays the connected representation of enterprise knowledge, and the inference engine stays responsible for deriving facts. The AI agent gains a second role: it doesn't just consume semantic knowledge, it also provides evidence of where the model can improve.

The outcome is not simply a larger knowledge graph, but a knowledge architecture that gets progressively better at representing what people, processes and AI systems need.

Shaping the document

My part was to give Michael’s ideas a form that invites reading. I set the paper up as a guided tour through seventeen short chapters, with key-idea callouts, diagrams for the architecture and the feedback loop, and comparison tables in place of dense paragraphs. The goal was a document that an information architect, a technical writer and a decision-maker can each read in their own way — and all come away with the same core message.

Read it yourself

The complete white paper is free to download. I hope it sparks as many ideas for you as the original conversation did for us.