Beyond better training data
AI systems need more than raw data. They need structured, semantically rich information. When concepts are captured in well-designed ontologies and connected in knowledge graphs, data stops being a raw ingredient and becomes a semantic map that both people and machines can navigate. Those same models that hallucinate on unstructured data behave far more deterministically when their retrieval pipelines draw on graph-based context.
But IA’s value extends beyond cleaner training data. A shared vocabulary expressed in ontology sharpens our own language. Teams formulate prompts and policies in precise terms, and AI receives signals that leave less room for ambiguity. This creates a feedback loop where better questions elicit clearer, more consistent answers.
Old expertise, new relevance
Time-tested methods like conceptual modeling, taxonomies, and metadata management are experiencing a renaissance through machine-readable ontologies and knowledge graphs. Research on knowledge-graph-augmented question answering demonstrates measurable gains in accuracy and stability when structured semantics guide retrieval and reasoning. In healthcare, ontologies like SNOMED CT power AI-driven diagnostic tools by providing standardised data that ensures precise medical insights.
The real breakthrough
We risk technology blindness once again, believing that sufficiently advanced AI will automatically solve everything. But the breakthrough doesn’t lie in a single technology. It lies in the composite of human expertise, proven IA practice, and modern AI techniques working as one.
The convergence of knowledge graphs with large language models is creating AI systems that can truly grasp subtle connections between concepts. Information architecture doesn’t just feed AI. It enables better human-AI communication and helps AI understand our intentions more accurately.
If we want AI that we can trust, we must first give it an information architecture that we can trust.
So, remember: No AI without IA. Information architecture isn’t just supporting. It’s essential for AI to deliver on its promises.
What are your experiences combining AI with information architecture?