Understanding

Understanding concerns the semantic interpretation of information: whether an AI system can form a coherent account of a cultural organisation, its activities and its significance. Accessibility asks, “Can the information be reached?” Understanding asks, “Can the information be interpreted correctly?” Retrieving an accurate passage does not, by itself, establish that its meaning will be preserved.

Generative AI systems construct meaning in their responses by combining learned patterns with the context available to them, rather than simply reproducing text. Here, understanding describes the adequacy of that interpretation, not human comprehension. A museum, its building, its collection and a temporary exhibition are connected, but they are not interchangeable.

In the AURAS-O Framework, this dimension therefore concerns clear identity, consistent descriptions and explicit relationships. Names, roles, dates and institutional context help distinguish what belongs together and what should remain separate. The aim is an account that retains cultural meaning, including what an institution represents, rather than an assembly of individually correct but disconnected facts.

Research on knowledge representation and generative search offers complementary perspectives on this distinction. Representing an institution requires more than collecting statements: identity, relationships and context determine how those statements fit together. Research on knowledge graphs provides a foundation for making these connections explicit, including distinguishing entities that share a name and describing relationships between them.^[1] This does not mean that every generative AI system consults a knowledge graph, or that structured information guarantees a correct answer.

Generative-search studies also indicate that retrieved material is used selectively. Experiments examining content preferences suggest that patterns of selection can be identified and vary across domains.^[2] Work incorporating search intent further suggests that the relationship between a passage and the reader’s information needs matters for its use in an answer.^[3] Together, these findings place interpretation within a particular question and source context, rather than treating visibility as a property of an isolated page.

These studies examine representation methods or visibility within experimental search settings; they do not directly establish whether systems understand a cultural institution’s mission. AURAS-O applies their insights to that question without equating source selection with accurate interpretation. Collectively, current research suggests that coherent identity, explicit relationships and relevant context provide a useful basis for examining how AI systems represent cultural organisations, while the accuracy of those representations still requires assessment.

For a museum, naming an artist alongside an object is only a starting point. A collection record can distinguish the artist from a previous owner, explain provenance and identify whether the work belongs to the permanent collection or a temporary exhibition. These relationships give facts their institutional and historical meaning.

In performing arts, a theatre can distinguish a dramatic work from a particular production, linking its director, cast and performance dates to that production. Repertoire pages can make clear which productions are current and which belong to the archive. A festival can similarly connect artists and themes to a named edition, so that different years do not become one undifferentiated programme.

A botanical garden can explain how its living collections relate to habitats, conservation priorities and educational programmes. This context helps describe its purpose beyond its role as a visitor destination. Across these examples, the practical task is to make relationships intelligible in ordinary language and consistent records. These are applications of the framework, not outcomes demonstrated by the cited studies; generated descriptions should still be checked against the institution’s own account.