Loading...
Loading...
Designing idea-native discovery for public libraries
Core Proposition
Public libraries have invested heavily in digitisation and metadata infrastructure, yet discovery systems remain focused on documents rather than the ideas they contain. Idea-native discovery treats atomic ideas—concepts, themes, symbols, patterns—as first-class navigational objects, connected to source materials through provenance-aware links.
Public libraries have long served as stewards of recorded knowledge, developing sophisticated systems for the acquisition, description, preservation, and circulation of information resources. Cataloguing standards, metadata schemas, and discovery platforms have been refined over decades to support reliable access to diverse collections at scale.
These systems are highly effective at managing containers—books, journals, archival records, audiovisual materials, and digital objects—and at enabling users to retrieve known items or locate materials matching specific keywords, authors, or subjects.
But there’s a growing gap between how these systems organise knowledge and how people actually seek to understand it.
A student exploring “the commons” doesn’t necessarily want a ranked list of books. They want to follow an idea—to see how it connects to collective action problems, to Ostrom’s governance work, to digital commons debates, to enclosure movements in 16th-century England. They want to move through meaning, not through containers.
Container-centric discovery systems tend to fragment meaning across individual records. Even when enriched with subject headings, abstracts, or controlled vocabularies, they require users to infer conceptual relationships by moving manually between items, synthesising information outside the system itself. Questions such as “What ideas connect these texts?”, “How has this concept evolved over time?”, or “What other works explore a similar theme?” are difficult to answer within discovery systems that treat documents as the primary unit of access.
The paper introduces idea-native discovery as a complementary paradigm for library discovery systems—one that treats ideas, rather than documents, as first-class navigational objects. Rather than replacing existing catalogues or metadata infrastructures, idea-native discovery is proposed as an additional interpretive layer that enables users to explore the conceptual structure of collections while preserving the integrity and authority of traditional systems.
Definition
A discovery paradigm in which atomic ideas—such as concepts, themes, symbols, arguments, or patterns—are explicitly represented as discrete objects, with typed relationships linking them to each other and to source materials, enabling users to navigate collections at the level of meaning rather than solely at the level of documents.
Ideas in this model may be categorised into multiple types: concepts (abstract notions or constructs), themes (recurring topics across works), symbols (objects or motifs with layered meaning), and patterns (structural or argumentative regularities). Equally important are the typed relationships between ideas—embodies, contrasts with, evolves into, parallels, or is symbolised by—reflecting how meaning is constructed across texts.
Documents as primary units of access vs. ideas and their relationships as first-class navigational objects.
Key Transformation: Users navigate between ideas rather than documents, with source materials accessible as provenance links rather than primary discovery targets.
The paper is careful to distinguish idea-native discovery from several adjacent approaches:
Search presupposes users know the terms they’re looking for. Idea-native navigation supports exploration when users can’t yet articulate what they’re seeking.
The system doesn’t produce AI-generated text that substitutes for original works. It provides structured representations of ideas and relationships, always grounded in and linked back to identifiable sources.
Most library knowledge graph work has focused on machine interoperability, not human-facing navigation. The design principles here are specifically about making conceptual structure legible to non-expert public audiences.
Critically, idea-native discovery is also not a replacement for catalogues, finding aids, or metadata standards. Existing systems continue to perform essential functions related to acquisition, authority control, preservation, and access. Idea-native discovery operates as an additional layer that draws upon these systems while offering a different mode of engagement.
The paper proposes four principles for translating the idea-native paradigm into usable interfaces for non-expert public audiences:
Principle 1
Instead of overwhelming network visualisations, the interface shows a small constellation of closely related ideas around a focal node. Users move stepwise from one neighbourhood to the next—conceptual wayfinding without needing a mental model of the whole graph.
Principle 2
Every node and every relationship is typed and labelled. An idea is clearly marked as a concept, theme, symbol, or pattern. A relationship says “contrasts with” or “evolves into”—not just “is related to.” This makes the system’s structure visible and auditable rather than mysterious.
Principle 3
Initial views present minimal information—a focal idea and its immediate connections. Users choose when to go deeper, revealing extended relationships, contextual notes, or source links. Complexity is available but not imposed.
Principle 4
As users move from idea to idea, they build a traceable path—an informal learning trajectory. Because each step is user-initiated and semantically labelled, the logic of exploration remains transparent.
A focal idea node with bounded, semantically typed related ideas and user-controlled expansion paths.
Relationship Types
+24 more related ideas
Focal Idea
Supporting Evidence
Extension
Counter-Argument
Application
Related Theory
Historical Context
The currently selected idea, with full context and source attribution.
Semantically labeled connections enabling structured exploration.
User-controlled depth limiting prevents cognitive overload.
Libraries face a dual pressure: public expectations for AI-powered discovery are rising (driven by tools like ChatGPT), but the values that underpin public libraries—transparency, provenance, accountability, copyright compliance—make it difficult to simply adopt opaque generative systems.
Idea-native discovery threads this needle through several architectural choices:
Copyright-safe by design
The idea-layer operates at the level of non-copyrightable abstraction—concepts, themes, and relationships rather than reproduced text. Every idea links back to authoritative sources held or licensed by the library.
Provenance-first
Every idea and relationship is linked to identifiable source materials or curatorial decisions. Users can see not only what is connected, but why, and where those connections originate. This traceability enables both public understanding and internal audit.
No black boxes
AI may assist in identifying candidate ideas and relationships, but under human curatorial oversight. Results remain inspectable and governable. The system does not present inferred connections without explanation.
Incremental adoption
Libraries can pilot idea-native discovery with focused collections or themes, evaluating impact through qualitative engagement measures rather than high-stakes system-wide changes. The idea-layer can be modified or withdrawn without disrupting underlying catalogues.
This positions idea-native discovery as an institutionally compatible innovation rather than a disruptive technology—something libraries can adopt incrementally, beginning with pilots or thematic initiatives, and expand based on evidence and institutional readiness.
The paper outlines a modular reference architecture built on three architectural principles:
The idea layer stores structured representations and provenance links, never full texts or copyrighted content. Existing catalogues remain the systems of record.
Every idea and relationship traces back to specific works, chapters, or collections, preserving attribution and enabling audit.
The idea layer connects to existing discovery platforms through deep links, enriching current item-level views with related ideas rather than competing with them.
Idea-native discovery has implications that extend beyond interface design or technical architecture. By enabling navigation at the level of meaning, it reframes how public libraries position themselves within contemporary information ecosystems.
In an environment of information abundance and algorithmic mediation, a third role has become increasingly salient: that of sense-making institution. Rather than competing with commercial search engines on speed or scale, libraries can differentiate by offering environments that support understanding, reflection, and learning.
Path-based navigation, time spent exploring conceptual neighbourhoods, and repeated traversal across themes provide indicators of interpretive engagement rather than mere access. These signals offer a richer picture of how users interact with collections.
Rather than positioning AI as an autonomous interpretive agent, idea-native systems constrain its role to supporting the identification and organisation of ideas under human oversight. This functions as a mediating layer between traditional library systems and emerging AI capabilities, grounding innovation in explicit, inspectable structures.
Because ideas are represented as abstract entities that can carry multiple labels across languages, the model supports cross-lingual conceptual navigation without requiring a single global ontology. Conceptual relationships can be curated locally while remaining interoperable across linguistic contexts.
Idea-native discovery has been conceptually possible for years. What makes it practically buildable now is the same technology that appears to threaten libraries: large language models and AI-assisted knowledge extraction. The irony is that AI doesn’t replace the need for this system—it’s what makes it affordable to create.
Extraction that was impossible is now cheap
Identifying the core ideas in a book, the themes it explores, and the relationships between concepts used to require a subject specialist spending hours per item. AI can now produce a first-pass idea extraction in seconds—not perfect, but good enough to create a draft graph that humans can refine. The cost drops from “impossibly expensive at 4 million items” to “feasible as a pilot.”
Cheaper than the alternatives
Rebuilding an integrated library system costs millions. Licensing a commercial knowledge base creates dependency. Building an idea-native layer on top of existing infrastructure is comparatively modest—a graph database, an extraction pipeline, and an interface layer. It doesn’t replace what libraries already have; it adds a new way to navigate it.
Librarians gain a new role, not lose one
The concern that AI replaces librarians misreads the situation. AI can extract candidate ideas and relationships. It cannot determine which interpretive frame is appropriate for a community, resolve contested meanings, or make curatorial judgments about what matters. Librarians shift from cataloguing containers to curating meaning—a more intellectually demanding role, not a diminished one. Their expertise in classification, subject analysis, and knowledge organisation becomes more valuable in an idea-native system, not less.
The competitive landscape demands it
Users already navigate ideas through ChatGPT, Perplexity, and Google’s AI overviews—but without provenance, without curatorial authority, and without institutional accountability. Libraries can offer something these tools cannot: idea-level navigation that is transparent, attributable, and grounded in verified collections. The question isn’t whether idea-native discovery will happen, but whether libraries will be the ones to do it well.
The State Library of Victoria holds over four million items. A reasonable estimate is that fewer than 1% receive regular engagement. This isn’t a failure of the collection—it’s a failure of discovery architecture. The current system is structurally biased toward materials that are already known: searched by title, recommended by name, or surfaced through popularity metrics.
A first-time author writes a novel exploring intergenerational trauma in migrant communities. In a container-centric system, it surfaces only if someone searches for the author’s name (unknown), the exact title (unfamiliar), or a subject heading broad enough to return hundreds of results. In an idea-native system, it surfaces because it shares thematic connections with Viet Thanh Nguyen, with oral history collections, with psychological research on displacement—connections that exist in meaning, not in metadata.
This is the core practical argument for idea-native discovery: it makes the long tail navigable. When ideas rather than containers are the unit of access, discoverability decouples from fame. A book doesn’t need a marketing budget, a prize shortlisting, or an established author name to be found—it needs to be about something that connects to what a reader is exploring.
A tiny fraction of holdings receive most engagement. Idea-native discovery makes the long tail navigable.
~1%
Regularly accessed
~99%
Good but undiscovered
Idea-native discovery makes the long tail navigable by connecting items through meaning rather than popularity. A book doesn’t need fame to be found—it needs to be about something a reader is exploring.
Unknown authors surface alongside established ones
The idea graph has no concept of fame, sales figures, or publisher size. A self-published exploration of “the commons” sits in the same neighbourhood as Ostrom, Bollier, and Hess—discoverable by anyone following that conceptual thread. Merit of ideas replaces marketing reach as the discovery mechanism.
Cross-format bridging
A poetry collection about displacement connects to academic migration studies, oral history archives, documentary films, and novels—materials that would never appear in the same keyword search but share deep thematic resonance. Formats stop being silos.
Digital shelf-browsing at collection scale
Physical browsing creates serendipitous encounters—but only across the ~200 items visible on a shelf walk. Idea-native navigation recreates that serendipity across millions of items, surfacing unexpected connections that physical proximity never could.
Collection value becomes demonstrable
Libraries can show that deep holdings—the 3.96 million items rarely accessed—are conceptually connected to trending ideas, active research, and current public discourse. The collection’s value is revealed through its intellectual structure, not just its circulation statistics.
Reader-contributed discovery
If readers can annotate what a book is really about—not its official subject headings but its lived thematic content—the idea graph becomes a crowdsourced discovery layer. A reader who found unexpected connections between a gardening memoir and grief literature creates a navigational path that benefits every future explorer.
Author self-submission
New and independent authors could submit structured idea-mappings for their own works—declaring what concepts their book engages with, what it contrasts with, what conversations it joins. This creates discovery metadata far richer than keywords, and it doesn’t require a publisher’s marketing department to produce.
Curated idea trails
Librarians or algorithms can create thematic reading paths—“The Architecture of Trust,” “How Cities Remember”—that naturally weave lesser-known works alongside canonical ones. The trail format means an unknown author isn’t buried in a results list; they’re a waypoint on a journey the reader has already committed to following.
The fundamental shift is this: container-centric discovery asks “Do you know what you’re looking for?” and rewards the already-known. Idea-native discovery asks “What are you thinking about?” and surfaces everything relevant, regardless of the author’s visibility. For the millions of books that are good but unknown, that architectural difference is the difference between permanent obscurity and being found.
There’s a deeper structural problem that idea-native discovery is uniquely positioned to solve: the same word operates in entirely different conceptual neighbourhoods depending on the plane of discourse. Current systems either return all meanings indiscriminately (keyword search) or force you to pick one (subject headings). Neither reflects how knowledge actually works.
The same word occupies entirely different conceptual neighbourhoods depending on the plane of discourse. Keyword search flattens all meanings together. Subject headings force one. Multi-plane navigation lets them coexist.
Same word
“cell”
Biology Plane
Criminal Justice Plane
Telecoms Plane
Security Studies Plane
Keyword Search
Returns all meanings indiscriminately. 4 worlds mixed into one list.
Subject Headings
Forces you to pick one. Can’t see that the word lives in multiple worlds.
Multi-Plane
All meanings coexist. Context determines which neighbourhood you see.
Planes are separate but traversable. A reader can see that “cell” lives in multiple worlds and choose to cross over—discovering that biology’s cell division metaphor appears in organisational theory, or that prison cell architecture echoes monastic design.
A keyword search returns all of these indiscriminately. A subject heading forces you to choose one. But an idea-native system with multiple interpretive planes lets the same term exist in parallel conceptual neighbourhoods, each with its own relationship structure. The user’s navigational context—which plane they’re exploring—determines which neighbourhood they see.
This addresses the obvious concern about cultural interpretation head-on. Instead of one authoritative graph that pretends ideas have fixed, universal meanings, multi-plane discovery acknowledges that meaning is contextual. The same concept should have different neighbourhoods in different interpretive frames. That’s not a flaw to be resolved—it’s the structure of how knowledge actually works.
Cross-disciplinary surprise
A reader exploring “resilience” on the ecology plane encounters the concept on the psychology plane and realises the same structural pattern applies to ecosystems and trauma recovery. The planes are separate but traversable—you can see that a word lives in multiple worlds, and choose to cross over.
Fiction and non-fiction share conceptual space without collapsing into each other
“Justice” in a legal philosophy text and “justice” in a novel about wrongful conviction operate in different planes but can be connected through explicit cross-plane relationships. A reader following justice-as-concept in fiction can see that it connects to justice-as-theory in philosophy—without the two being flattened into a single undifferentiated results list.
Cultural and linguistic pluralism built in
Different cultural communities can maintain their own interpretive planes for contested concepts. “Land” in an Indigenous Australian context has a fundamentally different conceptual neighbourhood than “land” in property law. Multi-plane architecture doesn’t force a single interpretation—it lets multiple valid framings coexist, each with their own internal coherence.
The “who decides” problem becomes tractable
Rather than one authoritative graph requiring universal agreement, different communities, disciplines, or curators maintain their own planes. A children’s literature librarian curates one plane; a legal studies researcher curates another. Disagreements become visible structural features—two planes with different relationship maps for the same concept—rather than hidden editorial choices.
The strongest objection to idea-native discovery—that ideas are culturally situated and contested—becomes, with multi-plane architecture, the system’s most distinctive feature. The goal isn’t one true graph. It’s a system that makes the plurality of meaning navigable.
As libraries navigate the challenges and opportunities presented by artificial intelligence and digital abundance, they face a choice not between preservation and innovation, but between different modes of engagement. Idea-native discovery offers a path forward that honours libraries’ enduring commitments while extending their capacity to steward meaning in a rapidly changing information landscape.