Writing has expanded our ability to organize as groups by allowing us to transfer experiential content into an informationally stable medium. Whether etched into basalt or written in ink on paper, this recording process enables us to draw links between concepts, track notions that have faded from working memory, and perform operations that overcome some limitations of bare minds.
Writing can be understood as a process of encoding part of one's subjectivity: the content of an experiential stream. Reading then becomes a process of decoding that subjectivity. When the encoding is compact and vivid and the reader is attentive, the reader's interpretive frame can partially align with the writer's. Sasha Chapin describes books as “subjectivity-merging devices.”
Encoding and decoding both play a crucial role in cognitive development and the upkeep of cognitive acuity. Both require sustained attention, a faculty placed under growing pressure by attention-capturing technologies.
Design can work within existing technological habits to curb this decline in sustained attention, or use the mind's peripheral processing to guide attention toward the subject at hand. Joe Darko asks a related question in We Built Technology to Capture Our Attention. What If It Gave It Back?
This project investigates how ambient virtual content can ease people into long-form content written by a conscious subject, thereby helping to protect their attentional faculty.
One approach is a cognitive-mirroring utility. Such a system can model how minds relate and reorder concepts, move them into and out of working memory, and give the reader a spatial anchor for a process already underway, whether consciously or otherwise.
Concept Lattice extracts concepts from a text corpus and arranges them in an orderly grid-like structure: a lattice.
The arrangement uses axial movement—horizontal, vertical, and depth—which minds can track with relatively low spatial ambiguity. Each lattice cell has six immediate axial neighbors, bounding local relational complexity for recall at a glance and spatial reorganization. These constraints address the arbitrary placement and high node degree common to relational graphs and mind maps.
I have been working toward this implementation for three years. Earlier prototypes used syntactic parsers and categorial grammars, but those methods handled long context and ambiguity poorly.
Constrained LLM workflows have extended language processing across longer word horizons, supporting meaningful relation extraction across the breadth and irregularity of human language. Their integration into 3D engines such as Lens Studio through CLAD has also reduced the effort required to generate complex spatial arrangements and translate spatial transformation primitives into code.
Concept Lattice is therefore an LLM-native technological artifact. The demo below generates a lattice from The Origin and Development of Markets: A Business History Perspective by Mark Casson and John S. Lee.
In practice, a reader would open a book while wearing Spectacles and let the tool generate ambient content during reading. The content would remain relevant to the book's themes, provide a spatial reference for later review, and potentially support the reader's own notes.
The same approach could support AI interpretability, particularly in combination with Jacobian Lenses, by capturing and conveying language-processing primitives spatially and succinctly.
The prototype synchronizes a source-faithful e-reader with a persistent, camera-facing concept lattice. Reading-cursor progress selects source-grounded relations, materializes antecedent neighborhoods, and reorganizes concepts through whole-cell axial movement.
- Source grounding: every displayed relation retains its source statement, quote, and printed page.
- Cognitive mirroring: the lattice accumulates and reorders concepts as the reading position advances.
- Bounded locality: a rolling label window and six axial neighbors limit immediate visual complexity.
- Persistent space: concepts retain lattice cells, encounter order, recurrence state, and temporal depth across statements.
- Deterministic data: corpus, concept, coordination, reference, and audit assets can be regenerated from the retained source.
- Separate interpretation layers: source-derived semantic records remain distinct from spatial layout policy.
Open concept-lattice.esproj in Lens Studio 5.23.1 or a compatible later version, load Assets/Scene.scene, and start Preview with a Spectacles-compatible target. The authored scene connects the corpus assets to SourceEReader and MarketsConceptLatticeUI.
Dataset regeneration instructions are in Assets/Corpus/README.md.
| Path | Purpose |
|---|---|
Assets/Scripts/ |
Spatial reader and lattice runtime components |
Assets/Corpus/Runtime/ |
Lens-importable corpus, concepts, graph, and spatial grammar |
Assets/Corpus/Tools/ |
Deterministic corpus and graph builders |
Assets/Corpus/Audit/ |
Coverage, reference, and integrity checks |
Assets/Corpus/Source/ |
Source PDF retained for provenance and regeneration |
docs/ |
Architecture, submission, provenance, and demo media |
- Architecture
- Hackathon submission
- Provenance and third-party material
- Corpus data model
- Coordination spatial grammar
Concept Lattice is a hackathon prototype. Its semantic extractor is a rule-based interpretation of one paper and requires human evaluation before application to additional domains or high-stakes claims.