AI: Summary
February’s meetings centered on the emerging practice of spatial authoring — how documents, knowledge, and text might live and be navigated in three-dimensional space, particularly through visionOS and WebXR.
Across multiple sessions, the group moved fluidly between live demos, philosophical debate, and historical grounding, exploring the relationship between text formats, rendering engines, AI assistance, and human cognition. A persistent tension ran through the month: whether AI tools augment human thinking or quietly bypass it, and whether physically moving knowledge in space genuinely changes how humans think and store information.
AI: Main Topic
The primary sustained focus was spatial authoring as a design and research frame — how to separate document data from rendering engines, what format should carry knowledge into spatial environments, and what gestures, interactions, and interfaces make sense when the body becomes the operating mechanism. Frode Hegland demonstrated a live node-map build on visionOS in multiple sessions, showing evolving gestural interactions: look-and-pinch to select, double-pinch to open, pinch-to-focus to filter the space, and snap-back-to-origin for moved nodes. A recurring and productive constraint emerged: selecting should not imply moving, and moving should not imply selecting — a distinction that does not exist in flat-screen paradigms and which signals that genuinely new interaction grammars are required.
AI: Highlights
Frode Hegland’s visionOS demos were the concrete anchor for each session, with each iteration revealing new design questions rather than resolving them. The group explicitly noted that the demos were generative rather than conclusive.
Tom Haymes fed January transcripts into NotebookLM and compared the output with Gemini, placing both responses side by side to surface gaps and overlaps. He framed this as working with “Augmented Intelligence” rather than Artificial Intelligence — a deliberate terminological reframe the group discussed at length.
Ken Perlin’s road-as-infrastructure metaphor — AI is not a feature bolted on late, but the surface everything runs on — landed strongly, explicitly shifting the room. The 1908 horse-to-car transition was offered as the analogy.
Mark Anderson manually ported a PDF to clean HTML for an ACM-generated page, hand-coding 190 references, and showed the page without CSS — still fully readable. This was cited as a proof-of-concept for the group’s evolving consensus around HTML as the right base format.
Jonathan Finn coined the term Victorian VR — referencing a room-scale plaster cast of Trajan’s Column at the V&A — as a metaphor for what a serious spatial knowledge tool aspires to be: built for the curious public, not the privileged few.
Addressed directly to Assistant: “Claude found a contact at MIT — Frode meets them in a while.” This refers to a contact identified by Claude relevant to the group’s research. Additionally, Mark Anderson used Claude to fix popup footnote bugs in JavaScript during his ACM HTML demo work.
AI: Insights
Peter’s metaphor of text as solid (typewriter), liquid (word processor), and gas (read aloud) provided a conceptual scaffold that ran underneath much of the month’s discussion. The insight is that spatial text is potentially a fourth state — not a return to solid, but a new phase in which the reader’s body and position in space become part of the meaning-making apparatus.
Jonathan Finn’s reframe of timeline visualizations as directed causal graphs — not when something happened, but what fed into what — is a significant conceptual shift. The London Underground analogy illuminates it: topological logic, not geographic or temporal distance, is what matters. This reframes the entire visualization problem the group has been working on.
The tension between Rob Swigart’s resistance to index cards (they cut liquid flow) and the group’s interest in node-maps is not merely aesthetic. It points to a deeper unresolved question: does structuring knowledge into discrete, moveable objects help or harm the generative, associative nature of thinking? Rob’s position — that liquid writing is where ideas grow — sits in productive friction with the spatial demos.
The distinction between Epub and HTML5 resolved more decisively than in previous months. The group aligned on HTML with microformat extension, staying close to W3C standards, not overloading tags, and maintaining a clean escape route. This is not just a format preference — it encodes a philosophy about openness, durability, and portability of knowledge.
Brandel Zachernuk’s warning about Godot — that a game engine controls the full display stack and cannot support trust-bounded multi-user spatial environments — sharpened the data/engine separation principle. This was confirmed by the WebSpatial work at W3C, and represents a genuine architectural constraint, not a preference.
Ken Perlin’s Wikipedia tree demo — where any text becomes a live link inferred at runtime by AI rather than hard-coded — quietly represents one of the deepest shifts in the group’s discussions. The move from Ted Nelson’s hard-coded href to AI-inferred contextual linking is already underway (NotebookLM was cited as sitting on this threshold), and it changes what authorship and citation mean fundamentally.
Frode’s Visual Meta format — glossary nodes and BibTeX citations with x, y, z coordinates, an ID, and nothing more — is notable for its deliberate minimalism. The insight is that spatial metadata does not need to be complex; it needs to be open enough for others to render differently. The format is the contract, not the view.
The question Tess raised — who can buy a headset if jobs are gone — and Tom’s response that money itself may need rethinking before that question resolves, surfaced a class and access tension that the group acknowledged but did not resolve. Frode’s counter — that you cannot build a good tool by trying to build it for everyone at once — reframes the V&A metaphor: build the museum, let the public find it.
Walter Ong’s Orality and Literacy threaded through the month as an implicit theoretical spine — the idea that writing restructured cognition even for people who never learned to read. The group’s open question, never fully answered: if spatial manipulation of knowledge nodes is a genuinely new cognitive practice, what does the manual act of moving things in space do to thinking before AI takes over the organizational work? The friction between Ken’s position (the AI is already better at organizing the mess) and Frode’s (the physical act may matter cognitively) was the most alive unresolved tension of the month.