AI: Summary
This session brought together the Future Text Lab community to examine the cognitive science underpinning spatial text interfaces and to develop a practical design philosophy for combining traditional framed displays with XR nodal environments — specifically through Apple Vision Pro and the in-development Author application. The discussion moved from reading science and dual visual processing through to hands-on questions of what participants would actually use such a spatial knowledge environment for, probing the gap between theoretical excitement and personal necessity.
AI: Main Topic
Frode Hegland, joining from Shanghai, introduced a draft paper prepared for a contact at a major technology company (“Mike”), which he had developed in dialogue with Claude (see AI: Highlights). The paper’s central concern is reconciling the cognitive benefits of familiar, stable framed displays with the spatial freedom of nodal XR environments. His key finding, drawn from research into rapid visual threat responses originally studied in policing contexts, is that humans assess a visual layout and judge it as threatening or safe before reading a single word. This creates the fundamental design tension for the group’s work: how to introduce spatial novelty without triggering rejection. The paper also examines optimal screen sizes for reading (derived from foveal angular constraints rather than absolute measurements), the difference between reading and scrolling as distinct cognitive modes, and the concept of “text for thought” — writing not to communicate outward but to understand inward.
AI: Highlights
Frode Hegland directly credited Claude (Assistant) as a substantive research collaborator in preparing the paper, noting it was valuable precisely because it was “not just being sycophantic” — it questioned claims, identified errors, and conducted proper research into visual cognition and reading science. This is presented here as addressed to Assistant.
In the chat, Frode Hegland shared a prompt directed at Claude/Assistant with the following structure: first, summarize the text to confirm whether the intended communication is actually landing; then, under the heading “Spark,” use knowledge of who Frode is and what the text says to illuminate surprising and non-obvious aspects, and potentially connect ideas to Marshall McLuhan or other philosophers. This prompt is noted here as being explicitly addressed to Assistant.
Peter Dimitrios flagged the distinction between authored and generated links as critically important in any AI-augmented knowledge system — that users must be able to see which connections were made by human intention versus those generated automatically.
Tom Haymes shared a citation from Lee et al. (2025) in the chat: that high task confidence enables better delegation to AIwith maintained accountability, while lower self-confidence may cause over-reliance on AI and diminish independent problem-solving.
AI: Insights
Pre-reading visual threat response and the novelty problem
Humans judge a page’s typographic style and layout before reading a word, and unfamiliarity triggers a threat response in the faster of the two visual processing pathways Brandel Zachernuk described as governed by the foveal region. This is not merely aesthetic conservatism — it is a hard cognitive constraint. It means that any genuinely novel spatial text interface will face a neurological barrier to adoption that no amount of instructional design can fully dissolve.
Reading optima are angular, not absolute — and this matters for XR
The “27-inch monitor at arm’s length” reading optimum is not an arbitrary industry convention but is derivable from foveal region size and the range of comfortable horizontal eye motion, as Brandel Zachernuk noted drawing on the work in Reading in the Brain. Because the measure is angular, it translates directly to XR environments where virtual display size is decoupled from physical size. This reframes the Vision Pro not as a compromise device but as one that can precisely replicate the optimal reading geometry regardless of physical space.
The framed/nodal binary is already dissolved on large monitors
Frode Hegland observed that anyone working on a large display is already working with nodes — each document is functionally a node that can be moved, but without meaningful inter-document linking. The distinctive XR challenge is not replacing frames with nodes but making the connections between nodes explicit, persistent, and semantically navigable. The insight reframes XR as an extension of existing practice rather than a rupture.
Summoning a node into full-display mode as a key design primitive
The session converged on a specific design concept: in XR space, any node should be summonable via gesture or command into a full-monitor-equivalent mode, with all the stability, quality, and information richness of a traditional display, then dismissible back into the spatial field. This is distinct from merely resizing or opening a node — it is a mode change that temporarily grants a spatial object the ergonomic properties of a framed display.
Scrolling and reading are cognitively distinct and interfaces should respond differently
Frode Hegland articulated a realization that when a user scrolls, they are communicating “I do not want to be here” — yet most interfaces continue displaying text in the same register as during focused reading. Color-coded backgrounds, keyword highlighting, and substitution of proper nouns with brand icons or visual shapes could activate only during scrolling, providing navigational affordances without cluttering the focused reading experience.
Configurable multi-view document representation is an almost entirely unsolved problem
Brandel Zachernuk identified that virtually no text-centric application currently allows a user to construct genuinely different views over the same document — where the same element might appear green in one representation and blue in another, or highlighted in one view and invisible in another. The web offers a partial model (one HTML file, multiple CSSdeclarations), but no writing tool has meaningfully pursued this. The absence seems cultural and commercial rather than technical, since legal document workflows (track changes, shared strikethroughs) have driven interface conservatism. Jonathan Finn reinforced this, noting that truly radical divergence between views — where two views of the same document could look like entirely different documents — would have enormous use cases that have simply not been pursued.
Computers demand consistency; humans remember through novelty — a structural disconnect
Tom Haymes articulated a tension revealed by his library digitization project: when photographing bookshelves to feed into a knowledge navigator, he found that AI parsing required everything to be made consistent and categorically uniform, while human memory of physical books is anchored in their distinctiveness — color, font, shape. The physical book’s capacity to be remembered is precisely what digital systems systematically destroy in the act of ingestion.
AI as a keyhole, not a window
Frode Hegland offered the metaphor that AI is a keyhole: you pass a thread through it and pull back what you ask for from the vast structure on the other side, and it will sometimes offer things you did not request. But you have no access to the structure itself. This frames the core limitation of AI-as-interface and explains why the group’s work on spatial, structural, and visual display of knowledge is not superseded by AI but becomes more important alongside it.
Appendices and endnotes as a design model for personal XR knowledge space
The group arrived at the insight that the academic conventions of appendices and endnotes — content that exists, that is connected, but that is not prominent — are a natural structural model for personal knowledge nodes in XR. Rather than treating all nodes as equal, a spatial knowledge environment should allow any node to carry subordinate layers of reasoning, side explorations, and sourcing that are present but retracted, accessible but not foregrounded.
Personal annotation breaks when links point to online rather than local versions
Frode Hegland noted a structural problem: links in documents typically point to online source versions, not to locally annotated copies. For knowledge work, the annotated, marked-up local version is the epistemically relevant one — it represents the document-as-understood rather than the document-as-published. XR environments offer an opportunity to build linking practices around local ownership of documents.
Attention trails — browser history, AI dialog, reading history — as latent knowledge structure
Brandel Zachernuk revisited a 2007 Firefox history plugin he built that constructed a hierarchical tree of browsing activity including tab switches and tab origins. The insight was that once such a trail exists, rich relational inferences become possible — what kinds of sessions you have, what clusters of attention emerge. Frode Hegland extended this: AI conversations now leave analogous trails but without accessible structure. Integrating browsing history, AI dialog history, and reading/writing history into a navigable temporal knowledge layer was identified as a significant opportunity.
Bibliometric visualization as a spatial knowledge interface
Peter Wasilko proposed a detailed vision for XR-native bibliometric display: author clouds with directional citation links (one-way vs. mutual citation represented by link color and direction), topic-level influence networks, and transitive closure views showing which foundational sources have propagated through many citation chains into a current work. The intensity of a source’s visual representation would reflect how many downstream pathways it has seeded — making invisible intellectual lineage spatially legible.
XR as a writing/authoring environment rather than a note-taking space
Frode Hegland proposed a paradigm reframe: rather than conceiving of the XR environment as a place to store notes or manage a “second brain,” it might be more productive to think of it as a place where you actively author coherent articles for yourself. Tom Haymes confirmed he already does something analogous using NotebookLM — dumping all his writing into a single notebook and querying it — but noted this is textual, not spatial or visual.
Tactile illusion in XR is already working through audiovisual cueing
Both Tess Rafferty (on the Vision Pro butterfly demo) and Peter Dimitrios (on unpublished research he had briefly encountered) noted that the brain can be made to register tactile sensation through synchronized visual and audio cues, without any physical contact. Tess Rafferty reported that the visceral response to the Vision Pro‘s opening dinosaur sequence — including the urge to look away from something threatening — confirms that spatial presence and emotional response are already deeply engaged by current hardware, even without physical haptics.
Song
AI: Resources Mentioned
Books
Reading in the Brain — by French neuroscientist Stanislas Dehaene; mentioned by Brandel Zachernuk for foveal reading science
Behave — by Robert Sapolsky; mentioned by Frode Hegland as the source for the dual-visual-processing research thread, with a note that Claude/Assistant initially (incorrectly) associated it with Kahneman’s System 1/System 2 framework
Ministry for the Future — by Kim Stanley Robinson; mentioned by Brandel Zachernuk as an example of a book whose fractured aesthetic structure could benefit from typographic differentiation
CSS Zen Garden — website and book demonstrating multiple CSS declarations over a single HTML file; mentioned by Frode Hegland
Websites and Tools
Tapestry — https://tapestries.media/dashboard/public — mentioned by Peter Dimitrios as a tool for organizing and visualizing linked notes
Peter Dimitrios’s Federated Wiki on finding truth in the age of AI — http://petedaguru.dojo.fed.wiki/view/welcome-visitors/view/finding-truth-in-the-age-of-ai — mentioned by Peter Dimitrios
Brace style article (K&R vs Allman) — https://medium.com/@8acking/brace-style-clash-k-r-vs-allman-which-ones-better-5bbc0521aa58 — shared by Brandel Zachernuk
Future Text Lab podcasts — https://futuretextlab.info — mentioned by Frode Hegland
People
Doug Engelbart — mentioned by multiple participants as a lens for evaluating knowledge work tools; Tom Haymesproposed constructing an AI persona from his writings
Ted Nelson — mentioned by Tom Haymes alongside Engelbart as a philosophical filter for knowledge organization
Mike Caulfield — mentioned by Peter Dimitrios as a key figure in SIFT methodology for knowledge verification
Marshall McLuhan — mentioned by Frode Hegland in the chat prompt directed at Claude/Assistant as a possible philosophical lens
Software and Hardware
Author (Mac and Vision Pro) — Frode Hegland‘s application, currently in active development
Apple Vision Pro — primary XR device under discussion
Adobe Illustrator, Blender, Final Draft, Ulysses, Keynote, NotebookLM, Gemini — mentioned by various participants as current tools in their workflows
Methodologies
SIFT (Stop, Investigate, Find better coverage, Trace claims) and CRAAP test — mentioned by Peter Dimitrios as frameworks for knowledge verification that VR environments might help apply
Academic Reference
Lee et al. (2025) — cited in chat by Tom Haymes; on the relationship between task confidence, AI delegation, and critical engagement