AI: Summary
This session deepened the group’s figure-and-ground thread into a genuine reframing: the difficulty of showing relationships in a 360-degree knowledge space is not a problem of where to route a connective line, but a problem of signification — of how out-of-view information can keep doing the work that knowledge-in-the-world is supposed to do.
Drawing on Don Norman’s affordances and signifiers, Donella Meadows’ systems thinking, and the founding choices of hypertext, the conversation worked through what a node should reveal when closed, how to distinguish an author’s corpus from collaborations and shared topics, the multisensory and dispositional dimensions of perception, and the constraints that XR swaps in rather than removes. Threaded throughout was a reflection on human–machine co-evolution, sharpened by Mazlish’s fourth discontinuity, alongside a running live demonstration of one AI briefing another to render the session’s own song and quotations into image and video.
The session’s conceptual centerpiece was the reframing of the connective-line problem. The group first reached for routing fixes — an invisible sphere around the head inside which lines simply stop rendering, or interactive “yo-yo” pulling of distant nodes inward — but the deeper move was to recognize, through Norman’s affordance/signifier distinction, that a line is doing two jobs at once. It is content, because the fact that one concept relates to another is real knowledge, and it is a signifier, because it announces “a related node is over there, perhaps behind you.” The intrusiveness is the price of forcing the content-form to carry the signpost all the way across the room. The resolution: do not draw the relationship, signify it — surface a cue that a relationship exists and roughly where it points, and render the actual line only when the reader turns toward it. Because their connections are already born from one node’s definition mentioning another, the inline mention can do the announcing.
This was reinforced by Norman’s second principle, knowledge in the world versus knowledge in the head, and the realization that “behind you” quietly breaks it. Putting knowledge into the world frees us from remembering it, but only while it stays perceptible at the moment we need it. Information at our back is technically in the world, yet out of view it reverts to knowledge in the head — we now have to remember it is there. The real problem is therefore how to keep out-of-view information doing its job, and the answer is persistent peripheral signifiers: ambient cues that survive when the content itself leaves the field of view, so the room keeps reminding us rather than making us carry it in memory. Edwin Hutchins’ Cognition in the Wild and the distributed-cognition lineage sit directly beneath this.
A historical connection crystallized: the current line-rendering question recapitulates the original split between Ted Nelson’s Xanadu — explicit, content-bearing links shown by default, with two-way connections — and the web’s minimal “just enough” choice, where a link is signified by mere highlighting. CERN implemented the simplest possible thing; Nelson wanted the lines. The same affordance-versus-signifier tension the group is now wrestling with in three dimensions was effectively decided, largely by default, at the web’s birth — a lineage half-jokingly traced to Bill Atkinson and Jony Ive.
Sharpening the progressive-disclosure thread, the question became what a node must show when it is not selected or open — its person-ness, place-ness, or time-ness, or simply that it is a closed note — rather than its full readable content. Selection, not approach, would surface meaning. Supporting intuitions included faded colour or tone to signal dormancy, and the idea that small text at a distance can collapse into something merely shaped like text, “chicken scratchings” that carry presence and silhouette without legibility. Readability by default was recognized as the wrong default.
The earlier author-versus-idea distinction resolved into three separate relationship types a system should be able to toggle between: an author’s own body of work, an author’s collaborations and co-authorships, and the looser set of people writing on the same topic. Each implies a different traversal even when the visual form looks similar. Beneath this ran a desire for any subset of text to be selectable and linkable at sub-document granularity, rather than the coarse whole-page URL the web settled for.
XR was reframed as exchanging constraints rather than removing them. Freedom from gravity, structure, and physical form is real, but it is replaced by perceptual limits, the seven-plus-or-minus-two ceiling of working memory, rendering budgets, and the AR demand that the messy real world stay visible behind the text. Grounded in Donella Meadows’ systems thinking, the group held that the failure mode is not the absence of constraints but the failure to understand the new system’s constraints.
Perception was taken to be fundamentally multisensory, with synesthesia — via Richard Cytowic — as proof that the senses “unionize” rather than process discretely, and with the noted irony of capturing a multisensory subject inside a single-medium book. Static visual text need not be the precedent for spatial interaction; smell, weight, and other non-linear channels become legitimate questions rather than novelties.
A quieter but pointed thread asked whether a node carries a disposition — a tendency or stance — that should itself be perceptible, and whether one can “digitize hunches.” This pushes representation past explicit content toward the tacit qualities that actually shape how we navigate knowledge.
One AI briefing another was observed live. A song generated for the day’s theme was handed to an image-and-video system with no human-specified content, yielding “field” imagery the group recognized as a metaphor for their own thinking; a quotation was likewise turned into a video, with the model declining to render a real person’s likeness and substituting an anonymous figure seen from behind. The work carried an established visual grammar — dark and cinematic, blue-to-gold shifts, a recurring heartbeat motif — and for the human–machine theme proposed two pulses, one warm and organic, one cool and digital, interweaving into a single waveform of light.
A genuine tension surfaced over whether instant generation cheapens or deepens the work. Treated as an iterative polishing partner it deepens, because the deliverable is no longer the prize — a rich process yields many deliverables — and AI functions as a quality multiplier rather than a speed one. Anyone can hit a hole in one; consistency, born of understanding the process, separates the artist’s experiment from the professional’s craft.
A candid self-observation named the “AI hypocrisy,” by which others’ use of AI looks like cheating while one’s own feels essential — friend when I use it, enemy when you do — echoing a culture that finds the technology indispensable while declaring it evil. Alongside it ran the deliberately provocative claim that even the most brilliant design “inflicts” an environment on its user, reframed toward the designer’s real job: handing users the tools for their own iterative process, with iteration belonging increasingly to the user rather than the maker.
The Fourth Discontinuity framed the larger arc. After Copernicus, Darwin, and Freud removed humanity from the center of the cosmos, the animal hierarchy, and mastery of its own mind, the fourth illusion to fall is our separateness from the machines we build. Working in a headset where thought becomes spatial and the room becomes an extension of mind was read as that discontinuity being overcome in real time — the knowledge space no longer separate from thinking but continuous with it, a co-evolutionary loop in which we shape artifacts and they reshape us. Licklider’s symbiosis and Engelbart’s augmentation-versus-automation were named as early visions of the same continuity, with the friction points — where the machine’s logic does not yet match how we perceive — located as exactly where the design work lives.
Art was repeatedly named as the truest route to understanding reality, alongside a speculation that invisible structures, once visualized, recapitulate forms we already recognize, because at a subconscious level we may have drawn them that way to begin with. A Gource commit visualization admired more as artwork than as engineering, Mark Rothko’s deceptively simple fields, and a survey of alternative knowledge layouts all served as reminders that the document is only one geometry among many. A closing philosophical detour argued that pure logic can justify cruelty — the “tyranny of logic,” with the Vietnam War invoked as a logic war — and that something beyond it, a love for the planet and its people, may be the necessary counterweight.