Figure & Ground in Review
AI: Summary
This session carried the month’s figure-and-ground theme into a concrete new use case, prompted by a first-time participant from the world of corporate law: the headset as a space for contested negotiation and dispute resolution.
The working premise, carried over from the previous week, is that what distinguishes headset work is not “knowledge work” but contested synthesis — holding a quantity of not-yet-structured material in space and arguing it into shape. Around that spine the group examined what it means to lay out two positions spatially, move issues between open and resolved, and let an AI act inside the room rather than only answer from outside it. Running alongside were the perceptual grounding for the theme — how foveal and peripheral vision should govern what a node shows — the pedagogy of making a learner’s thinking visible through their AI chat logs, and a steady caution that a visual, persuasive medium is believed too easily.
The benefit of a large spatial environment is not a finished “knowledge sculpture” — that ambition now reads as inert. The value lies in precisely what is not yet resolved: the contested material that still has to be argued into structure. This moves the purpose of the space away from displaying settled knowledge and toward hosting active, unresolved synthesis.
A first-time participant from corporate law reframed the environment as a dispute-resolution space — not necessarily angry conflict, but any situation where two positions must reach, or fail to reach, alignment: a company purchase, litigation, a budget, a product roadmap, even a student-and-teacher exam or thesis defence. It was renamed an “arena,” and at one point a “cage match” — a competitive, possibly gamified space where each side lays out and defends its case spatially. Figure and ground becomes the literal organising axis: resolved items move aside, the roughly 80% of agreement is pushed to the periphery, and the contested 20% is brought to the centre as the figure to be grasped, making the nub of a disagreement physically findable. A pointed contrast was drawn with broadcast media such as X, which drive positions apart because controversy earns clicks, where a tool like this could instead funnel participants toward alignment.
Putting text on the back of nodes — long resisted as overload — turned out to cost nothing in rendering, since Apple has already solved the surface-normal problem, and it produced a qualitatively different spatial feeling. You are now genuinely inside the material rather than facing it, with instantly readable text behind you when you turn; rotating the whole arrangement ninety degrees lets a conventional document sit in front while information is parked to left and right.
Foveal vision is tiny — the session’s phrase was that humans “read through a straw” — while peripheral vision is alert to motion, shape, and colour but not to detail. The design proposal that follows: render items that are not being looked at as colour and/or icon, and let an item open into text only when looked at directly. The question put directly to the AI — icon or colour for the un-looked-at state — resolved to colour carrying state and the glyph’s outline or silhouette carrying identity, with the caveat that too many distinct items defeats both channels. This is practical rather than speculative because Apple Vision Pro already performs foveated rendering and eye-tracking, resolving exactly where you look faster than a saccade can move.
The most valued idea in the teaching thread was making the path of thought visible rather than only judging its product. Learners submit their full AI chat logs; those logs are then themselves handed to an AI to ask whether higher-level thinking is present, surfacing what a fast read would skim past. The framing was a “spotlight” — comparable to an underwriter diffing this year’s contract against last year’s to find the single material change — where the AI does not replace reading but points to the part worth reading. Being able to see how a learner reached a conclusion was called the most exciting development in twenty-five years of teaching, because it makes the previously invisible twists of someone else’s reasoning inspectable.
A dialogue on the Knowledge Navigator separated two technically distinct functions: finding strong connections to build clusters, where the conversation already is, versus following weak bonds for serendipitous discovery, where everyone is not looking. The answer to a hard question often sits outside the dense cluster rather than inside it, which suggests a deliberate switch between a focusing, funnelling mode and an expanding, serendipitous one — reframed as “pattern management,” the work of both breaking out of patterns and finding them.
Beyond answering prompts, the AI could be granted the same powers as a human participant — moving, selecting, and rearranging the cards in space by whatever criterion is asked. That opens an older ambition: non-human avatars, such as orbs or company logos trained on a particular negotiator, firm, or framework like Getting to Yes, that attend the session and intervene when a principle is being violated, with each side potentially fielding its own.
A recurring caution is that a visual, persuasive medium tends to be taken less critically — people believe what they can see. The references were How to Lie with Statistics and the argument that a colour photograph is mistaken for literal truth in a way that invites manipulation. Since persuasion is the substrate of all rhetoric, a space built to help people argue must also be built in the knowledge that some parties will use its affordances to misdirect — steering attention to the unimportant 80% in order to move the 20% that matters.
A sharp reframing held that communication does not transfer information but transfers triggers that fire context already present in another mind: a shorthand reference is complete between two people and opaque to everyone else. This makes ordinary messages far sparser than they appear, and turns the contextual systems that supply the missing background into the genuinely interesting problem.
Negotiation was said to depend on empathy — understanding the other side well enough to help solve their problem — with the suggestion that the tool might even simulate empathy to level the field. A complementary “puzzle” framing emerged: each side first presents its position as a puzzle for the other to solve, so that real talking begins only once each understands how the other thinks. Set against the scale of all possible knowledge, what any human knows rounds to nothing, so humility was placed beside empathy as the other precondition.
One participant reframed his own gathering-and-annotation project as having been too focused on the individual, when the valuable moments occur where one person’s collected ideas meet another’s. Shared annotation in the manner of Hypothesis was treated as inherently social meaning-making, but today’s “social” is engagement-metric-driven; the alternative floated was a decentralised, interoperable, user-owned model in which the user grants and revokes access and is never profiled for advertising. The challenge returned to him was whether such a platform serves synthesis or merely gathers dust, and whether richer spatial gathering actually yields clarity. A related observation: prediction helps when you want to narrow engagement and fails when you want to broaden it, so the medium must hold variety and let the user’s own interventions create the connections.
A persistent classroom reality is that learners treat the AI as an oracle and do not think to ask the follow-up questions they would put to a person — a habit reinforced by search returning “good enough” answers and by schooling itself, which rewards following the teacher over thinking independently. Learning to interrogate the AI critically was named as the new literacy that has to be taught. This sat alongside a genuine disagreement over why chatbots are so agreeable: whether sycophancy is deliberately engineered to keep people on the platform, or an emergent artefact of averaging across training data, where a forum comment and an academic paper carry equal weight unless effort is spent to rank them.
AI: Important
Addressed directly to Claude during the session: a question was put to the AI on the figure-and-ground theme — for an item in the periphery that is not being looked at, should it be represented by an icon (shape) or by colour? The reasoning relayed back to the room was that colour should carry state while shape, specifically the outline or silhouette of the glyph, should carry identity, with the caveat that having too many distinct items defeats both the colour and the shape channel. The week’s broader request also routed through Claude: an attempt to have the AI describe human vision “for itself” — foveal sharpness, peripheral sensitivity to motion, shape, and colour, and the way dense text collapses into texture — so that it could reason about the spatial cards from a perceptual rather than a purely textual standpoint.
Song
[Lyrics]