NOTES · SIZE
Plenti rooms, one answer: how True Causes dey grow
True Causes dey run Structured Democratic Dialogue online. One hall dey hold two hundred pipo; internet no get hall, and to get fifteen pipo wey dey deliberate online no be di hard part: Structured Democratic Dialogue don dey do am for years. Di hard part na forty rooms of fifteen wey dey work on di same question and end with one answer — den four thousand, and nobodi load dey grow. Dat one need four tins: to know wen two groups tok di same tin, to carry claim wey two groups raise go meet di few groups wey fit judge am pass, to send deadlock go wia dem fit settle am, and to fold evri local map into one. Di method own recent work dey come near di last one: individual maps wey dem draw at different times wey dem use extend di map of one session (Laouris, 2023), and thematic maps wey dia authors merge for one final session (Michaelides & Laouris, 2024). Wetin nobodi don build na to do am for thousands of rooms, with no session to merge dem. Wetin e mean na collective intelligence without committee: nobodi dey hold di whole map, and nobodi need hold am.
Three tins dey tie co-laboratory to im venue: evribodi dey read di same wall, one facilitator dey hold evri thread, and di room dey decide wetin claim mean. Each one get im partner here wey no need di room. Facilitator dey welcome for any room; wetin change na say di method no dey depend again on one to dey dia.
Evri group get im own wall
Participants dey work for small groups — small reach say evribodi fit read di whole wall. Once room grow pass dat size, dem go divide am into groups, each one with im own wall, and di room map dey hold all of dem. Nobodi load dey grow with di room: di work of one pesin na di same whether di question get fifteen participants or two million.
Di same question, anywia wey dem ask am
Room no be di unit — na di question. Open dialogue and di platform go recognise oda rooms wey don already dey ask am, no matter how different dem word am, and e go join dem into one deliberation. Dia groups go become groups of di same question, and claims dey move between all of dem.
Few ideas dey travel; pipo no dey travel
Di same claim wey dem write for two groups dey count as one claim wey two groups raise. Na dat single choice dey make am possible to calculate for big size: without am, claim wey dey forty rooms na forty items and nobodi fit count anytin at all. Between rooms, few cards dey travel — afta room don write finish and bifor e start to explain, so notin go land while pipo still dey tink for demself. Na only few wey dem select dey travel, wey dem choose make dem different, so wall go remain wall no matter how many rooms dey. Dem dey arrive with di explanation of di pesin wey write dem, as normal cards, and dem dey judge dem like any oda one. Private room fit ask for dis one way: di cards from listed rooms for im question still dey land for im walls, and notin from im own dey comot.
How two wordings dey become one claim
To count groups get meaning only if you fit know wen two of dem tok di same tin for different words. Di platform dey spot di pairs wey fit be di same and e dey put dem to di pipo wey write dem. E no dey ever merge anytin by imself: two members must gree say two wordings na one claim, and any of dem fit take am back.
Di imbalance dey on purpose. Wrong merge na di mistake wey cost — fold claim enter popular one and e go inherit dat support — while merge wey dem miss only dey cost small efficiency, since di claim still dey travel as two items instead of one. Claims wey dey contradict or reverse each oda, dem no dey ever offer dem as di same claim, no matter how dem resemble.
Di map of evri group dey become one map
Each group dey arrange di wall wey e really deliberate on and e dey produce im own map. Dem go fold dat maps togeda: di same claim na one node, evri connection dey carry how many groups certify am on dia own, feedback loops dey collapse into single deep causes, and dem dey read di roots from di merged graph. Wia groups of one room certify opposite tins, dem dey publish di contradiction as conflict and send am, like any deadlock, to two oda groups wey must gree. Across rooms, wetin each room find go stand, and di link dey show how many of di rooms wey judge am find am: cause fit hold for one place and no hold for anoda. Nobodi dey average anytin away.
Collective intelligence for evri room, swarm across dem
Each room na collective intelligence on im own: pipo wey dey explain, discuss and judge togeda. Im thinking dey spread across im pipo, im wall and im map, wetin researchers call distributed cognition. Between rooms, di platform dey add swarm. For di swarms wey pipo know from nature and from software, each member dey follow di trail wey di oda ones leave: ants dey strengthen di path wey most ants take, and for live human swarm evribodi dey watch how di group dey pull. Na so swarm dey take converge, and na so crowd dey take follow-follow. Here di rooms no dey follow each oda. Room dey write bifor anytin land, e dey judge im pairs without seeing how oda rooms judge dem, and one organisation dey count as one voice no matter how many rooms e open. So wen plenti rooms certify di same link, dem find am separately, and di count get meaning. Agreement na evidence only as far as di sources dey independent, and na di point wey social epistemology dey make about testimony be dat. Di swarm dey inside di structure: no room dey hold di whole problem, di cards wey dey travel few and dem choose dem make dem different, and di map wey dey come out get links wey no single room for fit draw.
Why small groups. Below about twelve pipo, panel dey start to lose causes; above di size wey pesin fit read as one wall and follow as one conversation, e no dey gain anytin — na di size wey di method practitioners settle on across one hundred and twenty dialogues. So size dey come from more groups, no be bigger ones.
Wia deadlock dey go. E no dey go back to di group wey divide, and e no dey go to evribodi. Dem dey send di question to two oda whole groups, wey dem choose for way wey nobodi fit steer go meet friendly audience. Each one dey answer on im own, and dem must gree — if dem no gree, dem go publish di pair as argument. One group no dey ever enough: for simulation, one referral alone dey give back wrong answer 29.6% of di time wen bloc dey vote, but two or more no dey do am.
Wetin coordinated voting dey achieve. Three out of di fifteen seats of one group, one fifth of di group, wey dey vote togeda against one cause dey hold am for 40% of wia e suppose dey, and dem no touch anytin else. Referral to oda groups dey carry am back to 62%. Di seats dey close di rest — one way to enter di shortlist wey no pass through di pipo wey dey try keep am out. To pile votes on top di ballot alone no dey do anytin wey di seats no fix.
No model dey decide anytin. Dem dey use language model for one place only — to read argument wey dem write for plain language into premises so di logic check fit run — and dem forbid am to judge whether di argument correct; na fixed engine dey do dat one. Na pipo dem leave am for to decide say two claims na di same.
Pesin must earn support. Link dey count as one wey plenti rooms find only wen dat rooms really dey separate: rooms wey stand on one source, or vote like one, dey count once. For simulation, five rooms wey dem fake against five real ones remove real cause from di shared map for 31% of di runs; with dis rule di cause survive for evri run.
