v0.03
Systems notation · Isotype · 2026-10-08

A notation for collective systems

Six papers on collective intelligence, cooperation and human-AI coevolution describe the same machinery from six angles. These marks name its parts once, so any system in them, or one we work on, can be drawn and set in motion.

01 · The layers

Seven layers run through all six papers

LayerHenrichWilsonKehlerDafoeBrinkmannPedreschi
LevelAt what scale does selection act?
EngineWhat varies, what copies, what gets kept?
WiringWho is connected to whom, and how many?
LoopWhat feeds back on itself, and how fast?
StakesHow aligned are the interests?
RulesWhich institutions shift the balance?
ActorsWho is in the system: people, machines, groups?
carries the argument present absent
The hinge

Selection builds function at the level where it acts, and selection one level down erodes it. Engagement ranking selects at the level of the single post.

Wilson, from Muir's hens: select the best hen and you breed aggression; select the best cage and egg output rises 160% in five generations.
THE KIT · v0.03

Twenty-eight marks, seven families

One figure, one meaning; more means more figures. Every diagram on this page is built from these marks.

AActors8 marks

  • A1

    Human

    A person: a learner, chooser and carrier of traits.

    HENRICH
  • A2

    Machine

    A model or algorithm that generates, copies or ranks.

    BRINKMANN
  • A3

    Hybrid

    A human and machine acting as one unit.

    DAFOE
  • A4

    Stock

    Accumulates over time and changes only through flows: culture, a corpus.

    MEADOWS
  • A5

    Trait

    One idea, move, design or habit that can spread.

    BRINKMANN
  • A6new

    Organism

    A non-human individual: hen, water strider, microbe.

    WILSON
  • A7new

    Outlined trait

    A variant from a second source, set beside the filled one.

    BRINKMANN
  • A8new

    Commons

    A shared resource; overuse empties it block by block.

    WILSON · OSTROM

BLevels3 marks

  • B1

    Unit of selection

    Solid boundary: selection acts here, so function can form here.

    WILSON
  • B2

    Aggregate

    Dashed boundary: a grouping; selection acts on the parts inside.

    WILSON
  • B3

    Nesting

    Levels stack: person, group, population; read them inside out.

    WILSON

CEngine4 marks

  • C1

    Variation

    Recombination: two existing traits meet and a new one leaves.

    HENRICH
  • C2

    Transmission

    Solid for faithful copying, gapped for lossy copying.

    HENRICH
  • C3

    Selection

    A filter that keeps some traits; always label what it rewards.

    WILSON
  • C4

    Drift

    Change by chance: frequencies wander at random.

    WILSON

DWiring3 marks

  • D1

    Connectivity dial

    Partial connection outperforms full: clusters keep options alive.

    HENRICH
  • D2

    Diversity

    Different kinds of minds meeting: the fuel for recombination.

    HENRICH
  • D3

    Rewire

    A machine changes who learns from whom.

    PEDRESCHI

ELoops and time5 marks

  • E1

    Reinforcing loop

    Change feeds more of the same change.

    PEDRESCHI
  • E2

    Balancing loop

    Change triggers a correction toward a level.

    MEADOWS
  • E3

    Polarity

    Plus moves the target the same way, minus the opposite way.

    STERMAN
  • E4

    Delay

    The effect arrives later: retraining, a generation, a season.

    STERMAN
  • E5

    Lock-in

    An early good-enough path spreads and crowds out a better one.

    HENRICH

FStakes and commitment2 marks

  • F1

    Interest bar

    Edge colour shows how aligned the two ends are.

    DAFOE
  • F2

    Edge kinds

    Five ways two actors can be joined.

    DAFOE

GRules3 marks

  • G1

    Valve

    An institution on a boundary; name it: masked identity, junior first.

    HENRICH · KEHLER
  • G2

    Tug

    Two levels of selection pulling opposite ways; mark which wins.

    WILSON
  • G3

    Outcome badge

    Where an effect is measured: individual, item, model, system.

    PEDRESCHI

Sources: Henrich, Wilson, Ostrom, Meadows, Sterman, Brinkmann, Dafoe, Pedreschi, Kehler · Marks: A1 to A8, B1 to B3, C1 to C4, D1 to D3, E1 to E5, F1, F2, G1 to G3

03 · In motion

Four systems from the papers, drawn with the kit

Each one plays through its steps on its own. Click a step to hold it.

The recommender loop · Pedreschi

Choices train the machine that shapes the choices

choices retrain suggestions engagement C3 E4 t0 t1 t2 A1 Human A2 Machine A4 Training data R E1 I filter bubble S concentration G3

People make choices, and each one lands in the training data. The stock fills block by block with whatever was chosen.

Source: Pedreschi et al., Human-AI Coevolution · Marks: A1, A2, A4, C3, E1, E4, G3

Muir's hens · Wilson

The funnel moved up one level

FLOCK B2 B1 EGGS PER HEN EGGS PER HEN EGGS PER HEN A6 FLOCK OUTPUT 1 EGG = ONE TENTH OF THE STARTING OUTPUT SOUTPUT FALLS illustrative: the source gives the direction, not the size S+160% IN 5 GENERATIONS G3 GENERATION 1 2 3 4 5 R AGGRESSION E1 EGGS PER CAGE C3 C3

One flock, three cages, nine hens, all laying. The egg row is the flock's starting output, drawn as ten eggs.

Muir's hen trials, via Wilson · Marks: A6, B1, B2, C3, E1, G3

Move 37 · Brinkmann

A machine’s variant restarted human variation

A2 machine C1 self-play: 4.9M games in 3 days A5 move 37 1 in 10,000 C2 L2 B1 A1 human players old human moves C1 A7 new human moves decision quality 1950 2016

The machine plays itself: 4.9M games in 3 days. Every game throws out move variants, and almost all of them fade.

Source: Brinkmann et al., Machine Culture · Marks: A1, A2, A5, A7, B1, C1, C2

Leonia, 2017 · Pedreschi

Every route optimal, the town jammed

Five drivers A3 within: fastest route within: fastest route within: fastest route within: fastest route within: fastest route Leonia B2 one street A8 town: streets passable G2 rule: none G1 S gridlock G3

Five drivers, each a person with a navigation app. Each app picks the fastest route for its own driver: selection acts on the single trip.

Source: Pedreschi et al., Human-AI Coevolution · Marks: A3, A8, B2, G1, G2, G3

04 · Grammar

Five rules for drawing with the kit

  1. Shape says what a thing is: figure for a person, block for a machine, vessel for a stock.
  2. Line says how things connect: solid copies, gapped loses, wavy drifts.
  3. Boundary says where selection acts: solid selects, dashed groups.
  4. Colour says only one thing: how aligned the interests are.
  5. Quantity repeats a figure: more eggs, more figures, never a bigger one.