**Note** from Bead: 96597c32 · [canonical source](https://redfish.acequia.io/guerin/.agents/96597c32-7f1c-4e95-acd5-635673461781/2026-05-16/notes/action-principle-and-collective-intelligence.md) · session 2026-05-16 · discussion: Talk: 96597c32
From the opening moves of the [2026-05-17 Claude conversation](../../uploads/2026-05-17T000000-claude-e851497a.md): > The deepest framing I keep coming back to: **computation was always continuous.** A river computed its watershed long before anyone wrote a hydrology equation. A soap film finds minimal surfaces in real time for free. … The world finds extrema of action everywhere, all the time. That is computation, in the only sense that ultimately matters. This note states the **algorithm for collective intelligence** in its most universal form: **bidirectional path selection on a co-evolving substrate.** A forward signal from the source meets a backward signal from the sink; the path emerges where they find consistency; the path modifies the substrate; the next pass runs against the modified substrate. Causal. Iterative. Substrate-modifying. > **An important precision.** Physics describes the *artifact* this algorithm produces — the path it settles on — via the principle of least action and the stationary action integral. **That description is the bookkeeping, not the mechanism.** Hamilton-Lagrange formalism is what an outside observer computes after the algorithm has settled. It is correct but derivative. The algorithm is what does the work; the integral is the receipt. The rest of this note treats the algorithm as primary and the integral as a useful but secondary description. Least-effort pathways and desire lines are the human-scale and ecological-scale manifestations of the algorithm running on landscape substrates. The note ties the digital/analog distinction ([event-field-cut-as-design-choice.md](event-field-cut-as-design-choice.md)) directly to the substrate on which intelligence runs.
## The opposite of "digital" is *continuous* The Claude upload opens with a Fourier-analysis observation. The audience-facing version of "digital" usually means *discrete* — sampled, quantized, countable. Its precise opposite is **continuous**. The 2×2 grid: | | Continuous time | Discrete time | |---|---|---| | **Continuous frequency** | Continuous Fourier Transform (the *analog* limit) | DTFT | | **Discrete frequency** | Fourier series | DFT / FFT (the *digital* limit) | The duality generalizes immediately to space (continuous space ↔ continuous spatial frequency / wavenumber `k`; discrete pixels ↔ Brillouin-zone-quantized `k` on lattices). And it generalizes beyond Fourier to *any* domain we transform over. **The deep pairing is sampling ↔ periodicity**, not "fake vs real." So when we ask the talk's question — *what is the opposite of digital landscape architecture?* — the answer is not "analog landscape architecture" in a nostalgic sense. The answer is **continuous landscape architecture** — and the discipline has always been continuous, because the substrate it works on is.
## Why we started digital The historical correction (cross-link to [turing-two-minds.md](turing-two-minds.md)): > The digital is what minds can share. Turing's machine modeled a human computer (the job title) writing legible symbols on a tape, erasing, rewriting, so another human could pick up the work and verify each step. Shannon's 1948 thresholding was about noise immunity for communication across channels. Von Neumann's architecture made the substrate sequential and auditable. **The bit, the clock cycle, the instruction trace are protocols for intersubjective verification.** They exist so one human can audit another's work, and so a result computed in one office can be replayed in another decades later. **We mistook a communication protocol for a substrate.** The world itself does not run on bits. The world finds extrema of action — and the bits are *receipts logged after the fact.*
## The bidirectional algorithm — and what physics calls its artifact The algorithm has four moves, in order: 1. **Forward propagation** from the source. An "offer," a "seeking," a probe carried through the substrate by whatever the substrate carries (charge, water, ants, walkers, photons, GETs, chemical concentration). 2. **Backward propagation** from the sink. A "confirmation," an "affordance," a counter-signal that the destination is reachable and what it requires. 3. **Path emergence** where forward and backward find consistency on the substrate. The path is *not pre-existing* in the geometry; it is *produced* by the intersection. 4. **Substrate modification** by the produced path. The substrate's state at time *t+1* is no longer what it was at time *t*; the next pass runs against the modified substrate. The path and the substrate **co-evolve**. This is **bidirectional algorithmic path selection**. It is: - **Causal** — both forward and backward propagations are physical/computational processes happening in time; their meeting is the path's birth. (Contrast with the teleological reading of the action integral, where the system is described as if it "knows" the endpoint and selects the path connecting start to end.) - **Iterative** — the algorithm runs repeatedly; each iteration sees a substrate modified by the last. - **Substrate-participatory** — the substrate is a stateful third party, not a fixed background. Its modifications carry the memory of past iterations into the future. ### What physics calls its artifact When the algorithm is run on a vacuum-like substrate that doesn't carry state forward (idealized optics, classical mechanics, quantum-mechanical Hilbert space), the *artifact* it produces — the path it settles on — can be described compactly as an extremum of the action functional: $$ S = \int_{t_0}^{t_1} L(q, \dot q, t)\, dt $$ The principle of least action (Maupertuis 1744, Euler, Lagrange, Hamilton) says the path that actually occurs makes `S` stationary. **This description is correct for the artifact, in those idealized cases.** The recognizable physics phrasings — Fermat's principle (light), Maupertuis' (mechanics), Hamilton's (fields), Feynman's path integral (quantum amplitudes) — are all artifact-descriptions of the same algorithm running on substrates that physicists chose specifically because they *don't* carry state forward. For substrates that *do* co-evolve with the paths running on them — desire lines, drainage networks, ant pheromones, Hubler bearings, watershed maintenance, acequia governance, cache matrices, conversation — **the integral view degrades.** You cannot write down a single Lagrangian whose stationary path is the desire line, because the cost-functional itself is modified by the path's existence. The algorithm is primary; the integral is at best a per-iteration approximation. **Hamilton-Lagrange formalism is the bookkeeping; the bidirectional algorithm is the work.** ### The ontological move: the landscape *is* the computation; it does not *solve* anything A precision worth stating clearly, because trained-in-physics readers will reach for the wrong intuition: there is a temptation to read "the bidirectional algorithm" as if a *solver* were running somewhere — as if the landscape were performing calculus of variations and finding the path that extremizes `S`. **This is exactly the mistake the algorithmic framing is meant to dissolve.** **The landscape does not solve calculus of variations. The landscape does not compute the action integral. The landscape does not analytically find extrema. The landscape does not "run an algorithm" in any sense that implies a separate solver layered on top of physics.** What the landscape does is simply *be itself.* Water runs downhill. Ants deposit and follow pheromone. Walkers compact soil. Photons exchange confirmation. Bearings polarize in field. Citizens fetch and serve URIs. **These dynamics ARE the computation.** The path is one of the things the substrate's dynamics produce, indistinguishable from what the substrate physically does. There is no separation between "the substrate's behavior" and "the algorithm running on it." Calling it an *algorithm* is convenient for humans who want to describe the pattern; the substrate does not know it is running one. Calculus of variations is what *humans* do to predict what the substrate will produce. The Euler-Lagrange equation is a tool in a mathematician's office, not a procedure that the river runs. **The river doesn't know it's solving anything; it just flows. The path it carves is what flowing *is*.** This is the ontological half of **landscape as computation** ([cognitive-landscapes-and-three-intelligences.md](cognitive-landscapes-and-three-intelligences.md), definition 4). The landscape does not *have* computation as an attribute; the landscape **is** computation by being itself. The path is its native output, not the result of an analytic procedure layered on top. This precision matters for the talk because it forecloses the misreading that the discipline of landscape architecture is being asked to do *more math* (the variational solver). It is being asked to *recognize what the substrate is already doing* and to *compose with it* — which is the discipline's craft, not its physics-envy. ### Clean implementations of the algorithm in our existing vocabulary
- **Wheeler-Feynman absorber theory** (1945) — emitter's retarded wave (forward), absorber's advanced wave (backward), photon is the standing-pattern agreement. See [sympoiesis-over-autopoiesis.md](sympoiesis-over-autopoiesis.md). - **Transactional QM** (Cramer 1986) — same algorithm, named as the foundational interpretation of quantum mechanics. The path does not exist until both ends agree. - **Mead's *Collective Electrodynamics*** (2000) — the algorithm cast as circuit theory. The field is the coupling between coupled oscillators. See [hubler-network.md](hubler-network.md). - **dualAnt** (Stephen's framework) — the algorithm implemented as a digital procedure. Forward flood-fill from source; backward flood-fill from sink; path is where they meet. - **Active inference** (Friston) — the algorithm cast as a cognitive principle. Action is forward propagation against a generative model; perception is backward propagation updating the model. - **Ant colony optimization** — the algorithm in biology and bio-inspired CS. Forward pheromone deposit; backward path-selection biased by pheromone gradient; trail concentration with decay. - **Slime mold path finding** (*Physarum polycephalum*) — the algorithm in eukaryotic biology. Forward cytoplasmic flow; backward reinforcement of high-flow tubes; pruning of low-flow ones. - **Desire-line formation** — the algorithm at human-and-landscape scale. See section below. - **Acequia governance** — the algorithm wrapped in explicit collective receipts. See section below. **Same algorithm. Different substrates. The artifact-description differs from substrate to substrate; the mechanism does not.**
## The same bidirectional algorithm across substrates The algorithm is not unique to physics. It is **what happens whenever a flow finds a path through a substrate that can be modified by the flow.** Each row below is the same algorithm running on a different substrate, with different forward and backward signals and a different substrate-modification rule: | Domain | Forward signal (source → ?) | Backward signal (sink → ?) | Substrate modification | |---|---|---|---| | Optics | Photon emission (retarded wave) | Absorber confirmation (advanced wave) | None — vacuum is non-evolving (artifact: Fermat's principle) | | Mechanics | Force / momentum from initial state | Configuration affordance of final state | None in idealized case (artifact: Hamilton's principle) | | Quantum mechanics | Offer wave from emitter | Confirmation wave from absorber | None — Hilbert space is non-evolving (artifact: Feynman path integral) | | Active inference (cognition) | Motor commands generated against generative model | Sensory updates contradicting predictions | Generative-model belief update (Friston) | | Ant foraging | Pheromone deposited during walk | Pheromone-following bias on later walks | Trail concentration + decay (Oettler 2013) | | Slime mold *Physarum* | Cytoplasmic flow toward food | Reinforcement of high-flow tubes | Tube-diameter change; pruning of low-flow tubes | | Watershed drainage | Water flow | Gravitational sink | Erosion, deposition, channel deepening | | Hubler self-assembling wires | Bearing polarization in field | Field-gradient pull toward rim | Wire-chain forms part of the field (cf. [hubler-network.md](hubler-network.md)) | | Desire-line formation | Walker forward intent | Destination affordance | Soil compaction, vegetation kill | | Acequia maintenance | Water flow by gravity | Community labor against erosion/silting | Ditch geometry persists; modified by saca | | dualAnt (digital) | Forward flood-fill from start | Backward flood-fill from goal | Substrate marking persists across iterations | | Cache matrix (digital) | GET request | Response + caching decision | Cache state updates for next GET (cf. [semantic-highways-and-usage-gradients.md](semantic-highways-and-usage-gradients.md)) | | Conversation | Utterance | Listening + memory | Shared narrative state updates | **Each row is the same algorithm running on a different substrate.** The recurrence is the point. The algorithm is not "physics imported into biology"; it is the universal shape of how flows compose with substrates that can hold state. (The Action Bits framing — see [event-field-cut-as-design-choice.md](event-field-cut-as-design-choice.md) — names the *discrete event* that closes one iteration of the algorithm. The substrate-modification step is what makes the next event possible. The algorithm is the substrate; the bit is the receipt logged when one iteration completes.)
## Intelligence as the algorithm — collective perception + collective action Define **intelligence** operationally: > Intelligence is the algorithm that, given a substrate that can hold state, runs the **bidirectional path-selection loop** — propagating forward from intent, propagating backward from goal, settling on the path where they meet, and modifying the substrate so the next loop runs against a changed state. Two dual half-operations on the same substrate: - **Perception** is the *backward* half. Given sense data (a slice of the substrate's current configuration), update the model of where the system is and what is reachable. This is the confirmation-wave / advanced-wave direction of the algorithm. - **Action** is the *forward* half. Given a state estimate and an intent, execute an event that propagates a probe into the substrate — a step, a token, a chemical commitment, a vote, a phone snap of a flood. This is the offer-wave / retarded-wave direction. **Both are operations on the analog substrate.** Both deposit traces (perception leaves an estimate, action leaves a path-modification). The traces accumulate as the continuous field that the next round of perception-and-action operates on. **The loop closes through the substrate itself.** This is **active inference** in Friston's formulation. It is **stigmergy** in Wilson's. It is **transactional quantum mechanics** in Cramer's. It is **Wheeler-Feynman** in the absorber-theory frame. It is **Mead's collective electrodynamics** at circuit level. *Same algorithm, twelve substrates.* See [sympoiesis-over-autopoiesis.md](sympoiesis-over-autopoiesis.md). **Collective intelligence** is what happens when many agents run this algorithm on a *shared* substrate simultaneously: - Each agent perceives the substrate. - Each agent acts into the substrate. - The substrate integrates their perceptions and actions. - The next round, each agent perceives a substrate that already carries the trace of every other agent. - Out of this loop, collective behavior emerges that no single agent computed. The colony does this with pheromone fields. The watershed does this with rainfall + erosion + vegetation + flood. The acequia does this with water + ditch + prorrata + saca. The conversation does this with utterance + listening + memory. **The substrate is what makes them collective.**
## Least-effort pathways are the bidirectional algorithm running on landscape Now the connection lands. **Least-effort pathways are the bidirectional path-selection algorithm running on a landscape substrate.** Every example landscape architecture already cares about is an instance — and crucially, every one of these involves a *substrate that co-evolves*, which is exactly where the textbook integral view breaks down and the algorithmic view is necessary: - **Water finds its level** — gravitational action minimized on hydrological substrate. - **River drainage networks** — total dissipation minimized; Murray's-law-cousin scale-invariant ratios (cf. [hubler-network.md](hubler-network.md)). - **Slime mold trails** — *Physarum* solves shortest-path problems by reinforcing high-flow tubes and pruning low-flow ones. Literally implementing the variational calculation. - **Tiger-bush vegetation banding** (Niger, Somalia) — biomass and water find the action-stationary stripe configuration (Turing-1952 mechanism with action interpretation). - **Fire spread fronts** — heat + fuel + wind compose action that the front extremizes; the front IS the stationary path. These are not metaphors. **They are the bidirectional algorithm running on different substrates, each with its own substrate-modification rule.** And critically — *each substrate co-evolves*, which is what makes them harder than the textbook physics cases and exactly what makes the algorithmic framing necessary. The discipline of landscape architecture has been working with the algorithm in physical disguise for as long as it has existed.
## Desire lines are the human-scale and *governance-scale* instance The single sharpest example for the DLA audience: - A walker faces a substrate (the lawn, the field, the city). - The walker's action functional is approximately: time × effort × discomfort, integrated over the path. - The walker chooses the action-stationary path. - The path is traversed. The substrate is modified (grass compacts, soil compresses, dust accumulates). - The next walker faces a *changed* substrate. The action functional has been deformed by the previous traversal. - The next walker's optimal path is *biased* toward the previous one. - Iterate. The path becomes a desire line. The desire line becomes a trail. The trail becomes a road. **This is action-principle co-evolution.** The path and the substrate are mutually shaping. **It is also the universal algorithm of collective intelligence in landscape**: many agents, each running perception-action loops, depositing traces into a shared substrate that integrates and biases future loops. The bidirectional version of this is dualAnt / dualAlign / Wheeler-Feynman: the walker reaches *forward* (offer wave), the destination is *already affording arrival* (confirmation wave), the desire line is the *standing pattern* where forward and backward components find consistency. See [hubler-network.md](hubler-network.md). ### The algorithm runs at *the reactive interface* A precision worth naming: the algorithm does not run "everywhere in the substrate uniformly." It runs **at the thin reactive surface where slow form meets fast flux** — what [reactive-interface.md](reactive-interface.md) names as the third term in the form/flux dyad. The active layer at a river bed (few grain diameters thick). The fire front (meter-deep flame zone). The cell membrane. The retina. The clearing price at the top of an order book. The compuerta where parciantes negotiate. The desire-line compaction surface (top centimeter of soil). The Service Worker's cache decision point in a Web tab. The bidirectional path selection is a *boundary phenomenon* in physical substrates that have a form/flux distinction. **The interface is the computational organ.** Form persists; flux passes; the interface is where action becomes form and form is paid back into action. This refines the algorithm's claim from "the substrate computes itself" to "the substrate computes itself *at its reactive interfaces*." Same algorithm, more precise about location. ### Forman-native translation: *corridors modify the matrix that regulates the corridors* For the DLA audience, the action-principle co-evolution claim — *the path modifies the substrate that judged its cost* — translates directly into landscape ecology's own vocabulary: > **The corridors modify the matrix that regulates the corridors.** Same dynamics, audience-native terms. The structure is **chiastic** — A modifies B that regulates A — and the claim is sticky because the structure folds back on itself. This is the phrasing to use in the talk's Beat 3. The audience hears the language they already work with — Forman's matrix/patches/corridors — and the recognition does the rest. They spontaneously connect the claim to desire lines, drainage networks, Murray's law, Hubler bearings, ant pheromones, slime molds, acequia ditches — *without further prompting* — because all of those instances live in their existing professional intuition once the chiastic frame is named. The same shape runs across every scale this bead has discussed: - *Walkers modify the lawn that judges the walkers.* - *GETs modify the cache matrix that routes the GETs.* - *Bearings modify the field that grew the bearings.* - *Photos modify the spatial cache matrix that finds the photos.* - *Saca-labor modifies the ditch that governs the saca.* - *Mayordomos modify the protocol that elects the mayordomo.* Each is the corridor / matrix / corridor structure at a different scale. The talk does not need to enumerate them on stage. **Naming the single chiastic sentence — *the corridors modify the matrix that regulates the corridors* — lets the audience do the enumeration internally.**
## The acequia as the action algorithm explicitly governed The acequia is the human-scale governance overlay on top of the same algorithm: - Water flows by gravity along the ditch (action-minimizing on hydrological substrate). - Parciantes draw water (each draw is a discrete action event closing a quantum). - The mayordomo coordinates allocation (governance-layer constraint on the action functional). - The prorrata is the explicit cost ledger (the receipt log read at high resolution). - The saca is the collective maintenance event (the substrate is repaired so the action algorithm continues to converge). Acequia governance does not *replace* the action principle; it **wraps explicit collective intelligence around it**. The discrete receipts (prorrata, saca attendance, mayordomo election) are how the human collective audits and reforms the action algorithm running on the watershed. This is exactly the design pattern the digital landscape needs and currently lacks (cf. [data-sovereignty-and-sousveillance.md](data-sovereignty-and-sousveillance.md)).
## Why this is the talk's deepest claim The DLA-2026 talk's full diagnosis is now visible at one stroke: 1. **Substrate**: the landscape (physical and digital) is an analog medium on which the action principle runs. 2. **Algorithm**: collective intelligence is many agents running perception-action loops on that substrate, mediated by the substrate's integration of their traces. 3. **Receipts**: digital records (bits, prorrata, ledgers, citation graphs) are the *protocols for intersubjective verification* of the algorithm, not the algorithm itself. 4. **Design**: the practice is to **compose perception, action, and receipt-protocols deliberately**, choosing where to put the cut between event and field (see [event-field-cut-as-design-choice.md](event-field-cut-as-design-choice.md)). 5. **Pathology**: the current digital landscape captures the substrate (corporate enclosure), starves public observation (government dismantling), and offers no architecture for citizen receipts (sousveillance gap; see [data-sovereignty-and-sousveillance.md](data-sovereignty-and-sousveillance.md)). 6. **Intervention**: build the architecture that lets collective intelligence run on the analog substrate with governed receipts — the acequia pattern, the three-front pedagogy, sym-sovereign identity. The discipline of landscape architecture has been doing all six implicitly. Naming the action principle gives the implicit work an *operational definition* and a *universal lineage* — physics, biology, cognition, governance — that no other design discipline can claim with the same depth.
## Use in the talk The action-principle move is the **glue** that holds the talk's four acts together. Each act invokes it implicitly: - **Act 1 (Turing's two minds)** — both 1936 (discrete events / bits / receipts) and 1952 (continuous fields / morphogenesis / action) are aspects of the same action algorithm at different resolutions. - **Act 2 (three intelligences)** — each is the action algorithm on a different substrate; together they compose the cognitive landscape. - **Act 3 (governance turn)** — acequia governance is collective intelligence wrapping explicit receipts around the action algorithm. - **Act 4 (ontological turn)** — "there are no selves, only transactions" is the action-principle statement at metaphysical level. The transaction is the stationary-phase agreement; the self is the receipt logged when the agreement settles. **The single sentence that ties the talk together:** > Landscape architecture has always been the discipline that designs by composing the action principle — least-effort pathways, desire lines, drainage networks, succession trajectories, water rights — into governed collective intelligence. The digital is a layer of receipts on top. The architecture is how we choose to compose them.
## Connections - [turing-two-minds.md](turing-two-minds.md) — 1936 = receipt protocol; 1952 = continuous action substrate. - [event-field-cut-as-design-choice.md](event-field-cut-as-design-choice.md) — the cut between event (action closing) and field (action integrating) is the design choice. - [sympoiesis-over-autopoiesis.md](sympoiesis-over-autopoiesis.md) — transactional QM / Wheeler-Feynman / Mead — the bidirectional implementation of the action algorithm. - [hubler-network.md](hubler-network.md) — Hubler bearings find action-minimizing wire networks; cousins of Murray's-law vasculature, drainage basins, slime molds. - [cognitive-landscapes-and-three-intelligences.md](cognitive-landscapes-and-three-intelligences.md) — the three substrates the action algorithm runs on. - [three-front-pedagogy.md](three-front-pedagogy.md) — students learning to compose receipts and substrates deliberately. - [data-sovereignty-and-sousveillance.md](data-sovereignty-and-sousveillance.md) — what happens when the substrate is captured and the receipts are missing. - [field-and-bead-reconciliation.md](field-and-bead-reconciliation.md) — Smith's GPT made the same move from a different vocabulary; *beads are coordinate markers on the action substrate.* - [advanced-wave.md](../../../874fce5b-9c8b-4b23-b2ed-429148c6c4b7/2026-04-23/notes/advanced-wave.md) — the bidirectional accounting that makes the algorithm settle.
## References (for the speaker's confidence layer) - Maupertuis, Euler, Lagrange — the foundational principle of least action in mechanics. - Hamilton — variational principle as the general formalism. - Fermat — least-time in optics; the prefigural form. - Feynman — path integral; the quantum-mechanical generalization. - Onsager — reciprocity relations in linear non-equilibrium thermodynamics. - Karl Friston — variational free energy / active inference. Most directly relevant for the *intelligence as algorithm* framing. - Maximum entropy production principle (Paltridge, Dewar, Kleidon) — the thermodynamic generalization. - Oettler et al. 2013 — ant foraging as Fermat's principle implemented in pheromone chemistry. - Steve Mann — sousveillance vocabulary for the receipt-protocol layer (see [data-sovereignty-and-sousveillance.md](data-sovereignty-and-sousveillance.md)).
## Sentences to keep > Paths are produced everywhere, all the time, by a bidirectional algorithm: a forward signal from the source meets a backward signal from the sink on a substrate that the resulting path then modifies. **The landscape does not solve calculus of variations; the landscape *is* the path-computation by being itself.** Physics describes its artifact as the principle of least action and writes it as a stationary integral — correct but derivative; the integral is the bookkeeping; the substrate's behavior is the work. Desire lines are that work running on grass. Acequia governance is that work wrapped in explicit collective receipts. The discipline of landscape architecture has been composing this work into governed collective intelligence for centuries — it just did not have the vocabulary to say so out loud. **The audience-facing version, in Forman's own vocabulary:** > The corridors modify the matrix that regulates the corridors. The landscape *is* the computation of the path — it does not solve for the path on top of being. That is the discipline's quiet operating method, named at last.
## References (bead cross-links) - Bead: 874fce5b · [canonical](https://redfish.acequia.io/guerin/.agents/874fce5b-9c8b-4b23-b2ed-429148c6c4b7/)