Stigmergy
Coordination Mechanisms AIMech:Stigmergy DRAFT
A computational coordination mechanism in which actions modify an accessible environment or medium and those changes influence subsequent actions, allowing information to pass through the evolving state rather than requiring direct communication between the participating agents.
Identity Notes & Concept Scope
Selects environment-mediated coordination through action-produced state changes, as qualified by Heylighen 2016 and the computational treatment in Dorigo et al. 1999. It is a mechanism, distinct from the broader swarm design paradigm and from individual ant algorithms. The general mechanism does not require learning or beneficial coordination. Existing owner 262 is reused. The fresh collection on 15 October 2026 at 16:12:44–16:13:52 UTC covers 769 native and 442 pinned AIO records at merged base 41221e5f8916ff024296a8ed34622bda702c276d, with only the five reviewed schema/renderer/docs/test edits applied. All 4,825 identity fields (3,687 normalized values), including local IDs and the explicit Sarsa lambda source spellings, were compared. No candidate identity collision was found. Four ignored- and hidden-inclusive checkout scans retained 4,432 occurrences in 367 exact contents. Root read all 304 newly encountered contents, all 11 contextual fields and nine neighboring definitions, and seven complete matching frontier owners. The supporting assessment read all 367 contents, including 63 historical ones, and reconstructed every occurrence. All 909 main and 393 nested frontier objects were mechanically queried. This is a dated, bounded identity assessment, not global absence or discovery saturation.
Claims & Grounded Evidence
5 Claims
Mechanism
#action-trace-response
An action changes the medium, leaving a perceivable result that can trigger or modify a subsequent action. The defining relationship is the mediated action–trace–response loop; observing similar behavior without that link does not establish the mechanism.
Mechanism
#computational-local-state
The 1999 artificial-stigmergy construction associates state variables with problem states and gives artificial agents local access to their values. Agents modify these variables during activity, allowing later behavior to respond to changes left by other activity.
Scope
#action-coordination-scope
The 2016 formulation allows successive actions of one agent as well as actions of different agents, and allows an internal medium. Its medium must be affected by the relevant actions and perceptible as conditions for further action; this is the later author’s stated scope.
Mechanism
#feedback-forms
The 2016 account distinguishes responses that reduce perceived deviations from responses that amplify an emerging activity. These negative- and positive-feedback descriptions identify possible coordination dynamics; they do not specify one universal response law or guarantee convergence.
Application
#aco-instantiation-boundary
In the inspected ACO formulation, ants read and write local numerical pheromone information that influences later decisions. Pheromone deposition, usually evaporation, private ant memory and problem-dependent transition rules belong to that implementation framework; they are not universal requirements of the broader coordination mechanism.
⚠️ Curated Limitations & Negative Results
- The original Grassé 1959 full text was not inspected. Its reported definition and historical role are mediated by the later sources and publisher metadata. The 2016 author’s broader formulation is not presented as independently verified original wording or a priority finding.
- The 2016 display P(action|condition)>P(action) is not by itself a causal-identification criterion. For example, a common cause can determine both indicators and produce this inequality without the condition causing the action. The selected identity requires the actual mediating action–trace–response relationship.
- Shared access and aligned goals alone do not prove convergence. In a hand example, two agents observing the same real state x each propose the individually corrective increment −x; simultaneous application gives x_next=−x and a two-cycle for x≠0. Scheduling and response magnitudes require additional assumptions.
- The general mechanism does not fix a complete algorithm: trace representation, visibility, persistence, write conflicts, response timing, agent capabilities and stopping conditions depend on an implementation. A model must specify these before claims about its behavior or executable reproduction can be assessed.
- Broad source statements about perfect robustness, no wasted effort or unlimited scale are not adopted as universal guarantees. Access to a medium can have communication, storage, contention or processing costs. Indirect information exchange is not absence of information exchange or a guarantee of zero overhead.
- Pheromone reinforcement, evaporation, rewards, optimization objectives and learning are not mandatory for every instance. Positive feedback can also concentrate behavior prematurely. ACO, AntNet, telephone routing and other particular applications need their own evidence and do not make the general mechanism a reinforcement-learning task.
- Evidence scope is the complete ten-page 2016 article and the defining first six pages of the 36-page 1999 article, with selected method visuals. No complete audit of all cited biological experiments, all ACO variants, universal theorem, author implementation or benchmark reproduction is asserted.
Artificial Intelligence Ontology (AIO) Grounding
Explicit AIO Snapshot Assessment: No Match
Rationale: The candidate specifies how action-produced state changes mediate subsequent activity. The inspected AIO targets instead denote a whole network, neural layer, learning task, mathematical rule or application. None supplies a sufficiently informative counterpart for this general coordination mechanism without changing its scope; application-level mappings are not inherited automatically.
Search Scope: Current identity assessment covers 769 native and 442 AIO records at the merged cohort 344 base on 15 October 2026, with ignored and hidden files included in all four checkout searches. The pinned AIO commit is 397851c547d1c4d8f06ef184f63c1d40bc86937a. The retained source-specific mapping assessment examined the nine exact targets below and their 98 incident edges, alongside the separately documented native-neighbor context. Current identity collection and historical source/AIO reading retain distinct scopes. No informative mapping was established for this general concept in those inspected contexts; absence beyond this pinned vocabulary is not asserted.
| Examined AIO Candidate | Rejection Reason |
|---|---|
AIO:Network
|
The target denotes a system of interconnected entities; the candidate is a coordination mechanism that can operate within systems, rather than a specified network. General application alone does not establish a useful mapping of this subject. |
AIO:Layer
|
A deep-learning representation component is not the environment-mediated coordination mechanism. |
AIO:MachineLearningTask
|
Learning methods from data is not required by the mechanism. |
AIO:MathematicalFunction
|
No single input-output mathematical rule defines the general mechanism; implementations can use many different response functions. |
AIO:MetaLearning
|
Learning from metadata about machine-learning experiments is not inherent. |
AIO:ReinforcementLearning
|
An environment trace can affect later behavior without a learned policy, reward signal or exploration/exploitation objective. Pheromone reinforcement terminology does not supply that task identity. |
AIO:TrainingStrategy
|
The pinned class concerns preprocessing/training machine-learning models, which the mechanism does not require. |
AIO:UnsupervisedLearning
|
Learning patterns from unlabeled data is not inherent. |
AIO:Clustering
|
Clustering may be an application; grouping objects by similarity does not define the general coordination mechanism. |
Evidence Sources & Primary Literature
2 Sources| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
heylighen2016
|
Francis Heylighen
|
primary paper
full_text
|
ID: https://doi.org/10.1016/j.cogsys.2015.12.002
sha256:eca3052c7b21…
|
dorigo1999
|
Marco Dorigo, Gianni Di Caro, Luca M. Gambardella
|
primary paper
full_text
|
ID: https://doi.org/10.1162/106454699568728
sha256:7c6a5a0df5dd…
|
Curation Provenance
Curator: Codex
Method: ai_assisted
Model: GPT-6
Created: 2026-10-15T16:28:32Z
Updated: 2026-10-15T16:28:32Z
Source File:
data/curated/Stigmergy.yaml