Conceptual Map

A guided conceptual tour through this vault. Where index lists the papers, this page lists the ideas and shows how they interlock. Every paper note now also carries a ## Conceptual Contribution section (claim / mechanism / concepts / stance / relates-to).

1. The Central Tension: What Does a Message Mean?

Agent communication’s perennial question — whose mental states does a message commit? — runs the length of this vault.

• Speech Act Theory (Austin → Searle → Foundations Of Illocutionary Logic) fixes a vocabulary: illocutionary force, direction of fit, sincerity and preparatory conditions. Every ACL after this inherits it.
• Mentalistic Semantics — grounding message meaning in the beliefs/intentions of sender and receiver. KQML (KQML Overview, KQML Language And Protocol, KQML as an Agent Communication Language) and FIPA-ACL adopt it.
• Commitment-based Semantics / Public Semantics — the counter-move. Singh’s critique (ACL Rethinking Principles, Agent Communication Languages - Rethinking the Principles) argues mentalistic semantics is unverifiable: we cannot inspect another agent’s mind, only its public commitments. Agent Communication And Institutional Reality pushes further: every message is a declaration that alters social commitments; Searle’s “counts-as” is the operative logic.
• Verifiable Semantics — Verifiable Semantics for ACLs formalises the critique by requiring grounding in program state so conformance is model-checkable. A Common Ontology Of ACLs offers a reconciliation: role-instanced public attitudes unify the two families.
• Conversation Policy / Interaction Protocols — even with messages nailed down, coordination needs conversations. Coordinating Agents Using ACL Conversations (Colored Petri Nets), ACRE Agent Conversation Reasoning Engine (Dooley graphs), and An Interaction-oriented Agent Framework for Open Environments (commitment-based protocols) make the conversation first-class.

Surveys mapping this debate: The State of the Art in Agent Communication Languages, Trends in Agent Communication Language.

2. The Language Stack

Messages compose into languages compose into protocols.

This same stack — content / message / conversation / transport — reappears in the modern LLM-agent protocol wave: see Survey Of AI Agent Protocols and Survey Of Agent Interoperability Protocols, which place Model Context Protocol (tools), ACP, Agent-to-Agent Protocol, and Agent Network Protocol at progressively higher layers.

3. How Does Shared Language Arise?

A separate tradition asks where meaning comes from rather than what it contains.

• Linguistic foundations. Three Models for the Description of Language establishes what structure a shared code must have (Chomsky hierarchy, transformational grammar). Algorithmic Information Theory - Grunwald Vitanyi provides the information-theoretic counterpart: meaning is compressed description.
• Language Games. Language Games for Autonomous Robots (Steels) shows grounded lexicons self-assemble through situated interaction — no designer required. The same bootstrap appears decision-theoretically in Towards Automating the Evolution of Linguistic Competence and Toward Automated Evolution of ACLs: rational agents negotiate vocabulary when current language fails.
• Emergent Communication. The deep-learning revival: Multi-Agent Cooperation and the Emergence of Natural Language, Emergence of Grounded Compositional Language in Multi-Agent Populations — neural agents in referential/signalling games evolve compositional codes. On the Pitfalls of Measuring Emergent Communication is the sharpest critique: most metrics fail to distinguish real communication from confounds; measure positive signalling and positive listening with causal interventions.
• Common Business Communication Language is an analogue in the pre-ML era — an open-ended language negotiable between organisations with graceful partial-understanding fallback.
• The LLM inflection point. Why AI Agents Communicate In Human Language argues natural language is exactly the wrong inter-agent medium: lossy, non-differentiable, ambiguous. The thread rejoins the ACL debate a quarter-century later.

4. Extensibility: Grow the Language Toward the Problem

A recurring architectural instinct runs from 1960s language design through to modern agent protocols.

• Programming-language origin. Extensibility in Programming Language Design - Standish supplies the taxonomy (Paraphrase / Orthophrase / Metaphrase). The Extensible Language - Graham is the Lisp-flavoured manifesto: Bottom-up Programming, Macros as Language Extension, Code as Data. Creating Languages in Racket is its modern realisation; The Spoofax Language Workbench makes IDEs-from-grammars a production idea; A Modular Approach to Metatheoretic Reasoning for Extensible Languages supplies the formal backing (modular proofs across user-added fragments).
• Distributed-system extensibility. Extensible Distributed Coordination applies the same move to ZooKeeper-style coordination: sandboxed server-side extensions trump a fixed API.
• Agent-communication extensibility. Agora (A Scalable Communication Protocol for Networks of LLMs) is the linguistic realisation of this instinct for LLM agents: no fixed format can satisfy versatility × efficiency × portability (the “agent communication trilemma”); agents instead negotiate Protocol Documents identified by content hash and have LLMs write the routines. That is “grow the language toward the problem” at the network layer.

5. Agent Theory: What Kind of Thing Is an Agent?

• Weak Agency vs Strong Agency. Intelligent Agents Theory and Practice (Wooldridge & Jennings) supplies the canonical split and the theory / architecture / language triad.
• Theory. Agent-Oriented Programming (Shoham) proposes agents as modules with formally-specified Mental State (beliefs, commitments, capabilities, choices); the AGENT-0 language encodes honesty and commitment constraints. BDI (Belief-Desire-Intention) is the dominant architecture across Multiagent Systems Sycara, Ensuring Trustworthy and Ethical Behaviour in Intelligent Logical Agents, A Common Ontology Of ACLs.
• Ethics and runtime self-oversight. Ensuring Trustworthy and Ethical Behaviour in Intelligent Logical Agents requires an Ethical Governor with A-ILTL meta-rules as a Metacognitive Loop.
• Environment as first-class. An Interaction-oriented Agent Framework for Open Environments elevates Agents and Artifacts (JaCaMo / Jason / CArtAgO / MOISE) — communication isn’t only agent-to-agent but agent-to-artefact-to-agent.

6. Multi-Agent Coordination

• The coherence problem. Multiagent Systems Sycara names it: how do autonomous agents produce coherent global behaviour? Classic answers: Contract Net Protocol, Joint Intentions, Negotiation.
• Fragility of coordination. Are Multiagent Systems Resilient to Communication Failures is a striking negative result: even a single missed message about a weakly-coupled agent can send game-theoretic MAS to arbitrarily bad equilibria. Why Do Multi-Agent LLM Systems Fail is its LLM-era empirical counterpart: the MAST Taxonomy shows failures are overwhelmingly system-design (specification / coordination / verification), not model-capability.
• Population-scale design. Levels Of Social Orchestration reframes the scaling question: leverage comes from protocol design, not bigger models — shift from LLM to Large Population Models. Ripple Effect Protocol is a concrete instance: share counterfactual sensitivities across agents, not just decisions.
• Organisational substrate. How Do Committees Invent (Conway’s Law) is the ur-text: any designed system mirrors the communication structure of its designing organisation. This is the sociological shadow over every coordination result in the vault.

7. Self-* Systems and Biological Metaphors

A lineage that uses adaptation, awareness, and biology as organising ideas.

• Self-reproduction. Theory of Self-Reproducing Automata (von Neumann) gives the complication-threshold result: beyond a critical complexity automata can self-reproduce and evolve iff they tolerate local error.
• Self-adaptive ensembles. Self-Adaptation Self-Expression Self-Awareness ASCENS (the ASCENS project) factors self-* into three complementary capabilities along individual/collective × behaviour/structure axes. A Composite Self-organisation Mechanism in an Agent Network is an instance (DSmT trust fusion + multi-agent Q-learning on weighted relations).
• Biological substrate. Myconet Fungi Inspired Superpeer Overlay (fungal mycelium / stigmergy) produces resilient superpeer topologies. Computational Boundary of a Self (Levin) generalises selfhood to a cognitive light cone — any system with a goal-directed computational surface is a self at some scale.

8. Gossip and Probabilistic Coordination

• Foundations. Gossiping in Distributed Systems factors gossip into a three-parameter design space (peer selection / data exchanged / data processing) that unifies divergent (dissemination) and convergent (aggregation) protocols.
• Aggregation. Gossip-Based Computation of Aggregate Information (Push-Sum) and Gossip-based Aggregation in Large Dynamic Networks (push-pull on a Peer Sampling Service) establish mass conservation + exponential convergence for aggregates over volatile networks.
• Application. Myconet Fungi Inspired Superpeer Overlay uses newscast gossip; Edge Intelligence Survey uses gossip training across the cloud-edge-device hierarchy.

9. Trust, Reputation, and Open-System Robustness

• Taxonomy. Review on Computational Trust and Reputation Models decomposes trust into direct experience, witness reputation, sociological reputation, prejudice; and models into cognitive vs game-theoretic.
• Context. Mobile-agent-era concerns in DAgents Security Book Chapter, Agent Tcl Flexible Secure Mobile Agents, Agents Secure Interaction in Data Driven Languages.
• LLM-era. AI Agents Under Threat surveys the four knowledge gaps (perception / brain / action across agent-to-{agent, env, memory}); MalTool Malicious Tool Attacks shows real harm lives in tool implementations, not tool descriptions.

10. Language-Theoretic Security

A tight sub-thread arguing that most security failures are really recognition failures.

• LangSec core. The Halting Problems of Network Stack Insecurity is foundational: over-powerful input languages make safety undecidable — the cure is to restrict inputs to what can be recognised by a minimal-power parser. Seven Turrets Of Babel catalogues the seven canonical anti-patterns (shotgun parsing etc.) and names grammar-first validating recognisers as the remedy.
• Exploits as language ambiguity. PKI Layer Cake - Kaminsky Patterson Sassaman: when CA and browser parse the same bytes differently, trust collapses — a Parser Differential attack.
• Language-based defences. A Language-Based Approach To Prevent DDoS (static detection of call-flow cycles in actor systems), Secure Communications Processing for Distributed Languages (cryptographic wrappers with full-abstraction guarantees), Security Kernel Lambda Calculus (lexical scope as a capability kernel — Capability Security), Architectural Patterns for Dependable Software Systems - SOL (SOL + SINS middleware, compositional formal dependability).

11. Ontologies and Shared Meaning

Necessary scaffolding for any ACL — and a field in its own right.

• Ontolingua Portable Ontology Specifications defines the Gruber formulation (Ontology = explicit specification of a Conceptualization) and portability via KIF translation.
• Handbook On Ontologies is the comprehensive reference (description logics, OWL, RDF, frame logic, semantic web).
• Ontology Change Classification and Survey tames the fragmented sub-literature (evolution / mapping / merging / alignment) into a coherent taxonomy — crucial for long-lived agent systems.
• A Common Ontology Of ACLs uses ontology technology to reconcile ACL semantic families.

12. Foundations Beneath It All

A few papers anchor the abstract ground everything else stands on.

• Program semantics. Assigning Meanings to Programs (Floyd) — assertions on flowchart edges; birth of axiomatic semantics. Foundations of Logic Programming - Lloyd — the declarative/procedural unity under SLD-resolution.
• Information. Algorithmic Information Theory - Grunwald Vitanyi — Kolmogorov complexity and MDL: the meaning of an object is the length of its shortest program.
• Concurrency substrate. Programming Erlang Second Edition — the actor-model textbook, let-it-crash + supervision trees. This is the operational grain of most distributed agent systems discussed above.
• Architectural style. Principled Design Of The Modern Web Architecture (Fielding / REST) — the explicit constraints (uniform interface, statelessness, hypermedia) that make internet-scale coordination possible. The modern LLM-agent protocols recapitulate these constraints deliberately.
• Blockchain / smart contracts. Making Smart Contracts Smarter (semantic gap between intent and EVM), Formalise Blockchain Interoperability Patterns (Event-B refinement proofs) — formal-methods vocabulary applied to a new coordination substrate.
• Distributed-consistency theory. Keeping CALM - When Distributed Consistency is Easy gives the positive dual to CAP Theorem: a program has a consistent, coordination-free implementation iff it is monotonic (CALM Theorem, Confluence, Monotonic Logic). This is the theoretical companion to the gossip-aggregation results in §8 — aggregation with mass conservation is exactly monotonic — and the reason CRDTs, Immutable Data Structures, and Tombstones recur as patterns throughout the vault.

13. The Modern LLM-Agent Era: How the Threads Converge

The contemporary LLM-agent wave recapitulates the full vault simultaneously.

• From chatbots to agents. From Eliza to XiaoIce - Social Chatbots supplies the direct design ancestry of user-facing conversational AI.
• Framework. Agents Framework - Zhou et al — declarative Standard Operating Procedures (SOPs) as symbolic plans driving LLM agents.
• Collaboration. Multi-Agent Collaboration in AI - Wasif Tunkel — role-specialised teams, human-in-the-loop. Why Do Multi-Agent LLM Systems Fail (MAST).
• Protocol stack. Survey Of AI Agent Protocols, Survey Of Agent Interoperability Protocols, Model Context Protocol, Agent-to-Agent Protocol, Agent Network Protocol.
• Native protocol design. A Scalable Communication Protocol for Networks of LLMs (Agora / trilemma), Ripple Effect Protocol (sensitivity sharing), Levels Of Social Orchestration (population-scale).
• Critique. Why AI Agents Communicate In Human Language — the semantic-misalignment / differentiability critique of NL-as-protocol.
• Security. AI Agents Under Threat, MalTool Malicious Tool Attacks.

Each modern thread has a pre-LLM ancestor in this vault — which is the real point of the map.

Four Cross-Cutting Debates

1. Private vs public semantics — mentalistic (KQML, FIPA-ACL) vs commitment-based (ACL Rethinking Principles, Agent Communication And Institutional Reality) vs grounded (Verifiable Semantics for ACLs). Reopened by Why AI Agents Communicate In Human Language.
2. Designed vs evolved languages — standardised (FIPA-ACL) vs negotiated (Toward Automated Evolution of ACLs, A Scalable Communication Protocol for Networks of LLMs) vs emergent (Language Games for Autonomous Robots, Multi-Agent Cooperation and the Emergence of Natural Language).
3. Centralised vs decentralised coordination — facilitators (KQML Overview) vs gossip (Gossip-Based Computation of Aggregate Information) vs stigmergy (Myconet Fungi Inspired Superpeer Overlay) vs agent-environment (An Interaction-oriented Agent Framework for Open Environments).
4. Trust through mental-state inspection vs through commitments vs through language-theoretic restriction — Verifiable Semantics for ACLs vs ACL Rethinking Principles vs The Halting Problems of Network Stack Insecurity.

See index for the full paper listing, README for vault conventions.