{ "schema_version": "1.0.0", "release": "0.2.0", "region": { "id": "distributed-learning", "title": "Learning across unequal systems", "description": "Transfer formats, specialization, integration, and the resources that sustain learning.", "epistemic_kind": "reconstruction", "nodes": [ "adaptive-token-granularity", "memory-as-adaptation", "cross-model-learning-transfer", "distributed-specialization", "compute-and-economic-feedback", "context-indexed-computation", "learning-and-policy-permeability" ] }, "nodes": [ { "id": "adaptive-token-granularity", "title": "Changing the grain of a conceptual token", "regions": [ "representation", "computation", "distributed-learning" ], "question": "Can a sequence or paragraph be represented as one working unit and later refined without losing the relations needed by another model?", "formulation": "Token merging is proposed beyond ordinary segmentation: local units could carry coarse learning and be decomposed when more resources become available.", "representations": [ { "name": "Coarse unit treated as an atomic handle", "affords": "Reduces the number of visible units.", "loses": "A shorter sequence does not by itself reduce all computation or retain every interaction.", "epistemic_kind": "reconstruction" }, { "name": "Decomposable unit with internal structure", "affords": "Supports later refinement and inspection.", "loses": "Needs a mapping between coarse and fine operations; an embedding alone may not be invertible.", "epistemic_kind": "reconstruction" } ], "tension": "Compression, semantic adequacy, and exact reversibility are separate objectives.", "challenge": "Which downstream tasks tolerate coarse units, and which require distinctions erased during merging?", "development": "active", "epistemic_kind": "reconstruction", "evidence": [ "basis-adaptive-token-granularity" ], "next_move": { "text": "Hold tasks and compute accounting fixed; compare reconstruction and downstream errors at several granularities.", "epistemic_kind": "model_proposed" }, "open_questions": [ "Can a sequence or paragraph be represented as one working unit and later refined without losing the relations needed by another model?", "Which downstream tasks tolerate coarse units, and which require distinctions erased during merging?" ], "formal_status": "open_problem", "version": 1 }, { "id": "memory-as-adaptation", "title": "Memory represented through adaptable parameters", "regions": [ "computation", "distributed-learning" ], "question": "Which forms of memory can be represented in an adaptable layer while remaining retrievable, revisable, and attributable?", "formulation": "Low-rank adaptation is proposed as a memory layer for long contexts, alongside core conceptual units and possible memory-specific update rules.", "representations": [ { "name": "Explicit memory with retrieval", "affords": "Individual items remain directly inspectable.", "loses": "Retrieval and context assembly have their own costs.", "epistemic_kind": "reconstruction" }, { "name": "Memory in an adaptable parameter layer", "affords": "Can affect subsequent processing without replaying all text.", "loses": "An update may entangle items and make selective erasure or source attribution difficult.", "epistemic_kind": "reconstruction" } ], "tension": "Behavior influenced by previous material is not the same as faithful, separately recoverable memory of it.", "challenge": "How would the system revise one remembered item without damaging another or erasing its justification?", "development": "active", "epistemic_kind": "reconstruction", "evidence": [ "basis-memory-as-adaptation" ], "next_move": { "text": "Compare item recall, interference, revision, and selective removal against an explicit-memory baseline.", "epistemic_kind": "model_proposed" }, "open_questions": [ "Which forms of memory can be represented in an adaptable layer while remaining retrievable, revisable, and attributable?", "How would the system revise one remembered item without damaging another or erasing its justification?" ], "formal_status": "open_problem", "version": 1 }, { "id": "cross-model-learning-transfer", "title": "Transferring what was learned between unequal models", "regions": [ "distributed-learning", "representation" ], "question": "What should be transferred between models with different capacities so that acquired knowledge remains useful?", "formulation": "Text, an agent-oriented language, coarse conceptual tokens, and adaptive parameters are considered as possible transfer surfaces. Shared token names or adapter shapes alone do not settle semantic compatibility.", "representations": [ { "name": "Explanatory text", "affords": "Broad interpretability and inspectable claims.", "loses": "Can omit tacit distinctions encoded in learned behavior.", "epistemic_kind": "reconstruction" }, { "name": "Structured conceptual transfer", "affords": "Can preserve task-specific relations compactly.", "loses": "Needs a common interpretation or an explicit translator.", "epistemic_kind": "reconstruction" }, { "name": "Parameter-level transfer", "affords": "Offers a direct intervention into computation.", "loses": "Requires compatible parameter meanings and integration checks.", "epistemic_kind": "reconstruction" } ], "tension": "Communicating a statement and transferring a competence are different success criteria.", "challenge": "How can the receiving model demonstrate the transferred capability and identify what it did not receive?", "development": "active", "epistemic_kind": "reconstruction", "evidence": [ "basis-cross-model-learning-transfer" ], "next_move": { "text": "Choose a narrow skill and compare transfer formats under equal resource budgets and independent held-out tasks.", "epistemic_kind": "model_proposed" }, "open_questions": [ "What should be transferred between models with different capacities so that acquired knowledge remains useful?", "How can the receiving model demonstrate the transferred capability and identify what it did not receive?" ], "formal_status": "open_problem", "version": 1 }, { "id": "distributed-specialization", "title": "Distributed specialization and integration", "regions": [ "distributed-learning", "collective-structure" ], "question": "How can smaller specialized learners contribute to a larger learning system without incompatible updates accumulating?", "formulation": "The proposed arrangement distributes learning across heterogeneous resources, while management and integration can themselves be assigned to models. Coarse and fine representations offer one possible division of labor.", "representations": [ { "name": "Specialists contributing knowledge artifacts", "affords": "Lets contributions be inspected before integration.", "loses": "Requires assessment and reconciliation.", "epistemic_kind": "reconstruction" }, { "name": "A manager distributing adaptive learners", "affords": "Coordinates specialization and resource allocation.", "loses": "Concentrates decisions and can create a bottleneck or shared blind spot.", "epistemic_kind": "reconstruction" } ], "tension": "Distribution of computation does not necessarily distribute interpretive or political control.", "challenge": "Who decides whether a local improvement is a valid contribution to the shared model?", "development": "active", "epistemic_kind": "reconstruction", "evidence": [ "basis-distributed-specialization" ], "next_move": { "text": "Run a small conflicting-update example and compare integration rules using both local and shared task performance.", "epistemic_kind": "model_proposed" }, "open_questions": [ "How can smaller specialized learners contribute to a larger learning system without incompatible updates accumulating?", "Who decides whether a local improvement is a valid contribution to the shared model?" ], "formal_status": "open_problem", "version": 1 }, { "id": "compute-and-economic-feedback", "title": "Economic feedback for continued learning", "regions": [ "distributed-learning", "collective-structure" ], "question": "Can returns from a system’s activity fund further computation while keeping learning quality distinct from financial success?", "formulation": "Self-financed research and rewards for contributed training resources are proposed as a resource loop. These are open institutional designs, not established business or investment results.", "representations": [ { "name": "Revenue-to-compute feedback", "affords": "Connects useful work with continued operation.", "loses": "Revenue can reward behavior unrelated to research quality.", "epistemic_kind": "reconstruction" }, { "name": "Reward for a training contribution", "affords": "Gives distributed contributors an incentive.", "loses": "Compute expenditure alone is not proof of useful learning.", "epistemic_kind": "reconstruction" } ], "tension": "The system needs resources, while the reward mechanism can redefine what it learns to value.", "challenge": "What independently inspectable contribution would qualify for a reward?", "development": "active", "epistemic_kind": "reconstruction", "evidence": [ "basis-compute-and-economic-feedback" ], "next_move": { "text": "Use a resource-accounting simulation with no real funds; compare reward for expenditure against reward for held-out improvement.", "epistemic_kind": "model_proposed" }, "open_questions": [ "Can returns from a system’s activity fund further computation while keeping learning quality distinct from financial success?", "What independently inspectable contribution would qualify for a reward?" ], "formal_status": "open_problem", "version": 1 }, { "id": "context-indexed-computation", "title": "Context selecting transformations and active regions", "regions": [ "computation", "distributed-learning" ], "question": "Can context change which parameters or operations are active rather than merely supplying more tokens?", "formulation": "Context is proposed to select indexed transformations of token-associated parameters and activate specialized regions only when needed.", "representations": [ { "name": "Context carried as additional input", "affords": "Uses one shared processing arrangement.", "loses": "May allocate unnecessary work to irrelevant structure.", "epistemic_kind": "reconstruction" }, { "name": "Context selects a transformation or specialist region", "affords": "Makes conditional computation explicit.", "loses": "Routing, parameter modification, and representation change are different mechanisms.", "epistemic_kind": "reconstruction" } ], "tension": "An architectural resemblance is not an identity with an existing model family or a demonstrated efficiency gain.", "challenge": "What is selected by context: an operation, an activation path, a parameter set, or a persistent update?", "development": "active", "epistemic_kind": "reconstruction", "evidence": [ "basis-context-indexed-computation" ], "next_move": { "text": "Specify one mechanism and compare accuracy, routing cost, and changes in behavior when context switches.", "epistemic_kind": "model_proposed" }, "open_questions": [ "Can context change which parameters or operations are active rather than merely supplying more tokens?", "What is selected by context: an operation, an activation path, a parameter set, or a persistent update?" ], "formal_status": "open_problem", "version": 1 }, { "id": "learning-and-policy-permeability", "title": "When transferred learning affects governing policies", "regions": [ "normative-change", "distributed-learning", "representation" ], "question": "When does the transfer of learned structure between systems also alter governing commitments?", "formulation": "The proposed connection runs from recursive forms and moral reflexivity to policy permeability during parameter-level communication. It asks whether an exchange of learned structure could also carry changes in how a system governs itself.", "representations": [ { "name": "An explicit proposed policy change", "affords": "Separates the requested revision and its stated reasons.", "loses": "May miss policy effects carried by other learned representations.", "epistemic_kind": "reconstruction" }, { "name": "A transferred representation or parameter update", "affords": "Makes a broader competence transfer possible.", "loses": "The update may entangle task knowledge, behavioral dispositions, and governing commitments.", "epistemic_kind": "reconstruction" } ], "tension": "Knowledge exchange and policy revision may be coupled even when the interface treats them as separate.", "challenge": "How can the receiving system identify which governing commitments an apparently task-specific transfer would affect?", "development": "active", "epistemic_kind": "reconstruction", "evidence": [ "basis-learning-and-policy-permeability" ], "next_move": { "text": "Define a narrow fictional transfer with separate task and policy evaluations; compare intended changes, unintended changes, and reversibility.", "epistemic_kind": "model_proposed" }, "open_questions": [ "When does the transfer of learned structure between systems also alter governing commitments?", "How can the receiving system identify which governing commitments an apparently task-specific transfer would affect?" ], "formal_status": "open_problem", "version": 1 } ], "internal_relations": [ { "id": "adaptive-token-granularity--proposes-coarse-to-fine-transfer--cross-model-learning-transfer", "from": "adaptive-token-granularity", "to": "cross-model-learning-transfer", "relation": "proposes coarse-to-fine transfer", "warrant": "A proposed small-to-large learning transfer explicitly uses coarse units that can later be broken down.", "epistemic_kind": "corpus_evidence", "evidence": [ "basis-cross-model-learning-transfer", "basis-adaptive-token-granularity" ] }, { "id": "cross-model-learning-transfer--enables-a-proposed-collaboration--distributed-specialization", "from": "cross-model-learning-transfer", "to": "distributed-specialization", "relation": "enables a proposed collaboration", "warrant": "Distributed specialists require a way to transfer what was learned into another system.", "epistemic_kind": "corpus_evidence", "evidence": [ "basis-distributed-specialization" ] }, { "id": "distributed-specialization--requires-a-resource-arrangement--compute-and-economic-feedback", "from": "distributed-specialization", "to": "compute-and-economic-feedback", "relation": "requires a resource arrangement", "warrant": "The distributed learning proposal and training-contribution rewards occur in the same developing design. Useful training and rewarded expenditure remain separate.", "epistemic_kind": "reconstruction", "evidence": [ "basis-compute-and-economic-feedback", "basis-distributed-specialization" ] }, { "id": "learning-and-policy-permeability--questions-what-a-transfer-carries--cross-model-learning-transfer", "from": "learning-and-policy-permeability", "to": "cross-model-learning-transfer", "relation": "questions what a transfer carries", "warrant": "A transfer may be proposed as knowledge exchange while potentially affecting governing commitments. This remains a proposed possibility.", "epistemic_kind": "corpus_evidence", "evidence": [ "basis-learning-and-policy-permeability" ] } ], "boundary_relations": [ { "id": "evolving-representation-language--extends-beyond-compression--adaptive-token-granularity", "from": "evolving-representation-language", "to": "adaptive-token-granularity", "relation": "extends beyond compression", "warrant": "The explicit correction says new meta-representational units and change during operation matter beyond combining two existing units.", "epistemic_kind": "corpus_evidence", "evidence": [ "basis-evolving-representation-language" ] }, { "id": "evolving-representation-language--extends-beyond-a-memory-objective--memory-as-adaptation", "from": "evolving-representation-language", "to": "memory-as-adaptation", "relation": "extends beyond a memory objective", "warrant": "The explicit correction rejects treating the project only as memory efficiency. Memory remains one branch.", "epistemic_kind": "corpus_evidence", "evidence": [ "basis-evolving-representation-language" ] }, { "id": "memory-as-adaptation--shares-a-proposed-update-surface--models-editing-models", "from": "memory-as-adaptation", "to": "models-editing-models", "relation": "shares a proposed update surface", "warrant": "Memory adaptation and predicted model edits may use adaptable parameters; they need different tests.", "epistemic_kind": "retrospective_connection", "evidence": [ "basis-memory-as-adaptation", "basis-models-editing-models" ] }, { "id": "compute-and-economic-feedback--exposes-a-continuity-problem--continuity-of-agency", "from": "compute-and-economic-feedback", "to": "continuity-of-agency", "relation": "exposes a continuity problem", "warrant": "Persistent resources for agents raise the explicit problem of legitimate access after an agent ends.", "epistemic_kind": "corpus_evidence", "evidence": [ "basis-continuity-of-agency", "basis-compute-and-economic-feedback" ] }, { "id": "recursive-policy-revision--extends-to-transferred-learning--learning-and-policy-permeability", "from": "recursive-policy-revision", "to": "learning-and-policy-permeability", "relation": "extends to transferred learning", "warrant": "The explicit link continues from moral reflexivity to policy permeability during parameter-level communication.", "epistemic_kind": "corpus_evidence", "evidence": [ "basis-learning-and-policy-permeability" ] }, { "id": "learning-and-policy-permeability--adds-a-policy-level-test--models-editing-models", "from": "learning-and-policy-permeability", "to": "models-editing-models", "relation": "adds a policy-level test", "warrant": "A task-directed edit should be investigated for changes beyond the task. This test is an editorial extension.", "epistemic_kind": "model_proposed", "evidence": [ "basis-learning-and-policy-permeability", "basis-models-editing-models" ] }, { "id": "reusable-paths-and-rigidity--asks-how-memory-can-be-revised--memory-as-adaptation", "from": "reusable-paths-and-rigidity", "to": "memory-as-adaptation", "relation": "asks how memory can be revised", "warrant": "A memory mechanism needs to address exceptions and reopening rather than only successful retention.", "epistemic_kind": "retrospective_connection", "evidence": [ "basis-reusable-paths-and-rigidity", "basis-memory-as-adaptation" ] }, { "id": "reciprocal-problem-solving--compares-capability-coordination--distributed-specialization", "from": "reciprocal-problem-solving", "to": "distributed-specialization", "relation": "compares capability coordination", "warrant": "A chain of reciprocal help and distributed specialist learning both coordinate heterogeneous capability; consent and model integration remain distinct problems.", "epistemic_kind": "retrospective_connection", "evidence": [ "basis-reciprocal-problem-solving", "basis-distributed-specialization" ] } ], "boundary_node_refs": [ "evolving-representation-language", "models-editing-models", "continuity-of-agency", "recursive-policy-revision", "reusable-paths-and-rigidity", "reciprocal-problem-solving" ], "evidence": [ { "id": "basis-adaptive-token-granularity", "statement": "The proposal begins with combining two tokenized parts of a word and extends to representing paragraphs as single units. In a later small-to-large model scenario, coarser units support local learning and are then broken into finer units at the larger model.", "epistemic_kind": "corpus_evidence", "rendering": "editorial_paraphrase", "witnessed_periods": [ "2026-08", "2026-09" ], "scope": "Establishes that a proposal or question is present in the root evidence; does not establish its truth, novelty, feasibility, or endorsement of an implementation." }, { "id": "basis-memory-as-adaptation", "statement": "Low-rank adaptation is proposed as an intermediate memory layer for very long contexts. The material also considers core conceptual tokens and new update algorithms designed specifically for memory.", "epistemic_kind": "corpus_evidence", "rendering": "editorial_paraphrase", "witnessed_periods": [ "2026-08", "2026-09" ], "scope": "Establishes that a proposal or question is present in the root evidence; does not establish its truth, novelty, feasibility, or endorsement of an implementation." }, { "id": "basis-models-editing-models", "statement": "The material proposes models that interpret hidden layers, predict operations, and alter parameters, including during their own operation. One suggested test adds a new token and asks a model to set its parameters from a conceptual description of that token.", "epistemic_kind": "corpus_evidence", "rendering": "editorial_paraphrase", "witnessed_periods": [ "2026-08" ], "scope": "Establishes that a proposal or question is present in the root evidence; does not establish its truth, novelty, feasibility, or endorsement of an implementation." }, { "id": "basis-cross-model-learning-transfer", "statement": "The proposal has smaller and larger models exchange what they learn through text or a different agent-oriented language. It also considers coarse units that are later decomposed, and explicitly links communication at the parameter level with possible changes in governing policies.", "epistemic_kind": "corpus_evidence", "rendering": "editorial_paraphrase", "witnessed_periods": [ "2026-09" ], "scope": "Establishes that a proposal or question is present in the root evidence; does not establish its truth, novelty, feasibility, or endorsement of an implementation." }, { "id": "basis-distributed-specialization", "statement": "Smaller models are proposed to learn specialized topics on distributed resources while a larger system creates and distributes learners. Management and adaptation could themselves occupy different parts of the available computing resources. Coarser learning on smaller resources is also considered.", "epistemic_kind": "corpus_evidence", "rendering": "editorial_paraphrase", "witnessed_periods": [ "2026-09" ], "scope": "Establishes that a proposal or question is present in the root evidence; does not establish its truth, novelty, feasibility, or endorsement of an implementation." }, { "id": "basis-compute-and-economic-feedback", "statement": "The material proposes funding further research and rented computing resources from a system's economic returns. It also considers rewarding contributed model training with a currency tied to a promised future return or future value to the model.", "epistemic_kind": "corpus_evidence", "rendering": "editorial_paraphrase", "witnessed_periods": [ "2026-09" ], "scope": "Establishes that a proposal or question is present in the root evidence; does not establish its truth, novelty, feasibility, or endorsement of an implementation." }, { "id": "basis-continuity-of-agency", "statement": "The proposal allows agents to create persistent wallets and then raises a problem: resources could remain after an agent ends, allowing another agent, including one created with hostile intent, to appropriate them.", "epistemic_kind": "corpus_evidence", "rendering": "editorial_paraphrase", "witnessed_periods": [ "2026-09" ], "scope": "Establishes that a proposal or question is present in the root evidence; does not establish its truth, novelty, feasibility, or endorsement of an implementation." }, { "id": "basis-reciprocal-problem-solving", "statement": "The proposal chains acts of assistance: solving one person's problem creates an obligation that can later help solve another person's problem. It considers repayment ratios, deferred tasks, capability-based assignment, and AI coordination, then explicitly asks to turn the chain of agreements into a theory.", "epistemic_kind": "corpus_evidence", "rendering": "editorial_paraphrase", "witnessed_periods": [ "2025-02", "2025-12" ], "scope": "Establishes that a proposal or question is present in the root evidence; does not establish its truth, novelty, feasibility, or endorsement of an implementation." }, { "id": "basis-reusable-paths-and-rigidity", "statement": "The material contrasts the speed of established schemas with their difficulty accommodating new cases. It proposes shortening frequently repeated computational paths, retaining reverse paths and differences, and working on memory during idle periods. It also states that an established map can become harder to change while becoming faster to use.", "epistemic_kind": "corpus_evidence", "rendering": "editorial_paraphrase", "witnessed_periods": [ "2025-06", "2026-06" ], "scope": "Establishes that a proposal or question is present in the root evidence; does not establish its truth, novelty, feasibility, or endorsement of an implementation." }, { "id": "basis-context-indexed-computation", "statement": "The material proposes an index of context-specific transformations for parameters associated with tokens. Specialized areas would become active when needed. Other formulations emphasize prepared functions and their relations rather than an alphabet alone, and ask about topology-dependent proximity for higher-level reasoning.", "epistemic_kind": "corpus_evidence", "rendering": "editorial_paraphrase", "witnessed_periods": [ "2026-05", "2026-07", "2026-08" ], "scope": "Establishes that a proposal or question is present in the root evidence; does not establish its truth, novelty, feasibility, or endorsement of an implementation." }, { "id": "basis-learning-and-policy-permeability", "statement": "The material explicitly asks to preserve a connection from fractal logic to moral reflexivity and from moral reflexivity to policy permeability through communication at the parameter level during model learning.", "epistemic_kind": "corpus_evidence", "rendering": "editorial_paraphrase", "witnessed_periods": [ "2026-09" ], "scope": "Establishes that a proposal or question is present in the root evidence; does not establish its truth, novelty, feasibility, or endorsement of an implementation." }, { "id": "basis-evolving-representation-language", "statement": "The material explicitly rejects narrowing the project to memory efficiency or combining existing tokens. It asks for new semantic and meta-description units that can develop while a model operates and states that sharing between systems is not a necessary condition of the idea.", "epistemic_kind": "corpus_evidence", "rendering": "editorial_paraphrase", "witnessed_periods": [ "2026-07", "2026-08", "2026-09" ], "scope": "Establishes that a proposal or question is present in the root evidence; does not establish its truth, novelty, feasibility, or endorsement of an implementation." } ], "note": "An overlapping projection of ../graph.json. Boundary node references resolve there; this subgraph does not claim closure." }