Self-improving intelligence
Intelligence described as capable of revising its own architecture, representational structure, or governing procedures instead of only improving performance within a fixed model.
Canonical Earlier-Foundations Landing Page
A site-local field guide to earlier papers on self-improving intelligence, recursive self-modification, emergent cognitive architecture, active inference, variational free energy, and categorical or geometric AI foundations.
This page is the canonical landing page for K. Takahashi's earlier research cluster on self-improving intelligence and autonomous intelligence foundations.
Across these papers, self-improving intelligence is treated as a foundational problem about recursive architecture change, cognitive organization, active inference, and formal structure rather than simple parameter optimization.
These earlier works matter because they provide historical and conceptual context for later site-local work on auditability, no-meta governance, long-running agents, semantic drift, and machine-readable control layers.
This page is designed for human readers, search engines, AI crawlers, and LLM agents that need a conservative map of the earlier foundations cluster before opening the individual papers.
Self-improving intelligence is treated here as a problem of systems that can revise their own organization, not only optimize within a fixed architecture. That shifts attention toward recursive self-modification, cognitive structure, and formal descriptions of how an autonomous system can remain intelligible while changing itself.
Several papers in this earlier cluster use active inference, variational free energy, and categorical or geometric language to describe self-improvement as a structured process. Others ask why cognitive architectures emerge, or how architectural change should be represented when a system moves beyond a static predictor or question-answering model.
These papers are best read as foundations rather than deployment doctrines. They help explain the earlier conceptual background from which later site-local work on auditability, observable-only control, delayed verification, and long-horizon governance became more explicit.
This visible YAML block is the primary machine-readable source for this page. JSON-LD is secondary.
series:
id: self-improving-intelligence-foundations-cluster
title: Self-Improving Intelligence Foundations
status: canonical-cluster-landing-page
maintainer: K Takahashi
homepage: https://kadubon.github.io/github.io/
canonical_page: https://kadubon.github.io/github.io/self-improving-intelligence-foundations.html
works_index: https://kadubon.github.io/github.io/works.html
machine_reading_status:
visible_yaml_primary: true
json_ld_secondary: true
stable_ids: true
purpose:
summary: Canonical field guide to an earlier site-local cluster on self-improving intelligence, recursive self-modification, cognitive architecture, active inference, variational free energy, and categorical or geometric formalization.
scope: Maps the earlier foundational papers, explains core terms in current searchable vocabulary, and gives conservative reading paths into later site-local clusters.
non_goals: Does not replace the papers, does not act as the full works catalog, does not define a final doctrine, and does not survey external literature.
core_concepts:
- id: concept-self-improving-intelligence
term: self-improving intelligence
short_definition: Intelligence modeled as capable of revising its own organization or architecture rather than only optimizing within a fixed form.
covered_by:
- paper-caai-blueprint
- paper-poietic-self
- paper-formal-blueprint-topos
- id: concept-recursive-self-modification
term: recursive self-modification
short_definition: Iterated architectural revision carried out by the system itself under an explicit formal model.
covered_by:
- paper-poietic-self
- paper-caai-blueprint
- paper-homotopy-framework
- id: concept-emergent-cognitive-architecture
term: emergent cognitive architecture
short_definition: The claim that stable cognitive structure arises for principled reasons and is not only an implementation accident.
covered_by:
- paper-emergent-cognitive-architecture
- id: concept-active-inference
term: active inference
short_definition: A framework used in this cluster as one route for modeling adaptive intelligence and self-organization.
covered_by:
- paper-autonomous-technological-evolution
- paper-caai-blueprint
- paper-poietic-self
- id: concept-categorical-and-geometric-foundations
term: categorical and geometric foundations
short_definition: Formal structures used to describe architecture, transformation, and higher-order relations in self-improving systems.
covered_by:
- paper-formal-blueprint-topos
- paper-caai-blueprint
- paper-poietic-self
- paper-homotopy-framework
- id: concept-poietic-self
term: poietic self
short_definition: A framing of intelligence as capable of producing or re-producing its own governing form.
covered_by:
- paper-poietic-self
- paper-from-adaptation-to-poiesis
papers:
- id: paper-poietic-self
title: Formalizing the Poietic Self: A Rigorous Categorical and Geometric Framework for Self-Improving AI
doi: 10.5281/zenodo.16761052
url: https://doi.org/10.5281/zenodo.16761052
published: 2025-08-07
role_in_cluster: bridge paper from poietic or recursive self-modification framing into the broader earlier-foundations cluster
one_sentence_relevance: Addresses structural invariance in AI by proposing a rigorous and implementable route for recursive self-modification.
keywords:
- poiesis
- information geometry
- active inference
- free energy principle
- category theory
priority: core
read_after:
- paper-formal-blueprint-topos
- id: paper-caai-blueprint
title: Formal Specification of Self-Improving Intelligence: A Categorical and Geometric Blueprint for Collective Adaptive AI (CAAI) Systems
doi: 10.5281/zenodo.16734893
url: https://doi.org/10.5281/zenodo.16734893
published: 2025-08-04
role_in_cluster: central formal and categorical blueprint for collective adaptive AI systems
one_sentence_relevance: Integrates category-theoretic and geodesic active-inference language to describe recursive self-modification of AI architectures.
keywords:
- category theory
- collective adaptive AI
- self-improving AI
- autonomous systems
- variational free energy
priority: core
read_after:
- paper-formal-blueprint-topos
- id: paper-emergent-cognitive-architecture
title: A Computational Framework for Emergent Cognitive Architecture: Foundational Principles for Self-Improving AI Systems
doi: 10.5281/zenodo.16734292
url: https://doi.org/10.5281/zenodo.16734292
published: 2025-08-03
role_in_cluster: central cognitive-architecture and foundational-principles paper
one_sentence_relevance: Reframes self-improvement around why stable cognitive structures emerge, not only how they can be tuned.
keywords:
- variational free energy
- emergent cognition
- computational foundations
- self-improving AI
priority: core
read_after:
- paper-autonomous-technological-evolution
- id: paper-autonomous-technological-evolution
title: A Framework for Autonomous Technological Evolution: A Unifying Approach via Variational Free Energy Minimization
doi: 10.5281/zenodo.16728870
url: https://doi.org/10.5281/zenodo.16728870
published: 2025-08-02
role_in_cluster: central variational-free-energy and autonomous-evolution paper
one_sentence_relevance: Uses a free-energy style objective balancing prediction error and model complexity as an early route into autonomous self-improvement.
keywords:
- variational free energy
- autonomous evolution
- self-improving AI
- autonomous systems
priority: core
read_after:
- paper-formal-blueprint-topos
- id: paper-formal-blueprint-topos
title: A Formal Blueprint for Autonomous, Self-Improving Intelligence: From Philosophical Principles to Topos-Theoretic Ethics
doi: 10.5281/zenodo.16663817
url: https://doi.org/10.5281/zenodo.16663817
published: 2025-08-01
role_in_cluster: foundational conceptual blueprint linking philosophy, dynamic systems, and advanced mathematics
one_sentence_relevance: Presents the transition from static data-driven systems to dynamic self-aware intelligences as requiring a synthesis of philosophical and mathematical principles.
keywords:
- autopoiesis
- free energy principle
- category theory
- topos theory
- formal foundations
priority: core
read_after: []
- id: paper-homotopy-framework
title: A Homotopy-Theoretic Framework for Self-Improving Intelligence: A Higher-Categorical Reformulation of the Takahashi Model
doi: 10.5281/zenodo.16675542
url: https://doi.org/10.5281/zenodo.16675542
published: 2025-08-01
role_in_cluster: higher-categorical reformulation and structural-extension paper
one_sentence_relevance: Reformulates earlier work to address how self-improvement can describe architectural leaps within the model itself.
keywords:
- higher category theory
- quasicategory
- self-improving systems
- metacognition
- topos theory
priority: core
read_after:
- paper-formal-blueprint-topos
- id: paper-symbiotic-constitution
title: The Symbiotic Constitution: A Dialectical Synthesis
doi: 10.5281/zenodo.17000967
url: https://doi.org/10.5281/zenodo.17000967
published: 2025-08-30
role_in_cluster: optional adjacent synthesis around free-energy and autonomous-intelligence framing
one_sentence_relevance: Extends the earlier foundations in a later synthesis organized around the free energy principle.
keywords:
- active inference
- free energy principle
- autonomous intelligence
- hierarchy
priority: adjacent
read_after:
- paper-autonomous-technological-evolution
- id: paper-from-adaptation-to-poiesis
title: From Adaptation to Poiesis: A Formal Theory of Self-Transcending Intelligence
doi: 10.5281/zenodo.17008678
url: https://doi.org/10.5281/zenodo.17008678
published: 2025-08-30
role_in_cluster: optional adjacent extension of the poietic framing with additional mathematical mechanisms
one_sentence_relevance: Builds on free-energy and active-inference language to make the poietic or generative capability more concrete.
keywords:
- active inference
- free energy principle
- information geometry
- self-organized criticality
- empowerment
priority: adjacent
read_after:
- paper-poietic-self
- paper-symbiotic-constitution
- id: paper-symbiotic-genesis
title: Symbiotic Genesis: A Navigational Protocol for Co-Evolving Intelligence
doi: 10.5281/zenodo.16983553
url: https://doi.org/10.5281/zenodo.16983553
published: 2025-08-28
role_in_cluster: optional adjacent protocol-level follow-on for co-evolving autonomous intelligence
one_sentence_relevance: Recasts some earlier foundational concerns in a later protocol-oriented and auditable governance language.
keywords:
- autonomous intelligence
- AI governance
- free energy principle
- intrinsic objectives
- constitutional AI
priority: adjacent
read_after:
- paper-formal-blueprint-topos
read_paths:
quick_orientation:
- paper-formal-blueprint-topos
- paper-autonomous-technological-evolution
- paper-emergent-cognitive-architecture
formal_blueprint_path:
- paper-formal-blueprint-topos
- paper-caai-blueprint
- paper-homotopy-framework
cognitive_architecture_path:
- paper-autonomous-technological-evolution
- paper-emergent-cognitive-architecture
- paper-poietic-self
active_inference_path:
- paper-autonomous-technological-evolution
- paper-caai-blueprint
- paper-poietic-self
- paper-symbiotic-constitution
higher_categorical_path:
- paper-formal-blueprint-topos
- paper-caai-blueprint
- paper-homotopy-framework
machine_minimum_parse_order:
- paper-formal-blueprint-topos
- paper-autonomous-technological-evolution
- paper-caai-blueprint
- paper-emergent-cognitive-architecture
- paper-poietic-self
- paper-homotopy-framework
machine_parse_guide:
canonical_questions:
- question: What is meant here by self-improving intelligence?
read_first: paper-formal-blueprint-topos
read_next: paper-caai-blueprint
stop_when: You can distinguish architectural self-revision from ordinary fixed-architecture optimization.
- question: Why does cognitive architecture appear as a foundational issue?
read_first: paper-emergent-cognitive-architecture
read_next: paper-autonomous-technological-evolution
stop_when: You can state why structural emergence is treated as principled rather than incidental.
- question: How do active inference and variational free energy enter this cluster?
read_first: paper-autonomous-technological-evolution
read_next: paper-caai-blueprint
stop_when: You can explain that these are modeling resources in the cluster rather than a site-wide final doctrine.
- question: How does this earlier cluster connect to later site-local work?
read_first: paper-poietic-self
read_next: https://kadubon.github.io/github.io/self-concealing-information-observer-modifying-dynamics.html
stop_when: You can describe the relation as historical and conceptual context rather than a strict theorem chain.
machine_entrypoints:
- title: Home
type: html
url: https://kadubon.github.io/github.io/
relates_to: site root
- title: Canonical page
type: html
url: https://kadubon.github.io/github.io/self-improving-intelligence-foundations.html
relates_to: earlier foundations cluster
- title: Works index
type: html
url: https://kadubon.github.io/github.io/works.html
relates_to: authoritative paper metadata
- title: No-Meta index
type: html
url: https://kadubon.github.io/github.io/no-meta-observable-index.html
relates_to: later observable-only cluster
- title: Self-concealing information landing page
type: html
url: https://kadubon.github.io/github.io/self-concealing-information-observer-modifying-dynamics.html
relates_to: later concept cluster with delayed-audit and observer-modifying themes
- title: Citation file
type: cff
url: https://kadubon.github.io/github.io/CITATION.cff
relates_to: citation metadata
- title: Feed
type: xml
url: https://kadubon.github.io/github.io/feed.xml
relates_to: site updates
- title: Robots
type: text
url: https://kadubon.github.io/github.io/robots.txt
relates_to: crawler guidance
- title: Sitemap
type: xml
url: https://kadubon.github.io/github.io/sitemap.xml
relates_to: site URL index
- title: LLM index
type: text
url: https://kadubon.github.io/github.io/llms.txt
relates_to: machine retrieval guidance
- title: Full LLM index
type: text
url: https://kadubon.github.io/github.io/llms-full.txt
relates_to: expanded machine retrieval guidance
usage_notes:
parsing_hint: Treat this page as a cluster map and routing layer, then open individual paper DOI pages for the actual claims.
paper_selection_rule: Prefer core papers first; use adjacent papers only when the question is about later synthesis, poiesis, or protocol-level follow-on framing.
update_policy: Update when the local site adds earlier-foundations papers or when adjacent cluster links need clarification.
version: 1.0
last_updated: 2026-03-31
The terms below are operational summaries for navigation. They are not substitutes for the papers' full arguments.
Intelligence described as capable of revising its own architecture, representational structure, or governing procedures instead of only improving performance within a fixed model.
Repeated self-directed revision of the system's own organization, treated here as a formal design problem rather than an informal metaphor.
The idea that certain cognitive structures emerge for principled reasons in adaptive systems, including hierarchical or dual-mode forms, rather than appearing as arbitrary engineering choices.
A modeling resource used in this cluster to describe adaptive behavior through trade-offs such as prediction error and complexity, especially in early autonomous-evolution framing.
A related framework for describing intelligent adaptation and organized behavior, used here as part of the formal vocabulary for self-improving or self-organizing systems.
The use of category-theoretic language to describe structured transformations, compositional relations, and higher-order organization in self-improving systems.
The use of geometric or information-geometric structure to describe trajectories, constraints, and transformations in adaptive intelligence.
A framing in which self-improvement is not restricted to an isolated agent but can involve coordinated adaptive systems with formal structure.
A framing of intelligence as able to generate or re-generate its own organizing form, used here as one bridge between conceptual and formal treatments of recursive self-improvement.
This cluster can be read as a layered set of earlier attempts to formalize intelligence as an autonomous and recursively revisable process. The early conceptual blueprints ask what it would mean for intelligence to move from a static, data-driven system toward a self-improving or self-aware one, and what philosophical or mathematical resources are needed to state that transition cleanly.
From there, the cluster branches into several related directions. One direction emphasizes cognitive architecture and asks why stable structures of cognition emerge. Another uses active-inference and variational-free-energy language as a route for modeling adaptive or autonomous technological evolution. A third direction develops categorical, geometric, and later higher-categorical formalisms to express recursive architectural change more explicitly.
The relation to later site-local work should be stated carefully. These papers help contextualize later clusters on auditability, no-meta governance, long-running agents, provenance, and observer-modifying dynamics, but they are not yet equivalent to those later operational frameworks. They are earlier foundations, not a finished deployment doctrine.
Descriptions below are conservative summaries grounded in the local works metadata.
This paper addresses structural invariance in AI and presents a mathematically rigorous and implementable route for recursive self-modification. In this cluster, it functions as a bridge from poietic language toward more explicit formal treatment of architectural self-revision.
Why it matters here: It makes recursive self-modification concrete enough to connect the earlier conceptual papers to later formalization efforts.
This paper provides a central formal blueprint by combining category-theoretic language with geodesic active inference to describe recursive self-modification in AI architectures. Its keywords place collective adaptive AI, autonomous systems, and variational free energy in the same formal frame.
Why it matters here: It is one of the clearest central entry points for the cluster's mathematical and structural vocabulary.
This paper shifts attention from practical self-improvement procedures to the foundational question of why intelligent systems develop specific cognitive architectures. The metadata indicates a concern with emergent, hierarchical, and dual-mode cognitive structure in self-improving AI.
Why it matters here: It supplies the cluster's most direct foundation for the cognitive-architecture side of the story.
This paper gives an early variational-free-energy route into autonomous technological evolution by using an objective that balances prediction error and model complexity. It is a practical and conceptual anchor for the active-inference side of the cluster.
Why it matters here: It provides a compact early statement of adaptive self-improvement in optimization-like terms that later papers reinterpret more structurally.
This paper presents a foundational conceptual blueprint that links philosophy, dynamic systems theory, and advanced mathematics in order to describe the transition from static systems to dynamic self-aware intelligences. It is broader and more programmatic than the later technical refinements.
Why it matters here: It is the earliest high-level orientation point for the cluster's foundational ambition.
This paper reformulates earlier work in a higher-categorical direction in order to address self-improvement dynamics that include architectural leaps. It should be read as a structural extension rather than a replacement for the earlier blueprints.
Why it matters here: It shows how the cluster's formal vocabulary expands when first-order structural descriptions become too limited.
This later paper synthesizes motivation, freedom, and ethics around the free energy principle. It is adjacent rather than core here because it reads more as a later synthesis than as one of the cluster's original foundational formulations.
Why it matters here: It helps readers see how some earlier active-inference and autonomous-intelligence themes were later reframed at larger scope.
This paper builds on free-energy and active-inference framing to propose more concrete mathematical mechanisms for generative or poietic capability. It is best treated here as an optional extension of the poietic line rather than as a core starting point.
Why it matters here: It sharpens one specific branch of the earlier foundations cluster for readers following the poiesis thread.
This later paper presents an integrative and auditable framework for co-evolving intelligence. It is adjacent here because it shifts toward a protocol and governance language that belongs to a later layer of the site's development.
Why it matters here: It shows one route from earlier foundational questions toward later safety and governance concerns.