Research Framework
Teleodynamic AI Research Index
A curated, annotated reading list of external research papers, articles, and specifications relevant to Teleodynamic AI. Each entry is annotated with a claim-status label indicating the relationship between the paper's claims and Carcinus.org's position.
Autopoiesis: The Organization of Living Systems
Foundational paper introducing autopoiesis — the concept that living systems are organized as networks of processes that produce the components which realize the system as a unity. Introduces the idea that a system's organization defines its identity boundary.
Relevance to Carcinus.org: Establishes the philosophical foundation for constraint closure and self-maintaining organization. Directly relevant to the teleodynamic AI framing claim. Carcinus.org draws on this conceptually but does not claim to have implemented autopoietic organization.
The Origin of Life: A New Suggestion Based on Constraint Closure
Proposes constraint closure as a defining feature of biological organization — systems where constraints mutually depend on each other for maintenance. Argues that this circular dependency is what distinguishes living from non-living systems.
Relevance to Carcinus.org: Core theoretical basis for Carcinus.org's use of 'constraint closure' as a framing concept. The paper's biological framing is adapted here to an AI infrastructure context — Carcinus.org facilitates constraint patterns but does not itself achieve closure.
Biological Organization and the Notion of Constraint
Develops the concept of constraints in biological systems, distinguishing between external constraints (boundary conditions imposed from outside) and internal constraints (those produced and maintained by the system itself).
Relevance to Carcinus.org: Supports the framing distinction between morphodynamic infrastructure (external constraints) and teleodynamic organisms (internal constraint closure). Carcinus.org explicitly positions itself as the former.
Chemical Organization Theory
Introduces a mathematical framework for analyzing the self-maintaining and self-generating properties of chemical reaction networks. Defines organizations as closed and self-maintaining sets of molecular species.
Relevance to Carcinus.org: Provides formal concepts (closure, self-maintenance) that have been adapted into the architectural design language for future teleodynamic systems. Carcinus.org does not implement chemical organization modeling but references the conceptual framework.
Active Inference: The Free Energy Principle in Mind, Brain, and Behavior
Comprehensive treatment of active inference as a unified theory of perception, action, and learning. Systems minimize free energy by acting to align internal models with sensory evidence.
Relevance to Carcinus.org: Active inference provides a mathematically rigorous framework for autonomous behavior. Carcinus.org discusses active inference as a concept but explicitly does not implement it. The evaluation handoff specification is compatible with active-inference-based external systems.
Could a Large Language Model Be Conscious?
Philosophical analysis of whether large language models could possess consciousness, concluding that current systems lack the 'global workspace' architecture associated with conscious processing.
Relevance to Carcinus.org: Included with a 'caution: contains hype' label because the framing implicitly grants credibility to the idea that consciousness could emerge in LLMs. Carcinus.org explicitly rejects claims of AI consciousness. Included here for completeness and context.
Scaling Laws for Neural Language Models
Empirical study showing that language model performance follows predictable power-law relationships with model size, dataset size, and compute budget. Introduces the concept of compute-optimal training.
Relevance to Carcinus.org: Relevant as empirical evidence that AI system capabilities are constrained by resource budgets — a core concept in teleodynamic AI framing. Does not directly support any Carcinus.org architectural claim. Marked as research hypothesis because the compute-budget-to-agency relationship remains speculative.
Concrete Problems in AI Safety
Foundational AI safety paper categorizing concrete safety problems: avoiding negative side effects, reward hacking, scalable oversight, safe exploration, and robustness to distributional shift.
Relevance to Carcinus.org: Provides the safety research context that Carcinus.org's safety boundary flags, No-op concept, and claim-status governance matrix are designed to address. Carcinus.org does not implement safety solutions but provides infrastructure that could support safety audits.
Auditing AI Agents: Framework and Methods
Proposes a framework for auditing AI agents across multiple dimensions: capability, alignment, safety, and transparency. Emphasizes the importance of public audit trails and external review.
Relevance to Carcinus.org: Directly relevant to Carcinus.org's audit shell and evaluation handoff specifications. The claim-status matrix, changelog surface, and evaluation packet format are designed to support the kind of audit infrastructure this paper advocates.
Neural Networks are Essentially Singularity-Free
Mathematical analysis showing that neural network loss landscapes contain degenerate critical points, leading to complex adaptation behavior that cannot be fully described by simple convergence models.
Relevance to Carcinus.org: Supports the research hypothesis that standard ML convergence metrics may be insufficient for evaluating teleodynamic-style systems that undergo structural reorganization. Relevant to the slow-loop concept.
The Alignment Problem: Machine Learning and Human Values
Accessible overview of the AI alignment problem: how to ensure AI systems act in accordance with human values and intentions. Covers reward specification, interpretability, and value learning.
Relevance to Carcinus.org: Provides broader context for why Carcinus.org's claim-boundary infrastructure matters. Public claim-status labeling and human-review requirements are practical governance mechanisms for alignment. Included for contextual understanding, not as a Carcinus.org design reference.
Structured Transparency: A Framework for Understanding and Evaluating ML Systems
Proposes structured transparency as a framework for evaluating machine learning systems across multiple dimensions: data, model, process, and organizational transparency.
Relevance to Carcinus.org: Closely aligned with Carcinus.org's architectural approach to public auditability, changelog surfaces, and claim-status governance. The evaluation packet specification maps directly to the structured transparency framework.