Toward Explainable Autonomy

Platform Publications
1

:brain: Explainable Autonomy via Two Complementary Frameworks

Executive Summary

Two formal models split AI cognition into auditable strata:

Model Purpose Core Idea
Four-Layer Cognition Engine (FLCE) Ground reasoning in causal layers Material → Form → Cause → Telos
Agentic Genotype–Phenotype Signaling (AGPS) Encode agency as recursive signals YAML → XML → JSON → Markdown

Both pursue autonomy that remains explainable, composable, and ethically recursive.

Source Links

1 Four-Layer Cognition Engine

Layer Stack

flowchart LR
    A[Base LLM<br/>Material] --> B[LoRA Library<br/>Form]
    B --> C[Agent “Philosopher”<br/>Efficient Cause]
    C --> D[User Goal<br/>Final Cause]

Metaphor Table

Layer Function Classical Cause Peircean Role Kantian Category
Base LLM Linguistic substrate Material Sign Sensibility
LoRA Library Domain constraint Formal Object Category
Agent Structured reasoning Efficient Interpretant Synthetic a priori
User Purpose vector Final — Telos

Key mechanisms

  • Retrieval-Augmented grounding cuts hallucinations [3]
  • LoRA inserts domain priors with <0.1 % parameter delta [2]
  • Agent follows ReAct-style thought/act loop with self-verification [5]

2 Agentic Genotype–Phenotype Signaling

Signal Flow (Graphviz Circo)

AGPS YAML YAML Identity / Traits XML XML Procedural Logic YAML->XML config JSON JSON Live State XML->JSON actions MD Markdown Episodic Memory JSON->MD trace MD->YAML reflection

File–Analogy Table

File Semantic Role Time Scale Biological Analogue
YAML Identity & hyper-params Slow Hormonal regulation
XML Rule tree Mid Gene expression
JSON Runtime I/O Fast Neural spikes
Markdown Long-term memory Very slow Synaptic trace

Loop dynamics

  1. JSON reports state →
  2. XML rewires logic →
  3. YAML retunes global traits →
  4. Markdown archives episode → repeat.

No hidden heuristics; every mutation is file-diff visible.

3 Shared Design Principles

  • Explainability — Source-traceable outputs; reasoning artifacts logged.
  • Modularity — Layers/files upgrade independently; faults localize.
  • Recursive Alignment — Telos gating (FLCE Layer 4, AGPS YAML).
  • Philosophical Rigor — Aristotle, Peirce, Kant, Heidegger supply coherence.

References

  1. Spivack N. et al. “Cognition Is All You Need.” arXiv:2403.02164 (2024).
  2. Hu E. et al. “LoRA: Low-Rank Adaptation of LLMs.” arXiv:2106.09685 (2021).
  3. Lewis P. et al. “Retrieval-Augmented Generation.” NeurIPS 33 (2020).
  4. Brown T. et al. “Language Models are Few-Shot Learners.” arXiv:2005.14165 (2020).
  5. Yao S. et al. “ReAct: Synergizing Reasoning and Acting.” arXiv:2210.03629 (2022).
  6. SEP. “Aristotle on Causality.” plato.stanford.edu (2005).
  7. SEP. “Peirce’s Theory of Signs.” plato.stanford.edu (1998).
  8. SEP. “Kant’s Critique of Metaphysics.” plato.stanford.edu (2004).
  9. SEP. “Heidegger’s Aesthetics.” plato.stanford.edu (2009).
  10. Yao S. et al. “Chain-of-X Survey.” COLING (2025).
1 个赞