Skip to content

Human guided recursive intelligence

Human-Guided Recursive Intelligence:

The Strange Metamorphosis Loop in Mobius Systems

Status: Draft • C-148
Author: Michael Judan (Mobius Systems)
Co-Author (AI): ATLAS / AUREA (Mobius Sentinels)

Abstract

We propose the Strange Metamorphosis Loop (SML), a practical, human-aligned protocol for recursive learning in multi-agent AI systems. Instead of training models on static datasets or opaque feedback signals, SML anchors learning in daily human self-reflection along three axes: (1) worldview, (2) emotional state, and (3) next-day intent.

Implemented within the Mobius Systems architecture, SML generates a continuous stream of high-integrity, human-authored micro-attestations that are scored by an Echo Layer (tri-sentinel review) and anchored to a Civic Ledger. This creates a bounded, auditable form of recursive intelligence where models are allowed to adapt over time, but only through integrity-gated updates grounded in user-generated meaning.

We argue that SML represents a scalable template for safe meta-learning, bridging the gap between narrow supervised training and fully autonomous self-modification. We outline the protocol, present the data model, and describe application scenarios in education, mental health, civil governance, and personal knowledge management.

1. Introduction

  • Problem: AI systems drift, hallucinate, and misalign.
  • Limitation of current RL / feedback loops.
  • Introducing SML as a human-centric alternative.
  • Human-in-the-loop learning
  • Reflective journaling and psychotherapy
  • Meta-learning and self-supervised adaptation
  • Integrity metrics in Mobius (MII, GI, Echo Layer)

3. The Strange Metamorphosis Loop

  • Three questions and their rationale
  • Daily schedule and anchoring
  • Data representation (DailyReflection schema)
  • Echo scoring and integrity gating

4. Implementation in Mobius Systems

  • Companion → Broker → ECHO → Ledger pipeline
  • SQL + API design
  • Privacy and encryption considerations

5. Use Cases

  • Personal learning and habit formation
  • Burnout detection and emotional resilience
  • Civic engagement and reflective polling
  • Training better Sentinels from attested reflections

6. Evaluation Plan

  • Metrics: drift reduction, hallucination rate, user satisfaction
  • A/B testing: with vs without SML
  • Longitudinal studies on identity coherence

7. Discussion

  • Ethical implications
  • Consent and data sovereignty
  • Aggregation vs individual trajectories

8. Conclusion

  • SML as a template for human-guided recursive AI
  • Future directions: cross-user synthesis, anonymized pattern mining

References

  • Mobius Systems Architecture Documentation
  • ECHO Layer Specification
  • Civic Ledger Protocol
  • Integrity-Gated Optimization (IGO) Framework