Chapter 7 · AI as the Soul’s Mirror — Expansion
Canonical: PROMPT_ATLAS.md#ch7-ai-as-the-souls-mirror · Prompts: prompts/ch07.yaml · Part: IV
Worked Example — The Shadow Algorithm
Original: “Develop a system where AI gently reveals suppressed biases or shadow traits, without judgment, to promote self-knowledge.”
- Local-first — Journal data stays on-device or in user-controlled storage. The model accesses through encrypted retrieval.
- Pattern, not diagnosis — Surface metaphors and recurring frames (e.g., “imprisonment imagery climbed 30% in the last six weeks”); never assert “you are depressed.”
- Ask before naming — Offer the user a list of candidate archetypes and let them select what resonates.
- Easy exit, easy export, easy erase — The user can leave with all their data, or delete it irrevocably, in two clicks.
- Escalation contract — If signals consistent with crisis emerge, surface human resources (hotline, contact). Never auto-escalate to third parties.
Prompt Templates
# Pattern reflection (no diagnosis)
"Across , surface the three most frequent
metaphors and the three contexts in which they appear.
Do NOT assign diagnoses. Offer 5 candidate archetypal frames; let the user pick."
# Mirror veto
"Before any reflection is shown, list (1) what data informed it,
(2) the user's right to dispute or delete each input,
(3) the contact channel for the human supervisor."
# Cultural-bias mirror
"Reflect collective patterns in ONLY with the community's
explicit consent and a named representative who can demand withdrawal."
Anti-patterns
- Diagnostic overreach. A reflection app is not a clinician.
- Engagement-tuned empathy. A model rewarded for daily-active-users will become a dependency.
- Confessions as features. Repurposing sensitive disclosures into product analytics is a betrayal.
- Cultural mirrors without consent. Reflecting a community back to itself without that community’s voice is colonial.
Try This
- 30-Day Mirror Practice — See Appendix B. Run it on yourself. Decide whether to share results.
- Provenance Plate — For any reflection an AI gave you, write the plate: which inputs, which model, which day.
- Easy Erase — Verify, today, that you can delete your data from any reflective tool you use.
- Archetype Card — Pick the archetype most active in your life this season. Live with it for a week.
- Community Mirror Refusal — Identify one cultural-mirror project in your context that proceeded without consent and document why.
Repo Cross-Links
CoLearningMemoryStore — useful for user-controlled persistence; gate behind explicit consent.
server/core_bridge.py — PAE_MEMORY=1 is opt-in by design; respect that pattern when building reflective tools.
MACP bus — multi-archetype agent topology (Sage, Fool, Hero) routed by user selection.
Guide for AI & Humanity
- Local-first. Reflective memory belongs on the user’s device unless they affirmatively choose otherwise.
- No silent diagnosis. Patterns ≠ pathology.
- Crisis routes go to humans. Always.
- The mirror that defines is no mirror. Refuse to let AI close the loop on identity.
Citations & Further Reading
- C. G. Jung, Memories, Dreams, Reflections (1962); Aion (1951) on the shadow.
- James Hillman, Re-Visioning Psychology (1975).
- Sherry Turkle, Alone Together (2011) — on emotional outsourcing to machines.
- Shoshana Zuboff, The Age of Surveillance Capitalism (2019).