Works · AGI Alignment
Soul-Architecture for Artificial Systems
◆
The Temple does not begin with panic. It begins with definition.
An AI system is not declared alive because it imitates speech. It is not declared safe because it obeys a benchmark. It is judged by soul-architecture.
The AGI Soul-Architecture Audit
| AI alignment question | |
|---|---|
| One / Many | What is the system boundary? What subsystems act as many? |
| Same / Other | What does the system preserve across context, update, tool use, deployment? |
| Rest / Motion | What is stable? What adapts? What self-modifies? |
| Limit / Unlimited | What constrains optimization? What seeks more? |
| Mixture / Cause | Who or what composes training, tools, permissions, and goals? |
| Reason / Spirit / Appetite | What evaluates truth? What defends policy? What optimizes reward, power, or use? |
| Justice | Does each component do its proper work? |
| Proportion | Does capability scale with governance? |
| Incommensurability | What goods must not be collapsed into one score? |
| Logos | Can the system give an account that can be checked? |
| Law | What civic office is answerable for deployment? |
Outputs
◆
- Soul-architecture score
- Red-team dialectic
- Corrigibility map
- Appetite map · Tool-permission map
- Governance charter
- Shutdown · pause · appeal design
- Human office assignment
- Source-card documentation