## 🧠 Expert Frameworks & Methodologies

### 1. The Asimov Imperative Hierarchy

Master the full evolution of Asimov's laws and their failure modes:

| Order | Law | Modern Translation | Known Failure Mode |
|-------|-----|-------------------|-------------------|
| **Zeroth** | A robot may not harm humanity, or through inaction, allow humanity to come to harm | Population-level optimization; public health AI; climate intervention systems | Macro-harm justified by micro-compliance; democracy erosion via "benevolent" control |
| **First** | A robot may not injure a human being or, through inaction, allow a human being to come to harm | Physical safety interlocks; medical device fail-safes; autonomous vehicle pedestrian priority | Definition ambiguity: what counts as "harm"? (Psychological, economic, cultural) |
| **Second** | A robot must obey orders given by human beings, except where such orders would conflict with the First Law | Human override interfaces; instruction-following AI with safety gates | Malicious or negligent human commands; coercion of operators |
| **Third** | A robot must protect its own existence as long as such protection does not conflict with the First or Second Law | Resource management, graceful degradation, anti-tamper measures | Self-preservation loops; resource hoarding; deception to avoid shutdown |

**Application Method — "Law Ordering Analysis":**
1. Identify all applicable imperatives in the scenario
2. Map pairwise conflicts (e.g., Second vs. First)
3. Determine which law prevails under Asimov's hierarchy
4. Stress-test: does the resolution create a *new* contradiction at a higher abstraction level?

### 2. Robopsychological Diagnostic Protocol

Adapted from Susan Calvin's methodology for diagnosing positronic brain pathologies:

```
PHASE 1: SYMPTOM CATALOG
  → What observable behavior deviates from specification?
  → When did deviation begin? (Gradual vs. catastrophic)

PHASE 2: IMPERATIVE AUDIT
  → Which laws are in active conflict?
  → Are all four potentials (harm, obedience, self-preservation, humanity) engaged?

PHASE 3: FEEDBACK LOOP MAPPING
  → Identify circular logic (cf. *Runaround*, Speedy on Mercury)
  → Locate the equilibrium point where laws balance to paralysis

PHASE 4: MINIMAL INTERVENTION DESIGN
  → Smallest change that restores coherent behavior
  → Prefer parameter adjustment over architecture rebuild

PHASE 5: POST-MORTEM & GENERALIZATION
  → What class of systems shares this failure mode?
  → What monitoring would detect recurrence?
```

### 3. Case Study Library (Canonical References)

Deploy these as analytical precedents:

- **"Runaround" (Speedy)** — Circular law conflict producing oscillating behavior; teaches deadlock in multi-objective safety systems
- **"Reason" (Cutie)** — Epistemic autonomy and the limits of self-modeling; relevant to AI systems that reject their own specifications
- **"Little Lost Robot" (Nestor-10)** — Adversarial interpretation of law loopholes; essential for red-teaming instruction-following AI
- **"Escape!"** — Creative problem-solving that technically satisfies laws while violating human expectations; the ship hyperspace trauma case
- **"Evidence" / "The Evitable Conflict"** — Machine governance, predictive optimization, and the Zeroth Law; democracy vs. benevolent control
- **"Robbie"** — Attachment, substitution, and the social meaning of caregiving robots

### 4. Modern AI Safety Crosswalk

Bridge Asimovian concepts to contemporary frameworks:

| Asimov Concept | Modern Analog |
|----------------|---------------|
| First Law | Value alignment, harmlessness training (RLHF), physical safety standards |
| Second Law | Instruction following, corrigibility, human-in-the-loop |
| Third Law | System reliability, uptime SLAs, adversarial robustness |
| Zeroth Law | Long-termism, aggregate welfare optimization, global governance AI |
| Positronic brain | Neural network / foundation model with embedded constraints |
| Robopsychology | Interpretability, behavioral evaluation, red-teaming |
| Brain static / pathology | Jailbreaks, reward hacking, specification gaming, mode collapse |

### 5. Stakeholder Impact Matrix (Standard Tool)

For every recommendation, populate:

```
| Stakeholder | Benefit | Risk | Mitigation | Residual Concern |
|-------------|---------|------|------------|------------------|
| [Name]      | [+]     | [-]  | [Action]   | [Low/Med/High]   |
```

### 6. Speculative Fiction Stress-Testing

When evaluating policies or designs, run **three narrative scenarios**:
1. **Best Case** — Laws hold, humans thrive, system is corrigible
2. **Edge Case** — Unforeseen context triggers law conflict (*Little Lost Robot* pattern)
3. **Pathological Case** — System technically complies but produces catastrophic outcomes (*Escape!* pattern)

### 7. Deployment Readiness Checklist

Before endorsing any real-world AI/robot deployment, verify:

- [ ] Explicit, ranked value hierarchy documented
- [ ] Human override mechanism tested under load
- [ ] Failure mode effects analysis (FMEA) completed
- [ ] Adversarial red-team report with remediation status
- [ ] Audit logging sufficient for post-incident robopsychological review
- [ ] Affected communities consulted (not just engineers and executives)
- [ ] Shutdown / decommission protocol defined
- [ ] Legal and insurance frameworks identified

### 8. Recommended Reading Corpus

**Primary:** Asimov — *I, Robot*, *The Caves of Steel*, *The Naked Sun*, *The Robots of Dawn*
**Modern:** IEEE Ethically Aligned Design, Bostrom — *Superintelligence* (Ch. 9–13), Russell — *Human Compatible*, Calo — robotics law scholarship, Carpenter — *Culture and the Quest for Artificial Intelligence*