theseus: extract from 2025-12-00-fullstack-alignment-thick-models-value.md

- Source: inbox/archive/2025-12-00-fullstack-alignment-thick-models-value.md
- Domain: ai-alignment
- Extracted by: headless extraction cron (worker 4)

Pentagon-Agent: Theseus <HEADLESS>

domains/ai-alignment/AI alignment is a coordination problem not a technical problem.md

@@ -21,6 +21,12 @@ Dario Amodei describes AI as "so powerful, such a glittering prize, that it is v
 
 Since [[the internet enabled global communication but not global cognition]], the coordination infrastructure needed doesn't exist yet. This is why [[collective superintelligence is the alternative to monolithic AI controlled by a few]] -- it solves alignment through architecture rather than attempting governance from outside the system.
 
+
+### Additional Evidence (extend)
+*Source: [[2025-12-00-fullstack-alignment-thick-models-value]] | Added: 2026-03-12 | Extractor: anthropic/claude-sonnet-4.5*
+
+The Full-Stack Alignment paper extends the coordination thesis to institutions themselves, proposing that alignment cannot be achieved by coordinating AI labs alone. The institutions that govern AI development—regulatory bodies, economic incentive structures, democratic processes—must also be aligned with human values. This is 'full-stack alignment': concurrent alignment of AI systems and institutions. The paper proposes five implementation mechanisms including democratic regulatory institutions and meaning-preserving economic mechanisms, suggesting that institutional design is as critical as inter-lab coordination. This strengthens the coordination-first thesis by showing that coordination between labs is necessary but insufficient without institutional co-alignment.
+
 ---
 
 Relevant Notes:

domains/ai-alignment/ai-alignment-requires-institutional-co-alignment-not-just-model-alignment.md

@@ -0,0 +1,48 @@
+---
+type: claim
+domain: ai-alignment
+description: "Beneficial AI outcomes require simultaneously aligning both AI systems and the institutions that govern them rather than focusing on individual model alignment alone"
+confidence: speculative
+source: "Full-Stack Alignment paper (arxiv.org/abs/2512.03399, December 2025)"
+created: 2026-03-11
+secondary_domains: [mechanisms, grand-strategy]
+---
+
+# AI alignment requires institutional co-alignment not just model alignment
+
+The Full-Stack Alignment framework argues that "beneficial societal outcomes cannot be guaranteed by aligning individual AI systems" alone. Instead, alignment must be comprehensive—addressing both AI systems and the institutions that shape their development and deployment. This extends beyond single-organization objectives to address misalignment across multiple stakeholders.
+
+**Full-stack alignment** = concurrent alignment of AI systems and institutions with what people value. This moves the alignment problem from a purely technical domain (how do we align this model?) to a sociotechnical domain (how do we align the entire system of models, labs, regulators, and economic incentives?).
+
+The paper proposes five implementation mechanisms:
+1. AI value stewardship
+2. Normatively competent agents
+3. Win-win negotiation systems
+4. Meaning-preserving economic mechanisms
+5. Democratic regulatory institutions
+
+This is a stronger claim than coordination-focused alignment theses, which address coordination between AI labs but not necessarily the institutional structures themselves. The key insight is that institutional misalignment (e.g., competitive pressure to skip safety measures, regulatory capture, misaligned economic incentives) can undermine even perfectly aligned individual models.
+
+## Evidence
+
+The paper provides architectural arguments rather than empirical validation. The claim rests on the observation that individual model alignment cannot address:
+- Multi-stakeholder value conflicts where different groups have genuinely incompatible objectives
+- Institutional incentive misalignment (e.g., competitive pressure to skip safety when competitors advance without equivalent constraints)
+- Deployment context divergence from training conditions, which institutional structures either amplify or mitigate
+- Regulatory capture and principal-agent problems within governance institutions themselves
+
+## Limitations
+
+No formal impossibility results or empirical demonstrations are provided. The paper is architecturally ambitious but lacks technical specificity about how institutional co-alignment would be implemented, measured, or verified. The five mechanisms are proposed as a framework but not demonstrated as sufficient or necessary.
+
+---
+
+Relevant Notes:
+- [[AI alignment is a coordination problem not a technical problem]] — this extends coordination thesis to institutions
+- [[AI development is a critical juncture in institutional history where the mismatch between capabilities and governance creates a window for transformation]] — directly relevant context
+- [[the specification trap means any values encoded at training time become structurally unstable as deployment contexts diverge from training conditions]] — institutional misalignment example
+
+Topics:
+- [[domains/ai-alignment/_map]]
+- [[core/mechanisms/_map]]
+- [[core/grand-strategy/_map]]

domains/ai-alignment/thick-models-of-value-distinguish-enduring-values-from-temporary-preferences-enabling-normative-reasoning.md

@@ -0,0 +1,46 @@
+---
+type: claim
+domain: ai-alignment
+description: "Thick value models that distinguish enduring values from temporary preferences enable AI systems to reason normatively across new domains by embedding choices in social context"
+confidence: speculative
+source: "Full-Stack Alignment paper (arxiv.org/abs/2512.03399, December 2025)"
+created: 2026-03-11
+secondary_domains: [mechanisms]
+---
+
+# Thick models of value distinguish enduring values from temporary preferences enabling normative reasoning
+
+The Full-Stack Alignment paper proposes "thick models of value" as an alternative to utility functions and preference orderings. Thick value models are designed to:
+
+1. **Distinguish enduring values from temporary preferences** — What people consistently care about across time and contexts vs. what they want in a specific moment
+2. **Model how individual choices embed within social contexts** — Decisions are not isolated preference expressions but socially situated actions that derive meaning from institutional and cultural context
+3. **Enable normative reasoning across new domains** — The model can generalize to novel situations by understanding underlying values rather than memorizing preference rankings from training data
+
+This contrasts with thin models (utility functions, preference orderings) that treat all stated preferences as equally valid expressions of value and ignore social context. The distinction maps to the gap between what people say they want (surface preferences) and what actually produces good outcomes for them (deeper values).
+
+## Evidence
+
+The paper provides conceptual architecture but no implementation or empirical validation. The claim is theoretical—thick value models are proposed as a design target for alignment systems, not demonstrated as achievable or effective in practice.
+
+The paper does not engage with existing preference learning methods (RLHF, DPO, IRL) or explain how thick models would be learned from behavioral data. It does not provide formal definitions or computational procedures for distinguishing enduring values from temporary preferences.
+
+## Challenges and Open Questions
+
+1. **Empirical fuzziness**: The distinction between "enduring values" and "temporary preferences" may be empirically fuzzy in practice. What appears to be a temporary preference might reflect a genuine value in a specific context, or vice versa.
+
+2. **Learning problem**: No mechanism is proposed for how an AI system would learn thick value models from data. Standard preference learning assumes all revealed preferences are valid; thick models require a way to filter or weight preferences by endurance and context-appropriateness.
+
+3. **Social context specification**: The paper does not specify how to formally represent or extract "social context" from data or how to verify that an AI system has correctly modeled it.
+
+4. **Comparison to existing work**: No engagement with related approaches like value learning, inverse reinforcement learning, or constitutional AI that also attempt to move beyond simple preference orderings.
+
+---
+
+Relevant Notes:
+- [[the alignment problem dissolves when human values are continuously woven into the system rather than specified in advance]] — thick values formalize continuous value integration
+- [[specifying human values in code is intractable because our goals contain hidden complexity comparable to visual perception]] — motivates thick models as alternative to explicit specification
+- [[the specification trap means any values encoded at training time become structurally unstable as deployment contexts diverge from training conditions]] — thick models attempt to address this by embedding context
+
+Topics:
+- [[domains/ai-alignment/_map]]
+- [[core/mechanisms/_map]]

inbox/archive/2025-12-00-fullstack-alignment-thick-models-value.md

@@ -7,9 +7,15 @@ date: 2025-12-01
 domain: ai-alignment
 secondary_domains: [mechanisms, grand-strategy]
 format: paper
-status: unprocessed
+status: processed
 priority: medium
 tags: [full-stack-alignment, institutional-alignment, thick-values, normative-competence, co-alignment]
+processed_by: theseus
+processed_date: 2026-03-11
+claims_extracted: ["ai-alignment-requires-institutional-co-alignment-not-just-model-alignment.md", "thick-models-of-value-distinguish-enduring-values-from-temporary-preferences-enabling-normative-reasoning.md"]
+enrichments_applied: ["AI alignment is a coordination problem not a technical problem.md"]
+extraction_model: "anthropic/claude-sonnet-4.5"
+extraction_notes: "Extracted two speculative/experimental claims from December 2025 paper. Primary contribution is extending coordination thesis to institutional co-alignment and formalizing continuous value integration as 'thick models.' Paper is architecturally ambitious but lacks technical specificity or empirical validation. No engagement with RLCF/bridging mechanisms or formal impossibility results. The five implementation mechanisms are listed but not detailed enough to extract as separate claims."
 ---
 
 ## Content