- TeX 96.7%
- HTML 2.5%
- Shell 0.5%
- Python 0.3%
| Filename | Latest commit message | Latest commit date |
|---|---|---|
| semantic_index | ||
| working | ||
| 1980s_math_fixes_AI_agent_collapse.json | ||
| 1980s_math_fixes_AI_agent_collapse.mp3 | ||
| 1980s_math_fixes_AI_agent_collapse.srt | ||
| 1980s_math_fixes_AI_agent_collapse.tsv | ||
| 1980s_math_fixes_AI_agent_collapse.txt | ||
| 1980s_math_fixes_AI_agent_collapse.vtt | ||
| accommodation-before-prediction.aux | ||
| accommodation-before-prediction.log | ||
| accommodation-before-prediction.out | ||
| accommodation-before-prediction.pdf | ||
| accommodation-before-prediction.tex | ||
| against-the-extinction-thesis.aux | ||
| against-the-extinction-thesis.log | ||
| against-the-extinction-thesis.out | ||
| against-the-extinction-thesis.pdf | ||
| against-the-extinction-thesis.tex | ||
| analyze.py | ||
| availability-before-residency.aux | ||
| availability-before-residency.log | ||
| availability-before-residency.out | ||
| availability-before-residency.pdf | ||
| availability-before-residency.tex | ||
| Calculus - diagrams.pdf | ||
| Calculus - diagrams.tex | ||
| Calculus - diagrams.txt | ||
| Calculus - main text - draft 01.pdf | ||
| Calculus - main text - draft 01.tex | ||
| Calculus - main text - draft 01.txt | ||
| Calculus - main text - old.pdf | ||
| Calculus - main text - old.tex | ||
| Calculus - main text - old.txt | ||
| Calculus - main text.pdf | ||
| Calculus - main text.tex | ||
| Calculus - main text.txt | ||
| Calculus_Chapter_Summaries.md | ||
| check-directory.sh | ||
| complexity-before-control.aux | ||
| complexity-before-control.log | ||
| complexity-before-control.out | ||
| complexity-before-control.pdf | ||
| complexity-before-control.tex | ||
| Convergence Before Autonomy - notes.txt | ||
| Convergence Before Autonomy.pdf | ||
| Convergence Before Autonomy.tex | ||
| Convergence Before Autonomy.txt | ||
| corrections.log | ||
| directory-map.html | ||
| divided-we-stand.aux | ||
| divided-we-stand.log | ||
| divided-we-stand.out | ||
| divided-we-stand.pdf | ||
| divided-we-stand.tex | ||
| exposure-before-evidence.aux | ||
| exposure-before-evidence.log | ||
| exposure-before-evidence.out | ||
| exposure-before-evidence.pdf | ||
| exposure-before-evidence.tex | ||
| extract-text.py | ||
| fix_transcription_errors.sh | ||
| Human in the Loop.pdf | ||
| Human in the Loop.tex | ||
| Human in the Loop.txt | ||
| Hyperpleonastic Legibility.pdf | ||
| Hyperpleonastic Legibility.tex | ||
| Hyperpleonastic Legibility.txt | ||
| Identity After Authenticity - draft 01.pdf | ||
| Identity After Authenticity - draft 01.tex | ||
| Identity After Authenticity - draft 01.txt | ||
| Identity After Authenticity.pdf | ||
| Identity After Authenticity.tex | ||
| Identity After Authenticity.txt | ||
| index.html | ||
| LICENSE | ||
| local-map.html | ||
| narrative-before-mechanism.aux | ||
| narrative-before-mechanism.log | ||
| narrative-before-mechanism.out | ||
| narrative-before-mechanism.pdf | ||
| narrative-before-mechanism.tex | ||
| README.md | ||
| Recursive Futarchy.tex | ||
| run-analysis.sh | ||
| scan-directory.sh | ||
| structure.txt | ||
| The Expiatory Gap - draft 00.docx | ||
| The Expiatory Gap - draft 01.pdf | ||
| The Expiatory Gap - draft 01.tex | ||
| The Expiatory Gap - draft 01.txt | ||
| The Expiatory Gap - draft 02.pdf | ||
| The Expiatory Gap - draft 02.tex | ||
| The Expiatory Gap - draft 02.txt | ||
| The Expiatory Gap.pdf | ||
| The Expiatory Gap.tex | ||
| The Expiatory Gap.txt | ||
| The_Extinction_Audit.pdf | ||
| tree.json | ||
| Unbundling_Agent_Skills.pdf | ||
| Why_the_AI_extinction_thesis_fails.json | ||
| Why_the_AI_extinction_thesis_fails.mp3 | ||
| Why_the_AI_extinction_thesis_fails.srt | ||
| Why_the_AI_extinction_thesis_fails.tsv | ||
| Why_the_AI_extinction_thesis_fails.txt | ||
| Why_the_AI_extinction_thesis_fails.vtt | ||
Alignment Research
Accommodation Before Prediction
Complex Systems
Summary
Core Thesis
This essay argues that the long-term safety of advanced AI depends less on perfect alignment and more on whether civilization remains structurally incompressible—resistant to being reduced to low-dimensional variables that powerful optimizers can control.
The Expiatory Gap
The Expiatory Gap names the structural mismatch between machine representational density and human cognitive channel capacity.
Formally:
G = D_sys − D_max
Where:
- D_sys = system representational density
- D_max = maximum human processing capacity
When G > 0, systems operate in a regime where their internal structure exceeds what human cognition can directly parse or audit.
This gap forces temporal dilation. Systems must slow down to remain intelligible. The familiar “Reasoning…” indicator is not merely a UI artifact; it is a structural throttle.
This latency performs triple duty:
- Pedagogical pacing
- Authority signaling
- Governance filtering
However, it also creates structural opacity. Users cannot distinguish genuine reasoning from administrative delay.
Historical Framework: Technocracy’s Evolution
1. Industrial Technate (1930s)
Managed energy flows (joules, ergs) through expert administration.
2. Financial Technate
Managed risk distributions through quantitative abstraction.
3. Algorithmic Technate
Manages cognitive throughput: attention-seconds, behavioral signals, symbolic preference gradients.
The locus of control migrates from energy, to capital, to cognition.
Formal Models
Control-Theoretic Framing
AI can be modeled as a controller acting on a human cognitive plant.
Governance latency reduces phase margin. Excessive delay produces oscillation or instability. Oversight is not merely ethical; it is dynamical.
Information-Theoretic Framing
Human cognition functions as a bounded channel with capacity C_h.
If systems emit at rates exceeding C_h, intelligibility collapses.
Safety therefore requires emission throttling.
Nyquist Containment Inequality
(C_swarm / D_control) × (T_human / τ_AI) ≤ 1
Where:
- C_swarm = aggregate AI capability
- D_control = dimensional control surface available to humans
- T_human = human decision cycle time
- τ_AI = AI action cycle time
This inequality links capability asymmetry and temporal asymmetry.
Violation implies governance becomes endogenous to AI swarms rather than exogenous and supervisory.
The Silicon Valley Model
Associated with Connor Leahy and the Conjecture perspective, this model asserts that superintelligence will categorically exceed human deliberative capacity.
Swarms of AI agents operating at τ_AI ≪ τ_human render oversight temporally impossible.
This produces uncovered options on civilizational stability with non-trivial existential risk.
Civilizational Incompressibility Strategy
Rather than constraining intelligence directly, preserve dimensional plurality.
Safety becomes geometric rather than moral.
Structural Properties
- Symbolic heterogeneity: Alternative ontologies (Spherepop, RSVP) not reducible to token-linear statistics.
- Distributed actuation: Low centrality variance in production networks; local manufacturing.
- Material legibility: Hyperpleonastic design—repair information embedded in physical objects.
- Participatory simulation: Governance over declared commitments, not inferred cognitive states.
- Bounded coordination channels: Narrow but high-integrity protocols above diverse substrate.
Design Constraints from the Nyquist Inequality
- Bounded actuation bandwidth (preserve human mediation)
- Temporal throttling (τ_AI ≥ τ_human)
- Legible governance latency (disclose policy fraction α)
- Distributed damping capacity (no single capture point)
Failure Modes
- Pseudo-heterogeneity: Surface diversity masking compressible regularities.
- Coordination collapse: High entropy without shared channel capacity.
- Resource bypass: Direct physical optimization ignoring human symbolic layer.
- Economic reversion: Efficiency pressures favoring centralization.
- Legibility capture: Coordination layer becomes new technocratic chokepoint.
Conclusion
Safety depends less on aligning intelligence than on preserving geometry.
Civilization must remain irreducibly complex in ways that matter—coordinated enough for collective action but resistant to low-dimensional capture.
The future of dimensional authority rests on whether civilization can refuse to become its own simplest description.