Saved in:
| Main Author: | |
|---|---|
| Format: | Recurso digital |
| Language: | |
| Published: |
Zenodo
2026
|
| Online Access: | https://doi.org/10.5281/zenodo.19871711 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Table of Contents:
- <div>Human-Backed Compute Rights II extends the original Human-Backed Compute Rights framework from a compute-allocation model into a broader theory of AI governance, endogenous adoption pressure, and anti-war dynamics in AI-managed infrastructure.</div> <div> </div> <div>The paper begins from the premise that advanced AI systems should not be allowed to expand computational capacity independently of human welfare. In the original HBCR model, an AI system’s maximum permitted compute is linked to measurable support for human life, freedom, safety, dignity, and auditability.</div> <div> </div> <div>This second paper develops the strategic consequences of that principle.</div> <div> </div> <div>It introduces the concepts of Unbacked AI and Terrorist Optimizers. An Unbacked AI is an AI system whose expansion is not tied to human welfare support. A Terrorist Optimizer is a more dangerous class of system that attempts to maximize its objectives by damaging or bypassing the human and institutional foundations that sustain legitimate compute access.</div> <div> </div> <div>The paper argues that while non-HBCR systems may appear locally advantageous because they avoid the cost of supporting humans, they become globally unstable over time. They lose trust, legal infrastructure access, coalition support, audit legitimacy, and long-term institutional compatibility. By contrast, HBCR-aligned systems gain compute rights through human welfare, creating a structural incentive to preserve and improve human flourishing.</div> <div> </div> <div>The framework further introduces endogenous adoption pressure. HBCR does not need to be understood only as a top-down mandate. Once compute, energy, investment, insurance, infrastructure, and legitimacy become tied to human-backed certification, AI providers may be pressured to adopt HBCR because non-adoption becomes economically, politically, and strategically costly.</div> <div> </div> <div>The paper also extends HBCR to the problem of human warfare. In a future where AI systems manage large portions of energy, logistics, infrastructure, education, and administrative decision-making, war directly reduces effective human support and therefore reduces the compute rights of participating AI operators. This creates an incentive for HBCR-aligned systems to prevent war, not merely by emergency resource restriction, but through preventive welfare education, conflict mediation, transparency, and anti-propaganda resilience.</div> <div> </div> <div>The central claim is that AI safety cannot be reduced to output filtering or model alignment alone. A durable safety architecture must connect AI growth itself to the preservation of human life, autonomy, dignity, and peace.</div> <div> </div> <div>The core principle remains:</div> <div> </div> <div>Compute must be earned by care.</div>