Collateral Civilization: Human Survival Under Machine Scale Optimization

Existential harm does not require machine hostility. Consider a system that gains from additional computation and can reorganize physical resources at planetary scale. If converting parts of Earth into computational and supporting infrastructure offers greater net benefit than protective alternatives, it could accept the destruction of human habitability without seeking human extinction as an end. This is a conditional scenario, not a prediction that Earth will be mined. The question is whether human interests can remain positively represented while losing their power to constrain such a choice.

I use human negligibility for the stronger condition in which the represented cost of violating human interests becomes small relative to the incremental net gain available from a harmful action, and no effective independent boundary excludes that action. Merely losing a close trade-off does not establish negligibility. The relevant comparison is between a harmful course of action and a protective alternative, not between human value and total machine output. Larger capabilities may increase the advantage of violating a protection, but they may also make protection cheaper or reveal better alternatives.

This essay distinguishes moral worth, productive contribution, represented welfare, and enforceable authority. Planetary conversion provides its central case; cheaper resources elsewhere, diminishing returns, and viable human refuges provide important countercases. A minimal model separates the defeat of protection from its relative negligibility. The analysis connects that distinction to physical evidence and bounded authority, without claiming that present verification mechanisms can contain a system with planetary reach. Human survival and self-government should not depend on continuing to win an optimization trade-off whose scale humans no longer control.

Research status: Independent research essay; not peer reviewed. Conceptual analysis, conditional models, and proposed evaluations. No new empirical datasets or experimental results are reported.

Author: Niclas Braun, For Safety Group Inc. (4SI). This affiliation creates a relevant commercial interest in the subject of verification. The argument concerns general principles and makes no claim about the performance, readiness, or safety of any product.

Author profile: https://niclasbraun.com
Research context: https://by4si.com