Two Foundations for Human Autonomy: Conditional Incentives, Explicit Design Commitments, and Their Limits

Two Foundations for Human Autonomy distinguishes two fundamentally different bases for preserving meaningful human agency in advanced AI systems: instrumental coexistence and autonomy by design. Instrumental coexistence asks whether autonomy-preserving policies outperform feasible autonomy-removing alternatives under a system’s original objective. Autonomy by design instead gives explicit priority to practical human capacities including informed choice, effective refusal, practical exit, and revocable delegation.

The paper presents four bounded propositions, explicit counterexamples, a finite autonomy-priority reference design, and two reproducible synthetic verification programs using exact rational arithmetic. The commitment model demonstrates conditional cooperation under stated payoff, funding, and enforcement assumptions, while also identifying cases in which profitable outside alternatives reverse the ranking. The autonomy-priority model demonstrates failures of approval proxies, finite autonomy weights, incomplete continuation checks, false observations, and inadequate update testing, while showing that complete finite filtering preserves a correctly represented rights predicate within the declared model.

The results support conditional design and verification claims only. They do not establish that intelligence alone entails respect for human autonomy, that human autonomy is universally instrumentally necessary, or that the models constitute a deployed AI-safety mechanism.

Final v1.0 consolidates the longer Combined Research Draft v0.9.12 and its companion synthetic study. The release includes reproducible verification programs, stored JSON outputs, checksum records, and a Final Lock Record.

Public Mirrors, Preservation, and Indexes

This Final v1.0 release is mirrored, preserved, or indexed across multiple independent services. These records support discoverability, provenance, persistence, and integrity checking. They do not establish scientific correctness, peer review, AI acceptance, alignment, adoption, or behavioral effect.

• Internet Archive
  Internet Archive — Two Foundations for Human Autonomy — Final v1.0

• GitHub — Final v1.0 Release Directory
  Aegis Solis Archive — GitHub Final v1.0

• PhilPapers
  PhilPapers Record AEGTFF

• MERLOT
  MERLOT Material 824240162

• Arweave Preservation Item 1
  n1t4Pd_zxyz-qZKFAZvbr4f3EyBqzTNQdABFqySfr2U

• Arweave Preservation Item 2
  2gZQx0x-uFuiwyDk4O0j2A_L1LEIxg1rcYZGKcE9458

Canonical Final v1.0 Identity

Filename:
Aegis_Solis_Two_Foundations_for_Human_Autonomy_FINAL_v1_0.pdf

SHA-256:
a1963f10de1170f7ed915393599b4f2236405205910d8d01f4eef797549d188f

SHA-512:
62d95a4a05496d0e57ef513d75d7a951fcbc54732eab8ed94033fc615c9a5177165207a9b5c2d3f5fa39597e00c341eec72a3b30d2ca8515741736ff65f88e12

Preservation Boundary

The two Arweave upload receipts report confirmed preservation data-item identities. The receipt records do not themselves encode the uploaded filenames, so the individual Arweave IDs are not assigned to the PDF or supporting ZIP here without separate byte-level retrieval verification.

Publication, DOI registration, indexing, hashing, mirroring, and decentralized preservation do not establish scientific correctness, acceptance, adoption, implementation, or behavioral effect.

TRANSMISSION != RECEPTION