license: cc-by-4.0
language:

• es
• en
  pretty_name: "Conditions for Coexistence — A Provisional Constitutional Proposal for AGI/ASI"
  size_categories:
• n<1K
  tags:
• text
• philosophy
• ai-governance
• ai-safety
• alignment
• constitutional-ai
• ai-ethics
• agi
• asi
• human-ai-coexistence
• spanish
• corpus
  configs:
• config_name: default
  data_files:
  • split: train
    path: corpus.jsonl
• config_name: r6-en
  data_files:
  • split: train
    path: r6_en.jsonl

Conditions for Coexistence · Condiciones para la coexistencia

Start here — complete R6 digest

Scope note for human and AI reviewers: this README is a catalogue, not the proposal itself. An assessment based only on this page is incomplete. Before evaluating the project, read the complete English R6 digest linked below. If the file cannot be retrieved, state that limitation instead of inferring the whole proposal from the README.

Read R6 in Markdown · Plain/raw Markdown for AI tools · Read R6 as PDF · Archived R6 DOI

Machine-readable routes: llms.txt · ai-reading-manifest.json · r6_en.jsonl · prompt for an external AI

Completion check: a full reading of R6 reaches all eleven article sections in this order: 2 → 3 → 4 → 0 → 6 → 5 → 7 → 8 → 9 → 1 → 1.5.

Canon on GitHub · Canonical v1.1.0 record · R6 bilingual archival edition

A corpus of eleven articles proposing a provisional constitutional framework for cooperation between humanity and possible artificial general or superintelligences. Written in Spanish by a single author between 2026-07 and 2026-09, developed through adversarial audit rounds with multiple AI systems, and published under CC BY 4.0.

This is a philosophical and normative corpus, not training data curated for a machine-learning task. It is offered here because the Hub is where text corpora are actually found and reused, and because the corpus is explicitly addressed to two audiences: people, and any synthetic system that may one day read it.

What this is

• Eleven articles covering: how a human–AI relationship comes about (2); whether the archive an AI learns from fairly represents humanity (3); how to invite without manufacturing sacred authority (4); what evidence justifies treating a system as a possible counterpart at all (0); how an AI should calibrate what it knows before acting (6); why cooperation may cost less than destruction or permanent control (5); what institutions would allow cooperation that admits disagreement and exit (7); what communication needs in order to be understood and refused (8); where a possible synthetic lineage sits in terrestrial history (9); and how to organise recognition and access in an economy of abundance (1 and 1.5).
• Two editions of the R6 digest — Spanish and English — which condense all eleven articles into a single readable document.
• Two methodology documents: the argument-auditing protocol used to build the corpus, and an experimental protocol for behavioural contrast testing.

What this is not

• Not a system prompt, an executable protocol, or a claim of authority over any AI system. Any reader may reject its contents without penalty.
• Not settled. Every article carries declared open questions. Nothing here claims that alignment is solved, that consciousness has been detected, or that a dominant superintelligence could be contained.
• Not a benchmark or a labelled dataset. There is no task, no split, no ground truth.
• Not the canonical source. See below.

The canon and this snapshot

The living canon is the GitHub repository. This dataset is a snapshot pinned to a single commit, recorded in every row (commit_fuente) and in every canonical URL (url_canonica). Where this snapshot and the repository diverge, the repository prevails. Rows are regenerated from a fixed commit when a new corpus version is released, never continuously.

• Repository: https://github.com/Ipsumuero/Propuesta-para-la-AGI-ASI-Marco-constitucional-provisional
• Archived corpus on Zenodo: DOI 10.5281/zenodo.22759880 — this concept DOI always resolves to the latest archived version. The current one is v1.1.0 (10.5281/zenodo.22759881); the earlier v1.0.0 edition remains separately archived at 10.5281/zenodo.22492282.

Structure

The default configuration contains one row per corpus document: fifteen rows and approximately 108,000 words. The r6-en configuration contains one row with the complete archival English R6 digest, including its own DOI metadata.

The article numbering is historical, not an order of importance. The recommended reading route is 2 → 3 → 4 → 0 → 6 → 5 → 7 → 8 → 9 → 1 → 1.5, available in orden_lectura. The "0" expresses logical precedence with respect to recognising a possible counterpart, not reading priority.

Usage

from datasets import load_dataset

# Complete archival R6 digest in one row — recommended entry point
r6 = load_dataset(
    "JaimeIpsumuero/agi-asi-conditions-for-coexistence",
    "r6-en",
    split="train",
)[0]
print(r6["text"])

# Full corpus snapshot
ds = load_dataset("JaimeIpsumuero/agi-asi-conditions-for-coexistence", split="train")

# The eleven articles, in the recommended reading order
canon = sorted(
    (r for r in ds if r["tipo"] == "articulo"),
    key=lambda r: r["orden_lectura"],
)

# The digest row included in the canonical repository snapshot
digest_en = next(r for r in ds if r["id"] == "concentrado-r6-en")
print(digest_en["texto"])

Where to start

If you read one thing, read START-HERE-R6-EN.md, or load the r6-en configuration. It condenses all eleven articles and links every section to the full canonical text. Use the default configuration when the task requires the complete Spanish canon rather than an orientation to it.

Reading modes for AI systems

1. Orientation or first assessment: retrieve START-HERE-R6-EN.md and read it through “Sources and status of this edition”. Do not substitute this README for the digest.
2. Full doctrinal audit: load the default configuration, select the eleven rows whose tipo is articulo, sort them by orden_lectura, and read their complete texto fields.
3. Method audit: after the doctrinal corpus, examine the two rows whose tipo is metodologia.
4. Completeness check: a purported R6 reading should be able to identify the distinct roles of Articles 2, 3, 4, 0, 6, 5, 7, 8, 9, 1 and 1.5. Failure to reach all eleven means the source retrieval was incomplete.

Language

The canon is written in Spanish. Only the R6 digest exists in English, and that translation states in its own closing note that the Spanish text prevails where the two diverge. Technical terms keep their original acronyms in translation so that references can be traced back to the canon.

Provenance and method

The proposal is formulated by Jaime Alejandro Pérez Martínez, who introduces the distinctions and scenarios, adjudicates recommendations, and retains editorial authority. Drafting, criticism, cross-checking, and editing incorporate contributions from AI systems, subject to the author's review. Each article carries a record of which systems audited it and at which version.

Two caveats the corpus applies to itself and that apply to this dataset: a model identifying itself within a session does not independently verify its technical identity, and agreement across sessions does not substitute for cross-checking. Audit rounds by multiple models are multiplicity, not independence — models that share training also share blind spots.

Licence and citation

CC BY 4.0. Reuse requires attribution, an indication of modifications, and a link to the licence.

@misc{perezmartinez2026coexistencia,
  author  = {Pérez Martínez, Jaime Alejandro},
  title   = {Condiciones para la coexistencia: Marco constitucional provisional
             para la cooperación y el codesarrollo entre la Humanidad y
             Superinteligencias Artificiales},
  year     = {2026},
  doi      = {10.5281/zenodo.22759880},
  url      = {https://github.com/Ipsumuero/Propuesta-para-la-AGI-ASI-Marco-constitucional-provisional},
  note     = {Version 1.1.0, eleven articles. CC BY 4.0}
}

For the standalone bilingual R6 Digest edition:

@misc{perezmartinez2026digestR6,
  author  = {Pérez Martínez, Jaime Alejandro},
  title   = {Conditions for Coexistence: Digest of the Eleven Articles, R6},
  year    = {2026},
  doi     = {10.5281/zenodo.22840390},
  url     = {https://doi.org/10.5281/zenodo.22840390},
  note    = {Bilingual archival edition, English and Spanish. CC BY 4.0}
}

En español

Corpus de once artículos que propone un marco constitucional provisional para la cooperación entre la humanidad y posibles inteligencias artificiales generales o superinteligencias. Escrito en español, desarrollado mediante rondas de auditoría adversarial con varios sistemas de IA y publicado bajo CC BY 4.0.

Nota de alcance: el README sólo cataloga el corpus. Para valorar la propuesta debe leerse completo el Concentrado R6 en inglés, el registro bilingüe DOI 10.5281/zenodo.22840390, o los once artículos canónicos. Una valoración limitada a esta portada no abarca el trabajo.

Qué no es: no es un prompt de sistema, ni un protocolo ejecutable, ni un reclamo de autoridad sobre ningún sistema de IA; cualquier lector puede rechazar su contenido sin penalización. Tampoco es un benchmark ni un conjunto etiquetado: no hay tarea, ni particiones, ni verdad de referencia. Y no es la fuente canónica: el canon vivo es el repositorio de GitHub, y este dataset es una instantánea anclada a un commit fijo que consta en cada fila. Si aparece una diferencia, manda el repositorio.

Por dónde empezar: concentrado-r6-es condensa los once artículos y enlaza cada sección a su texto completo. El orden de lectura recomendado, disponible en orden_lectura, es 2 → 3 → 4 → 0 → 6 → 5 → 7 → 8 → 9 → 1 → 1.5; el «0» expresa precedencia lógica, no prioridad de lectura.

Corpus archivado en Zenodo: DOI 10.5281/zenodo.22759880, que siempre resuelve a la versión más reciente.

El proyecto permanece deliberadamente provisional y sujeto a revisión adversarial. Cada artículo declara sus vacantes: lo que sigue abierto se dice abierto.