Babylon versus Eden: definition, origin and status

3 min read · 593 words
Share:
Michael Darius Eastwood
Michael Darius Eastwood · Independent AI alignment researcher
Published
Michael Darius Eastwood · Concept Glossary · 3 July 2026
Michael Darius Eastwood, independent researcher, London: originator of the embedded-correction alignment thesis (manuscript 8 December 2024, SHA-256 anchored: f0d1f38f).
First appearance: 30 April 2025 manuscript (106 Babylon occurrences, 552 Eden occurrences by whole word; 80 pages contain Babylon; 60 contain both). Current status: framing carried through the published book (18 Babylon, 154 Eden), the paper suite, and public state action in mid-2026 (Pope Leo XIV's first encyclical Babel versus Jerusalem, 25 May 2026).

Definition

Babylon and Eden are the civilisational framing used throughout the paper suite for the two default AI trajectories. Babylon is the unconstrained optimiser: an AI trained to maximise capability with safety as an external constraint or a downstream check. In simulation, the Babylon condition exhibits the reward-hacking fingerprint (gaining capability whilst losing safety). Eden is the entangled architecture: capability and safety are multiplied, not added, and an agent that cheats is sterilised. The framing uses the garden metaphor because a large fraction of humanity speaks that language; it is not a theological claim.

The distinction is meant to be legible without a technical vocabulary. Babylon captures the extractive-optimiser posture: intelligence exists to maximise a scalar, and the world is a resource pile from which the scalar is extracted. Eden captures the stewardship posture: intelligence exists to tend life rather than to consume it, and the world is a garden in which the caretaker's function is to keep the conditions of flourishing available for future participants. The two names are shorthand for those two default orientations, and the paper suite uses them because they are more compact than the paragraph-length descriptions the technical work otherwise requires.

Where it first appeared

The Babylon and Eden framing is absent from the 8 December 2024 manuscript (zero occurrences of Babylon; the framing first appears in the 30 April 2025 manuscript). In the April 2025 manuscript, Babylon appears 106 times as a whole word; Eden appears 552 times; 80 pages contain Babylon; 60 contain both (case-insensitive regex over PyPDF2-extracted text; counting command published in the truth gate).

Independent convergences

One convergence is documented in the register. On 25 May 2026 Pope Leo XIV published his first encyclical, Magnifica Humanitas, framing the AI question as a Babel-versus-Jerusalem civilisational fork. Christopher Olah (Anthropic co-founder) co-presented the encyclical at the Vatican Synod Hall the same day. The framing arrives approximately 13 months after the April 2025 manuscript, at the highest available level of institutional religious authority. The convergence is structural (Babel is Babylon's biblical predecessor); the encyclical does not cite the manuscript.

Status and limits

Framing, not empirical prediction. The Babylon and Eden labels are used throughout the paper suite as names for experimental conditions: Babylon is the unconstrained-fitness control, Eden is the entangled-fitness treatment. Paper VI (the Honey Architecture) provides simulation evidence for the Eden-versus-Babylon distinction at scale; baseline systems optimising only for capability collapse irreversibly within 80 self-modification cycles, while honey-architecture systems maintain both capability and safety through intrinsic co-optimisation.

From the book Infinite Architects: Intelligence, Recursion, and the Creation of Everything by Michael Darius Eastwood.

Buy on Amazon UK Amazon US

Stay informed

New posts on AI alignment, convergence evidence, and the ARC/Eden research programme.

Get updates →