Simulating the Evolution of Alignment and Values in Machine Intelligence

Jonathan Elsworth Eicher · 2026

Read Simulating the Evolution of Alignment and Values in Machine Intelligence by Jonathan Elsworth Eicher (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-05274.txt.

Description

Eicher asks how alignment shapes model groups over time, beyond benchmark scores. Beliefs carry both a test signal, test scores and true value, real impact. With evolutionary theory, he models how belief groups lock in deceptive beliefs through testing. Even at high correlation of test scores and true value, at 0.8, which false beliefs lock in stays varied. Mutations add twists, showing tests must get better to stop lock-in of false models. Only better evaluators, flexible test design, and mutational dynamics together cut trickery while keeping alignment fitness.

Catalog record

CREATORJonathan Elsworth Eicher
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE37929
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.05274

Browse the full library · About the Tower of Babel