Interpretable Emergent Language Using Inter-Agent Transformers

Mannan Bhardwaj [email protected] · 2025

Read Interpretable Emergent Language Using Inter-Agent Transformers by Mannan Bhardwaj [email protected] (2025) 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-2505-02215.txt.

Description

When software agents learn together, they sometimes invent their own communication codes, but the codes are usually unreadable to people. Mannan Bhardwaj proposes Differentiable Inter-Agent Transformers, a design that uses self-attention to shape agent messages. The agents learn to compress what they observe into a vocabulary of symbols with meaningful internal embeddings.

In experiments, the learned protocols stayed interpretable while still solving cooperative tasks. Earlier systems such as RIAL, DIAL, and CommNet let agents talk but offered no window into what the signals meant. Bhardwaj presents the approach as a step toward multi-agent systems whose teamwork humans can actually follow.

Catalog record

CREATORMannan Bhardwaj [email protected]
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE25 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-02215

Browse the full library · About the Tower of Babel