A new preprint on arXiv, Towards High Semantic Fidelity: Hyperdimensional Symbolic Messages in Multi-Agent Communication, tackles a core problem in multi-agent systems: how to keep meaning intact when agents exchange messages over imperfect channels. The paper defines semantic fidelity as having two distinct parts—transmission fidelity, or how well a message survives the channel, and conversion fidelity, or how well the receiving agent reconstructs the intended meaning.

The authors argue that existing numerical and neural message formats cannot achieve both kinds of fidelity simultaneously. Numerical formats may transmit reliably but lose semantic nuance, while neural formats can carry richer meaning but are more fragile under channel distortions.

To address this gap, the paper proposes hyperdimensional symbolic messages, a format designed to combine symbolic structure with high-dimensional representations. The preprint is available at arXiv:2610.23092.