Saying the Unsaid: Revealing the Hidden Language of Multimodal Systems Through Telephone Games

TL;DR

This paper investigates the hidden language of multimodal systems by using a telephone game approach to analyze their concept understanding and biases, providing insights into their interpretability and generalization.

Contribution

It introduces a novel telephone game framework and dataset to study the hidden concept connections and biases in multimodal systems, enhancing interpretability.

Findings

Identifies specific biases in multimodal systems' concept co-occurrence

Constructs a global map of concept connections through iterative telephone games

Uncovers unexpected concept relationships using Reasoning-LLMs

Abstract

Recent closed-source multimodal systems have made great advances, but their hidden language for understanding the world remains opaque because of their black-box architectures. In this paper, we use the systems' preference bias to study their hidden language: During the process of compressing the input images (typically containing multiple concepts) into texts and then reconstructing them into images, the systems' inherent preference bias introduces specific shifts in the outputs, disrupting the original input concept co-occurrence. We employ the multi-round "telephone game" to strategically leverage this bias. By observing the co-occurrence frequencies of concepts in telephone games, we quantitatively investigate the concept connection strength in the understanding of multimodal systems, i.e., "hidden language." We also contribute Telescope, a dataset of 10,000+ concept pairs, as the database of our telephone game framework. Our telephone game is test-time scalable: By iteratively running telephone games, we can construct a global map of concept connections in multimodal systems' understanding. Here we can identify preference bias inherited from training, assess generalization capability advancement, and discover more stable pathways for fragile concept connections. Furthermore, we use Reasoning-LLMs to uncover unexpected concept relationships that transcend textual and visual similarities, inferring how multimodal systems understand and simulate the world. This study offers a new perspective on the hidden language of multimodal systems and lays the foundation for future research on the interpretability and controllability of multimodal systems.