RAcQUEt: Unveiling the Dangers of Overlooked Referential Ambiguity in Visual LLMs

TL;DR

This paper introduces RACQUET, a dataset for studying referential ambiguity in visual question answering, revealing that current models often overconfidently misinterpret ambiguous references and produce biased responses.

Contribution

The work presents RACQUET, a novel dataset for analyzing ambiguity in multimodal models, and demonstrates the limitations and biases of state-of-the-art models in handling ambiguity.

Findings

Models exhibit overconfidence in ambiguous scenarios.

Current models often produce stereotypical, biased responses.

Addressing ambiguity is crucial for fair and accurate AI systems.

Abstract

Ambiguity resolution is key to effective communication. While humans effortlessly address ambiguity through conversational grounding strategies, the extent to which current language models can emulate these strategies remains unclear. In this work, we examine referential ambiguity in image-based question answering by introducing RACQUET, a carefully curated dataset targeting distinct aspects of ambiguity. Through a series of evaluations, we reveal significant limitations and problems of overconfidence of state-of-the-art large multimodal language models in addressing ambiguity in their responses. The overconfidence issue becomes particularly relevant for RACQUET-BIAS, a subset designed to analyze a critical yet underexplored problem: failing to address ambiguity leads to stereotypical, socially biased responses. Our results underscore the urgency of equipping models with robust strategies to deal with uncertainty without resorting to undesirable stereotypes.