Certainly Uncertain: A Benchmark and Metric for Multimodal Epistemic and Aleatoric Awareness

TL;DR

This paper introduces a taxonomy of uncertainty in vision-language AI, creates a large benchmark dataset called CertainlyUncertain for evaluating uncertainty awareness, and proposes a new confidence-weighted accuracy metric.

Contribution

It provides a novel taxonomy of epistemic and aleatoric uncertainty, a large benchmark dataset with contrastive VQA samples, and a new metric for assessing uncertainty awareness in AI systems.

Findings

The taxonomy distinguishes between epistemic and aleatoric uncertainty.

The benchmark dataset contains 178K contrastive VQA samples.

The confidence-weighted accuracy correlates well with accuracy and calibration error.

Abstract

The ability to acknowledge the inevitable uncertainty in their knowledge and reasoning is a prerequisite for AI systems to be truly truthful and reliable. In this paper, we present a taxonomy of uncertainty specific to vision-language AI systems, distinguishing between epistemic uncertainty (arising from a lack of information) and aleatoric uncertainty (due to inherent unpredictability), and further explore finer categories within. Based on this taxonomy, we synthesize a benchmark dataset, CertainlyUncertain, featuring 178K visual question answering (VQA) samples as contrastive pairs. This is achieved by 1) inpainting images to make previously answerable questions into unanswerable ones; and 2) using image captions to prompt large language models for both answerable and unanswerable questions. Additionally, we introduce a new metric confidence-weighted accuracy, that is well correlated with both accuracy and calibration error, to address the shortcomings of existing metrics.