VisAlign: Dataset for Measuring the Degree of Alignment between AI and Humans in Visual Perception

TL;DR

This paper introduces VisAlign, a new dataset designed to measure how well AI models' visual perception aligns with human perception, especially in image classification, to improve AI safety and reliability.

Contribution

The paper presents a novel dataset with diverse scenarios and gold human perception labels to evaluate AI-human visual alignment in image classification tasks.

Findings

Analyzed five visual perception models for alignment with human perception.

Evaluated seven abstention methods using the dataset.

Validated dataset's reliability through statistical and expert analysis.

Abstract

AI alignment refers to models acting towards human-intended goals, preferences, or ethical principles. Given that most large-scale deep learning models act as black boxes and cannot be manually controlled, analyzing the similarity between models and humans can be a proxy measure for ensuring AI safety. In this paper, we focus on the models' visual perception alignment with humans, further referred to as AI-human visual alignment. Specifically, we propose a new dataset for measuring AI-human visual alignment in terms of image classification, a fundamental task in machine perception. In order to evaluate AI-human visual alignment, a dataset should encompass samples with various scenarios that may arise in the real world and have gold human perception labels. Our dataset consists of three groups of samples, namely Must-Act (i.e., Must-Classify), Must-Abstain, and Uncertain, based on the quantity and clarity of visual information in an image and further divided into eight categories. All samples have a gold human perception label; even Uncertain (severely blurry) sample labels were obtained via crowd-sourcing. The validity of our dataset is verified by sampling theory, statistical theories related to survey design, and experts in the related fields. Using our dataset, we analyze the visual alignment and reliability of five popular visual perception models and seven abstention methods. Our code and data is available at https://github.com/jiyounglee-0523/VisAlign.