Abstract

This study explores the classification of hate speech by artificial intelligence (AI) and its errors, to explore its impact on society. The online sphere in modern society is full of hate speech. To address this, hate speech classification AIs have been developed and have positively impacted the self-regulation of online platforms. However, their errors often cause significant harm to certain individuals by failing to filter out hate speech or by censoring incorrect sentences. Therefore, we aim to examine the misclassification of hate speech classifiers and discuss the potential societal impact of these errors, as well as explore methods for improvement. This study employed the hate speech classifier and date provided by SmileGate to examine the patterns of misclassification within SmileGateʼs test data. Therefore, we hope that this study can lay the groundwork for aiding hate speech classifiers in making accurate classifications, thus mitigating their detrimental impact on society.