| 概要 |
Despite recent advances, reports suggest that artificial intelligence (AI) chatbots and image generators demonstrate biases against White people when crafting responses to their place in history and o...n utterances demonstrating White identity. Until now, however, there has yet to be a systematic analysis, using corpus data, that characterizes the contexts in which these topics are covered in text-based online news reporting to determine if the anti-White sentiment demonstrated by AI is merely a reflection of the textual data on which it has been trained. Using virtual corpora generated from English-language online news articles (via the NOW Corpus) to extract frequency-based keyword lists, the current study demonstrates that news reporting on phrases such as “White lives matter” and “It’s okay to be White” was characterized by terms with inherently negative connotations. Overall, the most frequently occurring vocabulary consisted largely of evocative, highly-subjective and threatening nouns and adjectives, with what could be described as ad hominem attacks featured conspicuously throughout. Terms such as “supremacist,” “neo-Nazi,” “racist,” and “fascist” occurred repeatedly across multiple corpora and at high frequency. These findings provide a possible explanation for the mechanism by which ChatGPT and other AI content-generation modalities display anti-White biases and double standards.続きを見る
|
Mendeley出力
