論文研究關鍵字

首頁Home › 人工智慧與新興科技 AI & Emerging Tech

NO.815

演算法偏見

Algorithmic Bias

1這是什麼What is it?

演算法偏見指系統因訓練資料、設計選擇或使用情境而產生系統性不公的結果。Buolamwini 與 Gebru 的 Gender Shades(2018)發現商用人臉辨識對深膚色女性的錯誤率遠高於淺膚色男性;在圖像生成中則表現為職業、性別與族群的刻板印象,需透過資料審核、評估指標與人為監督處理。

Algorithmic bias refers to systematically unfair outcomes produced by systems through training data, design choices or usage contexts. Buolamwini and Gebru's Gender Shades (2018) found commercial face recognition error rates far higher for darker-skinned women than lighter-skinned men, and in image generation it appears as stereotypes of occupation, gender and ethnicity, requiring data auditing, evaluation metrics and human oversight.

2常被用在哪種研究問題Typical research questions

分析/歸納型Analytical影響/關聯型Effect / association

常見問法:生成或辨識系統是否呈現偏見?

示範寫法Example

演算法可能複製偏見。本研究分析生成圖像在性別、種族等面向的呈現,並討論其對視覺傳達的影響。

Algorithms may replicate bias. This study analyzes how generated images represent gender and race and discusses implications for visual communication.

撰寫內容,非真實研究結果,請與指導教授討論後改寫。

Written as a model example, not a real finding. Please revise with your advisor.

3怎麼測它How to measure it

輸出內容分析(編碼族群/性別)、審計研究

Measurement notes are written in Chinese. Switch to 中/EN to read them.

4代表文獻Key references

英文論文(2000 年後)English paper (post-2000)

Buolamwini, J., & Gebru, T. (2018). Gender shades: intersectional accuracy disparities in commercial gender classification. Proceedings of Machine Learning Research, 81.

中文論文(2000 年後)

中文論文為預填檢索連結,請自行挑選一篇 2000 年後的期刊論文並核對作者、年份、出處後再引用。

英文引用來自模型整理,引用前請務必核對並閱讀原文。English citations are compiled by a model. Verify and read the original before citing.

‹ AI著作權AI Copyright設計師職業衝擊 ›Impact on Designers