馬伊納克·馬祖姆達爾: 劣質數據如何令我們遠離優秀 AI

劣質數據如何令我們遠離優秀 AI

數據科學家馬伊納克·馬祖姆達爾 (Mainak Mazumdar) 指出，未來經濟不會由人類和工廠，而是由演算法和人工智能建構的。但如果這些演算法是在有偏誤的數據基礎上訓練出來，會發生什麼事？馬祖姆達爾從上海和紐約的實例，展示了品質不佳的數據如何導致 AI 做出錯誤決策和預測，並揭示為實現符合道德規範的 AI，所需的三個基礎設施重置。

Artificial intelligence promises to transform the economy, yet its power can reinforce bias when built on poor data. In one case, an AI model mistakenly turned a nonwhite face into a Caucasian one because African-American images were rare in the training set. Skewed data from censuses and market studies often leaves out marginalized groups, leading to decisions that ignore their needs. Ensuring fair AI starts with investing in representative, high-quality data, not just smarter algorithms.