{"resolved":true,"match_type":"uuid","source_type":"original_book","book_id":"094ce2a7-d7e6-4de6-a84c-9994a96a3c4a","title":"被算法操控的生活：重新定义精准广告、大数据和AI","author":null,"edition":null,"publisher":null,"year":null,"isbn":"9787571003012","requested_by":"printed_page","book":{"source_type":"original_book","book_id":"094ce2a7-d7e6-4de6-a84c-9994a96a3c4a","id":"094ce2a7-d7e6-4de6-a84c-9994a96a3c4a","uuid":"094ce2a7-d7e6-4de6-a84c-9994a96a3c4a","title":"被算法操控的生活：重新定义精准广告、大数据和AI","normalized_title":"被算法操控的生活重新定义精准广告大数据和ai","author":null,"edition":null,"publisher":null,"year":null,"isbn":"9787571003012","isbns":["9787571003012"],"parsed":true,"ai_readable":true,"document_type":"book","page_count":257,"navigation_url":"https://workshop.closeclaw.top/ai/book/094ce2a7-d7e6-4de6-a84c-9994a96a3c4a","title_url":"https://workshop.closeclaw.top/ai/title/%E8%A2%AB%E7%AE%97%E6%B3%95%E6%93%8D%E6%8E%A7%E7%9A%84%E7%94%9F%E6%B4%BB%EF%BC%9A%E9%87%8D%E6%96%B0%E5%AE%9A%E4%B9%89%E7%B2%BE%E5%87%86%E5%B9%BF%E5%91%8A%E3%80%81%E5%A4%A7%E6%95%B0%E6%8D%AE%E5%92%8CAI","fixed_page_url_template":"https://workshop.closeclaw.top/ai/book/094ce2a7-d7e6-4de6-a84c-9994a96a3c4a/page/{printed_page}","path_api_url_template":"https://workshop.closeclaw.top/api/ai/book/094ce2a7-d7e6-4de6-a84c-9994a96a3c4a/page/{printed_page}","read_url":"https://workshop.closeclaw.top/api/ai/read/094ce2a7-d7e6-4de6-a84c-9994a96a3c4a","page_url":"https://workshop.closeclaw.top/api/ai/books/094ce2a7-d7e6-4de6-a84c-9994a96a3c4a/page","search_url":"https://workshop.closeclaw.top/api/ai/books/094ce2a7-d7e6-4de6-a84c-9994a96a3c4a/search"},"page":{"printed_page":54,"pdf_page":74,"physical_index":74,"physical_index_base":0,"printed_page_status":"recognized","page_sha256":"2dcc601a3dff2b31c7dd82e8af9837c919e576394cf6a49e9d6079ed8a10739f"},"chapter":"第3章你以为自己了解朋友？脸书比你更懂他们","content":"此分解表显示了COMPAS算法中的风险评估项目（列）和两年内犯人是否会再次犯罪的预测情况（行）。关于“高风险”和“低风险”的定义及其他细节请参见 ProPublica 的分析报告\n\n这种差异本身并不意味着该算法存在偏见，因为在黑人和白人犯人中，再次犯罪的比例有所不同：52.6%的黑人罪犯在两年内因另一项罪行被逮捕，而仅有39.4%的白人罪犯因另一项罪行被抓。归类为高风险或低风险的犯罪者，其总体比例差异并不构成ProPublica对该算法的批评基础。朱莉娅和她的同事们意识到黑人罪犯的惯犯率比白人罪犯高，于是认为这个算法一定犯了某种类型的错误。\n\n在评估算法时，我们用“误报”(FalsePositive，也称假阳性)和“漏报’（False Negative，也称假阴性）来思考通常都是有效的。对于COMPAS算法来说，误报指的是一个不会在未来犯罪的人被列为高风险罪犯的情况，也就是说模型做出了肯定却错误的预测。误报率指那些没有再次犯罪却被列为高风险罪犯的人数除以未再犯的总人数。\n\n黑人罪犯的误报率是805/1714≈46.9%，白人罪犯则是23.5%，因此黑人罪犯的误报率比例比白人被告高很多。\n\n被算法操控的生活OUTNUMBERED\n\n54","previous_page":53,"next_page":55}