一年前被解僱的谷歌AI倫理專家宣佈成立研究所，專注防範AI傷害，對抗科技公司“作惡”

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"原谷歌AI倫理專家Timnit Gebru表示，其所成立的AI研究所DAIR致力於成爲一家提供不同觀點、並防止人工智能傷害的獨立機構。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"原谷歌AI倫理專家宣佈成立人工智能研究所"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"一年前被谷歌解僱的頂級AI倫理專家Timnit Gebru ，近日宣佈成立人工智能研究所。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Timnit Gebru曾是谷歌“倫理人工智能團隊”（Ethical AI Team）的技術聯合負責人 ，致力於解決人工智能算法存在的偏見等問題，同時還從事數據挖掘領域的工作。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"去年12月4日，Timnit Gebru 在其個人推特上發佈消息稱：自己突然遭到公司的“郵件解僱”，原因似乎與她批評谷歌的政策相關。起因是，Timnit Gebru 所撰寫的一篇關於批評大型人工智能模型危害的論文（包括谷歌開發的模型）引發爭議。此外，格布魯還曾公開指責過谷歌對待女性和黑人員工的方式。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"自從離開谷歌以後，Timnit Gebru 就一直致力於成立自己的人工智能研究機構。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"12月3日，據外媒報道，在Timnit Gebru 被谷歌解僱即將滿一週年之際，她宣佈成立了一家獨立的人工智能研究所 —— “Distributed AI Research”（分佈式人工智能研究，簡稱DAIR）。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"DAIR致力於成爲一家提供不同觀點、並防止人工智能傷害的獨立機構，該機構專注研究人工智能技術對某些邊緣化羣體的危害，這些羣體往往面臨着人工能系統所帶來的不同程度的不利影響。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"DAIR旨在對抗大型科技公司在人工智能研究、開發和部署方面的普遍影響。例如此前，Timnit Gebru曾開展了面部識別軟件對有色人種的偏見的研究，這令亞馬遜改變了其政策。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Timnit Gebru表示，DAIR將加入現有的小型獨立機構的生態中去，如關注數據與社會、算法正義聯盟和黑人生活數據等。她希望DAIR能夠從外部影響谷歌等大型科技公司內部的人工智能政策和實踐。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"未來一年，DAIR計劃聘請5名研究人員。該研究所的首位研究員Resetje Sefala在南非工作，她一直在研究如何使用計算機視覺算法和衛星圖像來追蹤“空間種族隔離”（某些地區的少數羣體隔離）相關問題。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Timnit Gebru稱，她必須持續關注大型科技公司，因爲它們的危害可能非常嚴重。“此前我關注大型語言模型的危害問題，現在我關注社交媒體以及我家鄉發生的事情，例如Facebook在埃塞俄比亞煽動種族暴力的報道”。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"DAIR是一家非營利性組織。目前，DAIR已經獲得了福特汽車、MacArthur、Kapor Center、洛克菲勒基金和開放社會基金會等共計370萬美元的資金。後續，Timnit Gebru 希望開展一些與"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"DAIR自身研究相關的諮詢工作，以增加該機構的營收來源。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Timnit Gebru 表示，作爲一家獨立機構，DAIR的目標是比大多數人工智能實驗室更具包容性，DAIR將更自由地質疑人工智能的潛在不利影響，而且不會受到“學術政治”和發表論文壓力的阻礙，她認爲，發表論文會使大學研究複雜化。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Timnit Gebru 還是著名的技術多樣性倡導者，她聯合創立了一個名爲“Black in AI”的黑人研究人員社區。一直以來，她都在直言不諱地對科技公司對待黑人員工的方式提出批評。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"加州大學洛杉磯分校(UCLA)教授Safiya Noble是DAIR的顧問委員會成員，她主要從事"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"科技平臺如何塑造社會的研究，她認爲，Timnit Gebru 的機構是人類在理解和應對技術對社會的影響方面取得新型進展和更具包容性的範例。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“黑人女性一直是幫助我們瞭解大型科技公司和各種對社會有害的技術的危害的主要貢獻者，但我們知道黑人女性在美國企業界和學術界面臨諸多限制。”“Timnit Gebru意識到了谷歌的危害，並試圖干預，但沒有得到大量支持 —這家公司迫切需要這種洞察力。” Safiya Noble說。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"專注研究人工智能的不安全使用問題"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"本週二，Timnit Gebru在歐洲議會的聽證會上表示，政府需要遏制科技公司的力量，增加員工的力量，以保護人們減少人工智能被不安全使用所導致的風險。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"她認爲，監管必須到位。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"此前，加利福尼亞州推出的《沉默不再法案》(Silence No More Act)規定，公司阻止員工公開談論騷擾或歧視是非法的。“我認爲這需要普及。違反這些法律的公司應受到更嚴厲的懲罰，就像我們在亞馬遜看到的大規模激進的工會解散活動一樣。”"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"今年歐盟發佈的全球第一部人工智能法律 ——《人工智能法案》，該法案對最有可能造成傷害的人工智能技術的使用進行了限制，法案提出了AI統一監管規則，旨在從國家法律層面限制AI技術發展帶來的潛在風險和不良影響。該法案的首席談判代表Brando Benifei呼籲《人工智能法案》建立一個補償機制，並呼籲工會更多地參與其中。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Timnit Gebru認爲，科技公司和軍事部門是人工智能技術開發的最大投資方。爲了停止人工智能危害的惡性循環，政府需要投資其他社區建設人工智能技術。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“我們希望看到能優先考慮公民福祉的技術，而不是一場場科技競賽，研究如何更有效地‘殺死’更多的人，或者爲世界各地的少數公司賺最多的錢”...."}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"一年前，Timnit Gebru因批評谷歌AI政策被解僱"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"一年前，Timnit Gebru 被谷歌解僱的事件廣受關注，並引發了一場關於員工質疑公司技術影響自由的爭議。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"當時，Timnit Gebru 在個人推特上發佈消息稱，自己突然被谷歌人工智能負責人 Jeff Dean 開除，個人公司賬戶也已經被公司切斷。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Timnit Gebru 稱，她的領導層對她的離職進行了“快車道”處理，原因是在本週早些時候，她向谷歌深度學習人工智能研究團隊“谷歌大腦”（Google Brain）的女性成員和“盟友”發送了一封電子郵件。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"郵件中，Timnit Gebru提到，自己提交論文後，2個月時間裏沒有得到任何反饋，而在休假期間，突然被命令需要在規定日期前撤回論文，且沒給出任何原因。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"她認爲，像這樣壓制邊緣化的聲音的做法有悖於谷歌一直所倡導的“構建負責人的 AI”原則，“我認爲，在谷歌，唯一有意義的就是等級制度。我看到我的專業知識被忽視。而且，任何有特權的人都可以決定他們不希望你的論文在沒有溝通的情況下發表”。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"隨着事件的發酵，作爲另一當事人的 Jeff Dean 通過郵件的形式正面迴應稱，Timnit Gebru被解僱事件的起因並非歧視，而是她的論文未達標。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"據Jeff Dean 描述，這件事發生的背景和經過是，Timnit 與四名谷歌同事以及一些需要經過審查的外部合作者共同撰寫了一篇論文。論文通常需要在內部審查過程中進行修改（或者甚至被認爲不適合提交便會被否決）。但Timnit這篇論文只在截止日期前一天才被分享出來——實際上我們通常需要兩週的時間進行審查——然後這篇論文沒有等待審稿人的反饋，而是被批准提交後進行了提交。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"隨後，一個跨職能團隊對這篇論文進行了審查，作者被告知，這篇論文沒有達到我們規定的發表的標準，並得到了關於原因的反饋。論文中忽略了太多相關的研究——例如，它談論了大型模型對環境的影響，但卻忽視了之後的研究顯示出了模型能產生更高效率的研究。同樣，它也提出了對語言模型偏見的擔憂，但沒有提到最近的一些研究正在試圖緩解這些問題。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Timnit Gebru 的經歷在社交平臺引發了熱烈討論，推特上有不少科學家紛紛轉發相關新聞並表態支持她。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"其中包括一個由谷歌現任及前員工聯合創立的“Google Walkout For Real Change”組織，他們在博客上發起了簽名運動。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"該組織還向谷歌高層提出，要求參與審查 Gebru 論文決定的人員解釋領導層單方面拒絕論文的原因及過程，Google 領導層命令 Gebru 及其同事撤回對大規模語言模型的研究的決定。此外，要求Google Research 承諾通過提供明確的指導方針來審查 Google 的 AI 原則等。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"參考鏈接："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"link","attrs":{"href":"https:\/\/www.washingtonpost.com\/technology\/2021\/12\/02\/timnit-gebru-dair\/","title":"","type":null},"content":[{"type":"text","text":"https:\/\/www.washingtonpost.com\/technology\/2021\/12\/02\/timnit-gebru-dair\/"}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"link","attrs":{"href":"https:\/\/techcrunch.com\/2021\/12\/02\/google-timnit-gebru-ai-research-dair\/","title":"","type":null},"content":[{"type":"text","text":"https:\/\/techcrunch.com\/2021\/12\/02\/google-timnit-gebru-ai-research-dair\/"}]}]}]}