Sinovation Venture首席執行官、谷歌中國前總裁李開復表示,隨著企業開始推進低級服務工作的自動化進程,他們或將制造“虛假任務”,以測試員工是否勝任高級職位。
李開復在Collective[i]主辦的一場線上活動上說:“我們可能需要這樣一個世界——在這個世界里,人們‘假裝在工作’,但實際上他們正在被評估向上流動的可能性。”Collective[i]是一家將人工智能應用于銷售和客戶關系管理系統的公司。
企業高層的工作需要更深入、更有創造性的思考,自動化難以做到,必須由人類來完成。但是,如果入門級工作全部能夠由自動化完成,企業就沒有理由雇傭和培養年輕人才。因此,李開復認為,企業們將需要找到一種新方法來招聘入門級員工,并打造一條晉升之路。
以上是李開復對人工智能系統的廣泛應用可能產生的社會影響做出的幾項預測之一。其中一些摘自他即將出版的新書《人工智能2041:我們未來的十個愿景》(AI 2041: Ten Visions for Our future),這是他與科幻作家陳楸帆合作編寫的10個短篇小說集,闡述了人工智能改變個人和組織的可能方式。李開復開玩笑說:“這幾乎是書版的《黑鏡》(Black Mirror),只不過形式更具建設性。”李開復是人工智能和機器學習領域的知名專家,曾經于2018年出版了《人工智能超級力量:中國、硅谷和世界新秩序》(AI Superpowers: China, Silicon Valley, and the New World Order)一書。
提到人工智能在社會行為中扮演的角色時,人們關注最多的是算法傾向于反映、加劇現有的社會偏見。例如,推特在一場旨在消除算法偏見的競賽中發現,它的圖像裁剪工具優先留下較瘦的白人女性,而不是其他人種的人。由數據驅動的模型存在加劇社會不平等的風險,尤其是當越來越多的個人、企業、政府部門依靠數據庫來作出決定。正如李開復所指出的,當一家“公司擁有相當多的權力和數據時,即使它在表面上優化了算法結構,迎合了用戶興趣,它仍然可能做一些對社會非常有害的事情”。
盡管人工智能存在潛在性傷害,但李開復相信,開發人員和人工智能技術人員可以進行自我約束。他支持指標的開發,以幫助企業判斷其人工智能系統的性能,其方式類似于通過環境、社會和公司治理(ESG)指標來確定企業的表現。“你只需要提供可靠的方法,讓這類人工智能倫理成為可定期衡量、可付諸實施的東西。”他說。
但是李開復指出,為了訓練程序員們,還需要做更多的工作,比如開發一些能夠幫助“監測潛在的偏見問題”的工具。更廣泛地說,他建議人工智能工程師采用“類似于醫生培訓中的希波克拉底誓言”的東西——所謂“希波克拉底誓言”,是指醫生們與病人們打交道時所需的職業道德,常被概括為“不傷害”原則。
“從事人工智能工作的人需要意識到,他們在編程時對人們的生活負有巨大的責任。”李開復說,“這不僅僅是為他們的互聯網公司雇主賺多少錢的問題。”(財富中文網)
編譯:楊二一
Sinovation Venture首席執行官、谷歌中國前總裁李開復表示,隨著企業開始推進低級服務工作的自動化進程,他們或將制造“虛假任務”,以測試員工是否勝任高級職位。
李開復在Collective[i]主辦的一場線上活動上說:“我們可能需要這樣一個世界——在這個世界里,人們‘假裝在工作’,但實際上他們正在被評估向上流動的可能性。”Collective[i]是一家將人工智能應用于銷售和客戶關系管理系統的公司。
企業高層的工作需要更深入、更有創造性的思考,自動化難以做到,必須由人類來完成。但是,如果入門級工作全部能夠由自動化完成,企業就沒有理由雇傭和培養年輕人才。因此,李開復認為,企業們將需要找到一種新方法來招聘入門級員工,并打造一條晉升之路。
以上是李開復對人工智能系統的廣泛應用可能產生的社會影響做出的幾項預測之一。其中一些摘自他即將出版的新書《人工智能2041:我們未來的十個愿景》(AI 2041: Ten Visions for Our future),這是他與科幻作家陳楸帆合作編寫的10個短篇小說集,闡述了人工智能改變個人和組織的可能方式。李開復開玩笑說:“這幾乎是書版的《黑鏡》(Black Mirror),只不過形式更具建設性。”李開復是人工智能和機器學習領域的知名專家,曾經于2018年出版了《人工智能超級力量:中國、硅谷和世界新秩序》(AI Superpowers: China, Silicon Valley, and the New World Order)一書。
提到人工智能在社會行為中扮演的角色時,人們關注最多的是算法傾向于反映、加劇現有的社會偏見。例如,推特在一場旨在消除算法偏見的競賽中發現,它的圖像裁剪工具優先留下較瘦的白人女性,而不是其他人種的人。由數據驅動的模型存在加劇社會不平等的風險,尤其是當越來越多的個人、企業、政府部門依靠數據庫來作出決定。正如李開復所指出的,當一家“公司擁有相當多的權力和數據時,即使它在表面上優化了算法結構,迎合了用戶興趣,它仍然可能做一些對社會非常有害的事情”。
盡管人工智能存在潛在性傷害,但李開復相信,開發人員和人工智能技術人員可以進行自我約束。他支持指標的開發,以幫助企業判斷其人工智能系統的性能,其方式類似于通過環境、社會和公司治理(ESG)指標來確定企業的表現。“你只需要提供可靠的方法,讓這類人工智能倫理成為可定期衡量、可付諸實施的東西。”他說。
但是李開復指出,為了訓練程序員們,還需要做更多的工作,比如開發一些能夠幫助“監測潛在的偏見問題”的工具。更廣泛地說,他建議人工智能工程師采用“類似于醫生培訓中的希波克拉底誓言”的東西——所謂“希波克拉底誓言”,是指醫生們與病人們打交道時所需的職業道德,常被概括為“不傷害”原則。
“從事人工智能工作的人需要意識到,他們在編程時對人們的生活負有巨大的責任。”李開復說,“這不僅僅是為他們的互聯網公司雇主賺多少錢的問題。”(財富中文網)
編譯:楊二一
As businesses begin to automate low-level service work, companies may start creating fake tasks to test employee suitability for senior positions, says Kai-Fu Lee, the CEO of Sinovation Ventures and former president of Google China.
“We may need to have a world in which people have ‘the pretense of working,’ but actually they’re being evaluated for upward mobility,” Lee said at a virtual event hosted by Collective[i], a company that applies A.I. to sales and CRM systems.
Work at higher levels of a company, which requires deeper and more creative thinking, is harder to automate and must be completed by humans. But if entry-level work is fully automated, companies don't have a reason to hire and groom young talent. So, Lee says, companies will need to find a new way to hire entry-level employees and build a path for promotion.
It was one of several predictions Lee made about the possible social effects of widespread adoption of A.I. systems. Some were drawn from his upcoming book, AI 2041: Ten Visions for Our Future—a collection of 10 short stories, written in partnership with science fiction author Chen Qiufan, that illustrate ways that A.I. might change individuals and organizations. “Almost a book version of Black Mirror in a more constructive format,” joked Lee, a well-known expert in the field of A.I. and machine learning and author of the 2018 book AI Superpowers: China, Silicon Valley, and the New World Order.
Talk of A.I. and its role in social behavior often centers on the tendency of algorithms to reflect and exacerbate existing social biases. For example, a contest by Twitter to root out bias in its algorithms found that its image-cropping model prioritized thinner white women over people of other demographics. Data-driven models risk reinforcing social inequality, especially as more individuals, companies, and governments rely on them to make consequential decisions. As Lee noted, when a “company has too much power and data, [even if] it’s optimizing an objective function that’s ostensibly with the user interest [in mind], it could still do things that could be very bad for the society.”
Despite the potential for A.I. to do harm, Lee has faith in developers and A.I. technicians to self-regulate. He supported the development of metrics to help companies judge the performance of their A.I. systems, in a manner similar to the measurements used to determine a firm's performance against environmental, social, and corporate governance (ESG) indicators. "You just need to provide solid ways for these types of A.I. ethics to become regularly measured things and become actionable."
Yet he noted that more work needs to be done to train programmers, including the creation of tools to help "detect potential issues with bias." More broadly, he suggested that A.I. engineers adopt something "similar to the Hippocratic oath in medical training," referring to the set of professional ethics that doctors adhere to during their dealings with patients, most commonly summarized as "Do no harm."
“People working on A.I. need to realize the massive responsibilities they have on people’s lives when they program," Lee said. "It’s not just a matter of making more money for the Internet company that they work for.”