算法伦理
For five years the British government used a racist algorithm in order to help determine the outcome of visa applications. Last week they announced that it “needs to be rebuilt from the ground up”.
五年來,英國政府一直使用種族主義算法來幫助確定簽證申請的結果。 上周,他們宣布它“需要從頭開始重建”。
“The Home Office’s own independent review of the Windrush scandal, found that it was oblivious to the racist assumptions and systems it operates. This streaming tool took decades of institutionally racist practices, such as targeting particular nationalities for immigration raids, and turned them into software. The immigration system needs to be rebuilt from the ground up to monitor for such bias and to root it out.”
內政部自己對Windrush丑聞的獨立審查發現,它沒有遵守種族主義的假設和運作的制度。 這種流媒體工具采取了數十年的制度上種族主義的做法,例如針對特定國籍的移民襲擊,并將其轉變為軟件。 移民系統需要從頭開始重建,以監測這種偏見并消除這種偏見。”
This is a colossal victory for everyone, the government has begun to identify parts of the racist machine and is now dismantling them. This is justice. But how did the government get their hands on a racist algorithm? Who designed it? And should they be punished?
對于每個人來說 , 這都是一個巨大的勝利 ,政府已經開始確定種族主義機器的各個部分,現在正在拆除它們。 這是正義。 但是,政府如何掌握種族主義算法呢? 誰設計的? 他們應該受到懲罰嗎?
‘Hand-Written’ AlgorithmsHistorically algorithms were ‘hand-written’ whereby the developers would manually set parameters that would dictate the outcome of an algorithm. A famous example is Facebook’s EdgeRank algorithm, which would consider a measly three factors (user affinity, content weighting and time-based decay).
“手寫”算法從歷史上看,算法是“手寫的”,開發人員可以手動設置指示算法結果的參數。 一個著名的例子是Facebook的EdgeRank算法 ,該算法僅考慮三個因素(用戶親和力,內容權重和基于時間的衰減)。
source)來源 )The score between users is calculated based on how ‘strong’ the developers of the algorithm perceive their interactions to be e.g. they dictate that if you shared your mate’s post last week, you like them 20% more than your other mate who made a similar post and you only liked it instead.
用戶之間的得分是根據算法開發人員對他們的互動的“強烈”程度來計算的,例如, 他們指出 ,如果您上周分享了同伴的帖子,那么您比其他發類似帖子的同伴要高20%而您只喜歡它。
At that point, it would’ve been quite easy for Facebook to be accountable for the results of their algorithms. They told it exactly what to do so they could be held liable for the outcomes of their algorithm. Unfortunately, this is no longer the case, popular commercial algorithms have undergone a radical change in design. The parameters that were once ‘hand-written’ are now decided by a ghost in the machine.
到那時, Facebook要對其算法的結果負責很容易。 他們確切地告訴了該怎么做,以便對算法的結果負責。 不幸的是,情況已不再如此,流行的商業算法在設計上發生了根本性的變化。 曾經“手寫”的參數現在由機器中的重影決定。
The Ghost in the MachineFrom the beginning of 2011 until around 2015 Facebook used its new machine-learning algorithm to dictate what users saw on their newsfeeds, instead of three parameters, this new beast considers at least 100,000 different factors that are weighted by the machine learning (ML) algorithm(s) (source — Facebook still use an ML algorithm today). Not a single one of these parameters is known to the developers of the algorithms, the AI is a black box that spits out an answer based on whatever information it has been fed previously.
機器中的幽靈從2011年初到2015年左右,Facebook使用其新的機器學習算法來指示用戶在其新聞源上看到的內容,而不是三個參數,該新的野獸考慮了至少100,000個由機器加權的不同因素學習(ML)算法( 來源 -今天的Facebook仍在使用ML算法)。 這些算法的開發者并不知道這些參數中的任何一個,AI是一個黑匣子,它會根據先前提供的任何信息吐出答案。
source — excellent read)來源 -優秀閱讀)In “The ethics of algorithms: Mapping the debate” the authors identify six ethical concerns that are raised by algorithms. The epistemic concerns (degree of validation of knowledge) arise from having poor datasets, without sound data your AI isn’t about to make sound decisions. The results and behaviors that these algorithms invoke create the normative concerns, they help create a new undesirable norm from the low-quality datasets. Then when it’s all said and done, none of these decisions can be traced back to their origins and no one is held accountable for the outcomes.
在“算法的倫理學:為辯論做準備”中 ,作者指出了算法提出的六個倫理學問題。 認知問題(知識的驗證程度)是由不良數據集引起的,如果沒有聲音數據,您的AI將無法做出正確的決定。 這些算法調用的結果和行為創建了規范關注點,它們有助于從低質量數據集中創建新的不良規范。 然后,當一切都說完了,所有這些決定都無法追溯到其起源,也沒有人對結果負責 。
Windrush DataThe ‘streaming tool’ (ML algorithm) used by the British government ranked each visa applicant red, yellow, or green — this would heavily influence the outcome of the government’s decision of whether or not to grant a visa. The (racist) datasets that were fed into this algorithm were definitely misguided, inconclusive and instructable. The outcomes were determined by datasets created in the past, created by people who judged someone a threat solely based on the color of their skin or their country of origin. This racist data was fed into the algorithm and the algorithm made racist decisions.
Windrush數據英國政府使用的“流工具”(ML算法)將每個簽證申請人排名為紅色,黃色或綠色-這將嚴重影響政府決定是否批準簽證的結果。 饋送到此算法中的(種族)數據集絕對是錯誤的,不確定的和可指導的。 結果是由過去創建的數據集確定的,這些數據集是由僅根據某人膚色或原籍國來判斷某人為威脅的人創建的。 該種族數據被輸入到算法中,并且該算法做出了種族決策。
So it’s not the algorithm’s fault that the creators of the dataset were racist, this algorithm was completely neutral, it only showed us a reflection of the shortcomings of the data that it has received. What can be done to improve the ethics of our algorithms and of our datasets in order to reduce unfair outcomes?
因此,數據集的創建者是種族主義者不是算法的過錯,該算法是完全中立的,它只向我們反映了它接收到的數據的缺點。 為了減少不公平的結果,可以采取什么措施來提高我們算法和數據集的道德標準?
Transparent Algorithms would help the developers and consumers understand the unethical biases within the machine, once identified, the model or dataset could be refined in order to eliminate the bias.
透明算法將幫助開發人員和消費者理解機器中不道德的偏差,一旦確定,可以對模型或數據集進行完善以消除偏差。
Improved Data Regulation is required in order to prevent data from being mistreated in order to create unfair outcomes for minorities and society as a whole.
為了防止數據被濫用以對少數民族和整個社會造成不公平的結果,需要改進數據法規 。
Education will help everyone see without rose-tinted spectacles. To understand that it’s YOUR data and you should have adequate rights to protect it. Failing to do so only feeds the unethical algorithms.
教育將幫助每個人沒有玫瑰色的眼鏡。 要了解這是您的數據,您應該擁有足夠的權利來保護它。 否則,只會滋養不道德的算法。
Final ThoughtsMachine Learning algorithms are popular for two very good reasons, they’re profitable and AI is sexy. Naturally, there is a commercial interest in the former and most computer science students have the latter on their mind.
《 Final Thoughts》機器學習算法之所以受歡迎,有兩個很好的原因,它們是有利可圖的,而AI則很性感。 當然,前者具有商業利益,大多數計算機科學專業的學生都對后者感興趣。
I encourage everyone to talk to their friends and family about this topic because these algorithms are far more powerful than most people realise.
我鼓勵每個人都與他們的朋友和家人談論這個話題,因為這些算法比大多數人意識到的要強大得多。
Popular algorithms are dictating how we think by choosing what we see on a daily basis, they pick our news, they pick our films, they pick our wardrobe, they pick our romantic partners, they pick our friends… And no one knows how they do it, so no one is accountable.
流行的算法決定了我們如何選擇每天看的東西,他們選擇我們的新聞,他們選擇我們的電影,他們選擇我們的衣櫥,選擇我們的浪漫伴侶,他們選擇我們的朋友……而沒人知道他們如何做它,所以沒有人負責 。
翻譯自: https://medium.com/swlh/the-ethics-of-algorithms-1c69b87a656
總結
- 上一篇: 索尼面向日美欧的相机生产转至泰国 中国工
- 下一篇: 腾讯《王者荣耀·世界》官方道歉!承认不当