數據科學 (Data Science)

Nina was close to tears when she accused Nick Gibb of ruining her life. Nina is an 18 year old about to leave school and go on to higher education; Gibb is the UK government’s schools minister.

妮娜(Nina)指責尼克·吉布(Nick Gibb)毀了自己的生活時差點流淚。 妮娜(Nina)今年18歲，即將離開學校繼續接受高等教育； 吉布(Gibb)是英國政府的學校部長。

The occasion was a live BBC radio programme called “Any Questions”, a popular show where politicians and others discuss questions posed by the public.

場合是英國廣播公司(BBC)的直播節目“任何問題”，這是一個受歡迎的節目，政界人士和其他人討論公眾提出的問題。

Nina was on track to be accepted by the most prestigious veterinary college in the country; she just needed to get the right grades in her A-level exams and her teachers predicted that she would indeed get those grades.

妮娜(Nina)有望被該國最負盛名的獸醫學院錄取； 她只需要在A級考試中取得正確的成績即可，她的老師預測她的確會取得這些成績。

But, because of the pandemic, the exams never took place and instead the government sanctioned the use of an algorithm to assign grades to students, instead.

但是，由于大流行，考試從未進行過，相反，政府批準使用一種算法為學生分配成績。

This algorithm reduced the grades for nearly 40% of students. Nina’s were so much lower than the ones predicted for her that, not only would she not be able to get into her preferred college, but no other college would accept her, either.

該算法降低了將近40％的學生的成績。 妮娜(Nina)的學歷比她預期的要低得多，不僅她不能進入自己喜歡的大學，而且其他任何大學也不會接受她。

Mr Gibb was reassuring and told the audibly upset Nina that this was a mistake that would be put right.

Gibb先生放心，并向Nina傾訴不安，這是可以糾正的錯誤。

And it was. A few days later, after much argument, protest and controversy, and to the great relief of many students who were in a similar situation to Nina, the algorithm was dumped in favour of student grades predicted by schools.

是的。 幾天后，經過多次爭論，抗議和爭議，并使處于與Nina相似境地的許多學生大為欣慰的是，該算法被棄用，以支持學校預測的學生成績。

So how did the government get itself into this mess?

那么，政府如何陷入困境呢？

The concern was that just using predicted grades would result in grade inflation because teachers tend to be more optimistic about their students’ abilities than is borne out in actual exam results.

令人擔心的是，僅使用預測的成績會導致成績膨脹，因為與實際考試結果相比，教師更傾向于對學生的能力持樂觀態度。

So an algorithm was devised that was based, not so much on the academic record of the individual student, but on the record of the school. A results profile was constructed for the school — how many As did previous cohorts achieve, how many Bs, C’s and so on — and the individual students were positioned on that profile. The ‘exam results’ were then calculated for each student depending on their position in the profile.

因此，設計了一種算法，該算法不僅基于單個學生的學習成績，而且還基于學校的成績。 為學校構建了一個結果概要文件-以前的隊列達到了多少，B，C的數目等等，并且每個學生都位于該概要文件上。 然后根據每個學生在個人資料中的位置為他們計算“考試結果”。

Using statistics from previous years to predict future results is not an unusual thing to do. Indeed, statistical models like these are very useful for government and corporate planning.

使用前幾年的統計數據來預測未來的結果并不是一件尋常的事情。 確實，像這樣的統計模型對于政府和公司計劃非常有用。

However, while the predicted results may have been in line with previous years and might well reflect what the outcome of this year’s exam results might have been, generally, there are at least two glaring problems.

但是，盡管預測結果可能與前幾年保持一致，并且可能很好地反映了今年考試結果的結果，但通常至少存在兩個明顯的問題。

The first is that some schools are getting better, so judging their performance on the previous year would produce a worse result than is fair. This would affect all of the students at that school.

首先是一些學校的狀況正在好轉，因此，從上一年的表現來看，其結果會不公平。 這將影響該學校的所有學生。

Secondly, not all cohorts are the same. There will always be a number of high flyers in a school but this number will vary. If there are fewer in a particular school this year, then the statistical approach taken would artificially promote some more mediocre students into that category. They probably won’t complain.

其次，并不是所有的隊列都是一樣的。 在學校中，總是會有很多高級傳單，但是這個數字會有所不同。 如果今年某所學校的學生人數減少，那么采用的統計方法將人為地將一些中等水平的學生晉升為該類別。 他們可能不會抱怨。

But if there are a larger number of gifted students in this year’s cohort then they will be unfairly downgraded and, like Nina, be awarded grades below those that they could have achieved in a real exam.

但是，如果今年的隊列中有更多的天才學生，那么他們將被不公平地降級，并且像Nina一樣，被授予的分數低于他們在真實考試中可以達到的成績。

These additional gifted students may well be cheated out of their place at a good university or college.

這些額外的有天賦的學生很可能會被一所好的大學或學院騙走。

In a further twist, the algorithm was not applied to small cohorts because the statistics for these smaller groups are not reliable. In the small classes, which are often private schools, the teacher predicted grades were used. Thus smaller private schools got the benefit of generous teacher assessment while the larger state schools did not. Not an equitable situation.

更進一步，該算法不適用于較小的同類群，因為這些較小的群體的統計數據不可靠。 在通常是私立學校的小班教學中，使用了老師預測的成績。 因此，較小的私立學校獲得了慷慨的教師評估的好處，而較大的公立學校則沒有。 并非公平的情況。

So should we chuck out algorithms altogether? No. There are many situations where the use of algorithms and statistics are entirely suitable for prediction purposes. How many beds should a hospital leave free during the flu season? Previous numbers of flu cases can inform the calculation. When Netflix or Amazon recommend a movie to watch or a product to buy, they are using statistics from people who they think are like you. But, if they get it wrong you end up watching a movie that you don’t enjoy or rejecting a product that you don’t want to buy. This is not life changing.

那么我們應該完全放棄算法嗎？ 不可以。在許多情況下，算法和統計信息的使用完全適合預測目的。 流感季節醫院應該騰出幾張床？ 以前的流感病例數可以為計算提供依據。 當Netflix或Amazon建議觀看電影或購買產品時，他們會使用他們認為與您相似的人的統計數據。 但是，如果他們弄錯了，那么您最終會看不喜歡的電影，或者拒絕您不想購買的產品。 這不會改變生活。

When statistics are used to make decisions that are profound, using someone else’s data just won’t do. It might produce an overall pattern that is satisfying to the designers, or policy makers, but it can unfairly disadvantage individuals.

當使用統計數據做出有意義的決策時，僅使用其他人的數據是行不通的。 它可能會產生令設計者或決策者滿意的整體模式，但可能不公平地使個人處于不利地位。

Making decisions about someone’s personal life by using data from people from a similar background is wrong and unethical. Nina’s grades were calculated by looking at students who were similar to her. But they were not her. If such important decisions are to be made about someone’s life then it is only that person’s data that should be taken into account.

通過使用來自相似背景的人的數據來決定某人的個人生活是錯誤和不道德的。 妮娜的成績是通過查看與她相似的學生得出的。 但是他們不是她。 如果要對某人的生活做出如此重要的決定，那么僅應考慮該人的數據。

And maybe it should not be an algorithm that makes that decision.

也許它不應該是做出決定的算法。