On Machine Intelligence 2（4’58）
— — Zeynep Yufekci
Machine intelligence is here.
機(jī)器智能來了义辕。
We're now using computation to make all sorts of decisions, but also new kinds of decisions.
我們不僅用計算機(jī)來做各種決策碱璃，而且包括各種新的決策螃宙。
We're asking questions to computation that have no single right answers, that are subjective and open-ended and value-laden.?
我們詢問計算機(jī)沒有單一答案的跪削，主觀的尔当、開放的蹭沛、價值取向的問題赂韵。
We're asking questions like, who should the company hire?
我們咨詢類似問題：“公司應(yīng)該雇傭誰?"
Which update from which friend should you be shown?
你的哪位朋友的更新應(yīng)該被你看到怖侦？
Which convict（定罪 有罪） is more likely to reoffend（再犯罪 再犯法）?
那個罪行更容易再犯篡悟？
Which new item or movie should be recommended to people?
那些新的事項或電影應(yīng)該被推薦給人們？
Look, yes, we've been using computers for a while, but this is different.
看匾寝，是的搬葬，我們使用計算機(jī)已經(jīng)有一段時間了，但是這是不同的艳悔。
This is a historical twist, because we can not anchor computation for such subjective decisions急凰， the way we can anchor computation for flying airplanes, building bridges, going to the moon.
這是歷史性的轉(zhuǎn)折，因?yàn)樵谶@些主觀決策上我們不能導(dǎo)向計算機(jī)猜年，我可以在飛行飛機(jī)抡锈，建造橋梁，登陸月球上錨向計算機(jī)乔外。
You know, our airplanes safer床三？
你知道，我們的飛機(jī)更安全嗎杨幼？
Did the bridges sway and fall?
橋會搖晃倒塌嗎撇簿？
There we have the great-upon（偉大的）, fairly clear benchmarks（基準(zhǔn)率）, and we have laws of nature to guide us.
我們有偉大的、非常清晰的準(zhǔn)則差购，我們有自然法則來引導(dǎo)我們四瘫。
We have no such anchors and benchmarks for decisions in messy human affairs.
在復(fù)雜的人事上我們沒有如此的錨向或準(zhǔn)則。
1. What does??Yufekci mean by historical twist?
...Computers are being used to solve subjective problems for the first time in history.
2. With the development of machine intelligence,...
....algorithms(算法 計算程序)?are now being used to answer subjective questions（主觀題）.
3. If someone reflects our personal values, it is ...value-laden(受的主觀價值影響的 主觀的).
4.排序
1）This is a historical?twist, because we can not anchor computation for such subjective decisions the way we can anchor computation for flying airplanes, building bridges, going to the moon.
2）Are our airplanes safer. ?Did the bridges sway and fall?
3）There we have the great-upon, fairly clear benchmarks, and we have laws of nature to guide us.
4）We have no such anchors and benchmarks for decisions in messy human affairs.
To make things more complicated, our software is getting more powerful, but it's also getting less transparent and more complex.
讓我們的事情變得更復(fù)雜的事欲逃，我們的軟件變得更強(qiáng)大找蜜，但是，它也變得更加不透明稳析，更加復(fù)雜洗做。
Recently, ?in the past decade, complex algorithms（算法 計算程序 ） have made great strides（大步走 大步發(fā)展）.
最近弓叛，在過去的數(shù)十年，復(fù)雜的算法取得了長足發(fā)展诚纸。
They can recognize human faces.
他們能夠識別人臉邪码。
They can decipher（破譯 辨認(rèn)）?handwriting.
他們能夠辨認(rèn)筆跡。
They can detect（發(fā)現(xiàn) 查明） credit card fraud（欺騙罪 欺詐罪）?and block spam（垃圾郵件） and they can translate ?between languages, they can detect tumors in medical imaging.
他們能夠發(fā)現(xiàn)信用卡欺詐咬清，阻擋垃圾郵件，他們可以語言翻譯奴潘，他們可以通過醫(yī)療影像發(fā)現(xiàn)腫瘤旧烧。
They can beat humans in chess and Go.
他們可以在棋類比賽中擊敗人們。
Much of this progress comes from a method called machine learning.
很多這種發(fā)展來自于一種稱作”機(jī)器學(xué)習(xí)“的方法画髓。
Machine learning is different than traditional programming, where you give the computer detailed, exact, painstaking（需細(xì)心的 辛苦的 需專注的）?instructions.?
機(jī)器學(xué)習(xí)與傳統(tǒng)程序不同掘剪，需要給計算機(jī)詳細(xì)的，精確地奈虾、精準(zhǔn)的指令夺谁。
It's more like, you take the system and you feed it lots data, including unstructured（結(jié)構(gòu)凌亂的 無條理的 紊亂的）?data, like the kind we generate in our digital（數(shù)字式的 數(shù)字顯示的）?lives.
它更像，你采用了這系統(tǒng)肉微，你投喂它很多數(shù)據(jù)匾鸥，包括非結(jié)構(gòu)化數(shù)據(jù)，例如我們數(shù)位生活中產(chǎn)生的那種碉纳。
And the system learns by churning（劇烈翻滾的 湍急的） through this data.
系統(tǒng)扎入這些數(shù)據(jù)中學(xué)習(xí)勿负。
And also, crucially, these systems don't operated under a single-answer logic.
但是，重要的是劳曹，這些系統(tǒng)也沒有在單一答案邏輯下運(yùn)行奴愉。
They don't produce a simple answer; it's more probabilistic（給予概率的 ）:" this one is probably more like what you looking for.
他們產(chǎn)生的不是簡單的答案，而是更概率性的铁孵，這個更像你正在尋找的锭硼。
Now, the upside（好的一面 正面的） is: This method is really powerful.
現(xiàn)在，好的一面是蜕劝，這個辦法真的很有力檀头。
The head of Google's AI systems called it, the unreasonable（不合理的 不公正的 期望過高的）?effectiveness of data.
谷歌AI系統(tǒng)負(fù)責(zé)人稱它，不合理的數(shù)據(jù)效率熙宇。
The downside is, we don't really understand what the system learned. In fact that's its power.
不好的一面是鳖擒，我們并不真正了解這些系統(tǒng)都學(xué)了什么。事實(shí)上這就是它的力量之所在烫止。
This is less like giving instructions to a computer, it's more like training a puppy-machine-creature we don't really understand or control.
這不像是給一臺計算機(jī)指令蒋荚，更像是訓(xùn)練一條機(jī)器小狗，我們并不了解和控制馆蠕。
So this is our problem.
所以這就是我們的問題期升。
It's a problem when this artificial intelligence system gets things wrong.
當(dāng)人工智能系統(tǒng)會出錯惊奇。
It's also a problem when it gets things right, because we don't even know which is which when it's a subjective problem.
當(dāng)他們得出正確答案時，這也是一個問題播赁。因?yàn)楫?dāng)是一個主觀問題時颂郎，我們并不知道那個是哪個
We don't know what this thing is thinking...
我們不知道這個東西它在想什么。容为。
1. Why is machine intelligence unpredictable?
..it is often unclear how it comes to its conclusions.
2. What is one characteristic of traditional programming?
.. it requires explicit instructions.
3. If a method or argument is probabilistic, it is..
....based on what is most likely to be true.
So, consider a hiring algorithm--a system used to hire people, right, using machine-learning system.
所以乓序，考慮一下招聘算法-一個使用機(jī)器學(xué)習(xí)系統(tǒng)來招聘職員的系統(tǒng)。
Such a system would have been trained on previous employee's data and instructed to find and hire people like the existing high performers in the company.
這樣的系統(tǒng)將會采用以前的員工數(shù)據(jù)進(jìn)行培訓(xùn)坎背，被指示去尋找和招聘那些公司里現(xiàn)存的表現(xiàn)非常好的職員替劈。
Sounds good.
聽起來不錯。
I once attended a conference that brought together human resources managers and executives, high-level people, using such system in hiring.
我曾參加一個會議得滤，聚集了人力資源經(jīng)理陨献、總監(jiān)、高層人士懂更，正是采用這個系統(tǒng)進(jìn)行招聘眨业。
They were super excited.
他們非常興奮。
They thought that this would make hiring more objective（客觀的 就事論事的 不帶個人情感的）, less biased, and given women and ?
minorities a better shot against biased human managers.
他們認(rèn)為這個將會使招聘更客觀沮协，少一些偏見龄捡，針對有偏見的人力資源經(jīng)理，給予女性和少數(shù)群體一個更好的機(jī)會
Look, human hiring is biased.
看慷暂，招聘職員是存在偏見的墅茉。
I know, I mean, in one of my early jobs as a programmer,
我知道，我的意思是呜呐，在我早期做程序員時就斤，
my immediate manager（直接主管）?would sometimes come down to where I was, really early in the morning or really late in the afternoon, and she'd say," Zeynep, let's go to lunch."
我的直接管管有時會屈尊來到我工作的地方，真的早晨很早或是下午很晚的時候蘑辑，她說：“澤伊內(nèi)普洋机，我們?nèi)コ晕顼埌伞?/p>
I'd be puzzled by the wired timing. It's 4pm. Lunch?
我被這奇怪的時間點(diǎn)給搞糊涂了，下午4點(diǎn)洋魂，吃午飯绷旗？
I was broke, so free lunch. I always went.
我那時很窮，所以副砍，免費(fèi)午餐衔肢，我總是會去。
I later realized what was happening.
后來我認(rèn)識到發(fā)生了什么豁翎。
My immediate managers had not confessed to their high-ups that the programmer they hired for a serious job was a teen girl who wore jean and sneakers to work.
我的直接經(jīng)理沒有向他們的高層要員坦白角骤，他們雇傭了一個穿著牛寨庫，運(yùn)動鞋的青少年小姑娘來做這項重要的編程工作心剥。
I was doing a good job, I just looked wrong and was the wrong age and gender.
我工作做的很好邦尊，我只是看起來不合適背桐，不合適的年齡和性別。
So hiring in a gender- and - race -blind way certainly sounds good to me.
所以蝉揍，在一個性別種族無法看到的方式下招聘链峭，聽起來對我是好的。
But with these system, it is more complicated, and here's why.
但是又沾，有了這些系統(tǒng)弊仪，它更復(fù)雜，這就是問什么
Currently, computational systems can infer all sorts of things about you from your digital crumbs, even if you have not disclosed those things.
當(dāng)下，計算機(jī)系統(tǒng)能從你的零星數(shù)據(jù)推斷出關(guān)于你的一切事情，甚至那些你沒有公開的事托修。
They can infer your sexual orientation, your personality traits, your political learnings.
他們能夠推斷出你的性別取向，你的性格特征，你的政治傾向喂柒。
They have predictive power with high levels of accuracy.
他們有高度精準(zhǔn)的預(yù)測能力不瓶。
Remember, for things you haven't even closed. This is inference.
記住，是那些你沒有公開的事情灾杰。這就是推斷蚊丐。
1. What does?Yufekci personal ?experience with her immediate manager suggest?
...Human bias is a problem in the workplace.
2. To make an inference means...
...to form an opinion based on the available information.
3. 選詞填空
Such a system would have been trained on previous employee's data and instructed to find and hire people like the existing high performers in the company.
4. 聽復(fù)述
Recently, in the past decade, complex algorithms have made great strides.
5. The downside is, we don't really understand what the system learned.
6. Hiring in a gender-and race-blind way certainly sounds good to me.
