完整的英文字幕见本页后半部分。另外,建议翻墙观看Youtube视频:http://www.youtube.com/watch?v=H_aLU-NOdHM,因为可以音、画、字幕同步

Christopher Steiner是《纽约时代》2009年畅销书《油价30元/升》的作者,他的新书《算法帝国》也将于2014年上半年在中国上市。2011年,Steiner与人联合创办了在线杂货零售公司Aisle50(https://aisle50.com/),该公司是国际著名创业孵化器Y Combinator的毕业生。在此之前,Steiner在《福布斯》杂志任资深技术主笔7年之久,《华尔街日报》《芝加哥论坛报》《快公司》《麻省理工科技创业》和《滑雪杂志》经常刊登他的作品。他拥有伊利诺伊大学香槟分校工学学士学位和西北大学的新闻学硕士学位。Steiner住在伊利诺伊州库克郡埃文斯顿市。

TEDx. TEDx was created in the spirit of TED's mission, "ideas worth spreading." The program is designed to give communities, organizations and individuals the opportunity to stimulate dialogue through TED-like experiences at the local level. At TEDx events, a screening of TEDTalks videos -- or a combination of live presenters and TEDTalks videos -- sparks deep conversation and connections. TEDx events are fully planned and coordinated independently, on a community-by-community basis

0:09
>>可能有人

0:09
>>一些疯狂的人会这样跟你说

0:13
>>比如一些夸夸其谈的作者

0:17
>>说什么算法正在接管世界

0:22
>>或许今天在这里,我们可用几分钟时间考量一下这种说法

0:26
>>我们知道算法已经接管了华尔街

0:31
>>控制着你的IRA(个人退休金账户)你的养老金(401 K)你的退休金的交易 70%通过算法完成

0:35
>>算法控制着你的活命钱

0:39
>>我们的股市

0:42
>>现在都是算法一层一层一层的叠加,Balbir说得好

0:47
>>我们试图在混乱中求一恒久次序

0:50
>>对人来说这几乎是不可能完成的任务

0:53
>>人做不到的,机器也做不到

0:56
>>发生在2010年5月6号的“闪电崩盘”就是例子(沃德尔和里德公司使用自动高频算法交易41亿美元期货导致)

1:00
>>万亿美元5分钟内消失

1:03
>>还有今年8月1号骑士资本暴跌亏损4.4亿美元

1:09
>>起因便是其某一算法的异常行为

1:12
>>我有一位朋友,他是基金经理

1:15
>>两年前,我们一块出去吃午饭,我问他

1:20
>>算法真的在接管金融界吗?他说,当然

1:24
>>真正有趣的是他还说算法正在接管一切

1:32
>>因为算法在华尔街的应用

1:40
>>人们便开始应用这种方法深入研究并把它推广到其他领域

1:44
>>引起了很多领域的革命

1:47
>>医药体育音乐

1:51
>>于是我们开始关注这些事,对算法产生兴趣

1:54
>>我们就像谈论自己的身体一样谈论着算法仿佛很了解它似的

1:58
>>有人会问你那个天气应用怎么样你说那不过是个算法而已

2:02
>>但算法到底是什么呢?

2:08
>>很简单算法就是用计算机语言编写的一组指令

2:11
>>它告诉计算机怎么处理信息

2:14
>>算法通过输入的一批数据产生输出结果

2:19
>>学工程的大学一、二年级的学生

2:22
>>在第一堂计算机科学上

2:25
>>都必须写花上一小时,写个井字棋游戏

2:29
>>输入是人的走法输出是机器的走法

2:35
>>但算法的能力已经不限于此了

2:39
>>算法可以学习可以适应可以进化它们已经进化到这样的地步

2:42
>>不仅是人类编写算法而算法也在改变我们

2:50
>>算法塑造我们的文化影响我们的话语

2:52
>>控制我们所听所闻改变我们的生活

2:55
>>我想大家都认为算法有其极限

2:58
>>无法跨越的极限

3:01
>>那便是最有人味的任务

3:06
>>也许没错但事实是这些领域也已然成为机器的天下

3:10
>>最有人味的任务我是指的比如研究生写论文

3:17
>>原创艺术创作至关重要的国家安全决策制定

3:21
>>书写法律文书

3:24
>>音乐界已经使用了算法来发现艺术家

3:27
>>算法非常善于发现流行歌曲

3:31
>>因为它们深谙最红的流行歌曲背后的数学原理总能百发百中地勾住你的心

3:34
>>它们越来越多地决定着电台放什么歌曲

3:40
>>但是我们得问问自己

3:41
>>是算法发现瓦尔纳的吗是算法发现披头士的吗

3:47
>>有一天你的医生会是算法

3:50
>>我们已经有了一个全自动化的药房

3:53
>>在旧金山加利福利亚大学运行着算法

3:57
>>这家药房完成了2百万份处方的配药没出一点儿差错

4:00
>>一般的人类药剂师在为同样的处方配药时可能会出2万次错

4:09
>>某一天算法也会出现在急诊室

4:13
>>问题在于这是坏事吗

4:17
>>答案尚待揭晓

4:19
>>未来20年是大数据和算法的时代

4:24
>>我们处在人类发展弧线的顶峰阶段

4:29
>>我们允许算法接管的范围有多广呢

4:33
>>可是更切合现在更切合当下的问题

4:37
>>是算法已经接管了多少领域

4:41
>>关于算法能够走多远以及它们已经走了多远我最喜欢的一个例子是

4:47
>>算法为人类做准确的心理评估

4:53
>>这通常是人类的专属领域

4:56
>>那些在医学院上了10年书的人的专属领域

4:59
>>连自己都不了解的算法怎么可能了解我们呢

5:04
>>这个故事始于50年前, 1960年代后期

5:07
>>NASA当时决定把科学家送上外太空

5:15
>>在此之前航天员都是从空军里面选拔的

5:17
>>比如试飞员

5:19
>>这些是能驾驶超音速飞机远赴重洋对苏联进行侦查的人

5:27
>>这些人沉着镇定临危不乱

5:29
>>这就是他们为什么能成为优秀的航天员

5:31
>>NASA知道他们不会被压力击垮

5:34
>>但是对科学家来说没有试飞项目

5:37
>>NASA怎么能知道哪些科学家经得住压力

5:41
>>哪些又承受不住呢

5:43
>>怎么知道何种性格的人一起呆在太空舱72小时会产生冲突矛盾呢

5:50
>>NASA清楚苏联不止一起航天任务由于宇航员之间的矛盾而夭折

5:55
>>所以NASA着手构建一个性格分类系统

6:00
>>一个可以提前预知人与人之间矛盾的预见性系统

6:05
>>可以预测在外太空谁能表现优异谁将被击垮的系统

6:11
>>NASA在此后20年的时间里创建了这个世界上最先进的心理分析系统

6:20
>>他们能通过10分钟的谈话知道你在想什么

6:22
>>你的性格如何以及你的承压能力有多大

6:29
>>而人需要数年才能掌握这些方法

6:33
>>但是这些方法的重要性以及我们今天谈论它的原因在于它是可以量化的

6:40
>>那么它是如何预测的呢

6:42
>>我们的用词我们的句子结构我们对介词和俚语的习惯用法

6:48
>>这些小细节都为探查我们内心世界及我们与他人的共事方式提供了线索

6:54
>>当然言语就好比股市背后的推动力或音乐背后的数学原理

7:00
>>言语即数据

7:02
>>可被整理归类储存分析的数据

7:05
>>所以可能有人会把NASA的科学和可应用于任何领域的算法科技结合起来

7:16
>>那么我们会在哪些地方遇到这些机器呢

7:18
>>它们在我们的生活中几乎无所不在

7:21
>>那么它们怎么知道我们在想些什么

7:23
>>它们知道你的性格知道你的言下之意

7:29
>>比如我的性格是思考支配型的

7:34
>>如果一个思考支配型的人在听别人解释某种事物

7:39
>>他也许会说

7:42
>>真有趣

7:44
>>那可真有趣

7:46
>>可我说这句话的时候真实的想法是很遗憾我不觉得有趣

7:50
>>我不知道这怎么样我需要更多的信息才能做出判断

7:55
>>而程序就知道一个思考支配型的人说有趣的时候真正的态度是什么

8:03
>>那么我们会在哪些地方遇见如此了解我们的机器呢

8:06
>>我想大家都听过那句客服电话的辞令吧

8:11
>>为了确保服务质量您的通话可能会被录音

8:15
>>我们还以为是某个上级时不时要听下谈话内容呢

8:19
>>也许吧

8:20
>>可是通常的情况你引来了6百万条算法监听你的通话

8:28
>>NASA的科学传授给了算法

8:31
>>那么它如何工作呢

8:34
>>你致电客户电话算法记下你的电话号码

8:37
>>随机为你分配一个客服代表然后监听你们的谈话

8:41
>>它们可听得很仔细

8:43
>>你刚开始说话的2分钟内它们就为你贴上了一个性格标签准确得惊人

8:49
>>下次你再打进来它们就会把你连线到具有和你相同性格的客服代表

8:56
>>它们之所以可以这样做

8:58
>>因为这些客服中心大都有1万5到2万名客服代表同时接听客服电话

9:02
>>当你连线到与自己脾性相投的客服代表你们的通话时间将缩短一半

9:07
>>而且你们达成一个满意的解决方案的几率为90%

9:11
>>而不是47%

9:14
>>客服中心所属的公司已经听过10亿通电话

9:20
>>公司花了6千万美元耗时10年构建了这个含有6百万条算法的算法库

9:25
>>对人类语言分门别类从而打造这个读心机器

9:31
>>在明尼苏达州有一个足球场大小的仓库

9:35
>>那里存储着所有这些通话这些数据

9:38
>>等着为你的大脑思维说话方式找到一个完全匹配的解读

9:44
>>如果我们可以这样处理数据应用算法

9:46
>>我们便知你是谁你为什么来电你在想什么

9:51
>>这并非遥不可及的

9:57
>>问题已不再是我们是否会通过这种方式对人们进行分类

10:00
>>问题是我们这样做的场合和频率

10:02
>>我们会以此方式挑选求职人员吗

10:06
>>我们会以此方式挑选大学申请者吗

10:09
>>甚或我们会以此方式挑选可能的伴侣吗

10:13
>>个人意见我可不希望这发生

10:17
>>我们会这样来分类孩子吗

10:20
>>每种工具其效用都有其局限性

10:24
>>我们已经看到自动化应用于华尔街的局限性

10:29
>>华尔街已经成了人类对发生之事没有洞察无法控制的场所

10:32
>>引领华尔街的人已经很难分清效用和危险的界限

10:40
>>在我们发展的道理上各个领域的数据科学家程序员宽客

10:46
>>都会遇到这样一个难题

10:48
>>如何划清效用与危险的界限

10:51
>>未来20年上演的是大数据和算法的故事

10:57
>>这界限如何画谁来画决定了这故事的走向

11:05
>>希望你们觉得这还挺有趣 (前面提到思考支配型的人这样说表明他不觉得有趣)

11:08
>>非常有趣

11:09
>>谢谢大家


0:09there are people
0:09who will tell you crazy people
0:13people like authors who tend to be prone to hyperbole
0:17that algorithms are taking over the world perhaps
0:22we should take the start and examine it is for just a few minutes here
0:26we already know of course that algorithms have taken over Wall Street
0:31they make seventy percent but the trades that control your IRAs
0:35your 401 K's your pensions the control your money
0:39all our stock markets
0:42have become one layer upon layer upon layer Balbir them's to the point
0:47where discerning order from the chaos for human
0:50is next to impossible sometimes that's impossible for the machines as well
0:56that's what happened on May 6th 2010 what we now call the flash crash
1:00when a trillion dollars disappeared in five minutes
1:03or in August 1st this year when I capital loss for did forty million
1:08dollars
1:09in 45 minutes 11 a bit algorithms went buzzer
1:12I have a friend who's a fund manager
1:15in two years ago we're out to lunch and I said on is it true
1:20algorithms are taking over he said of course
1:24but the interesting thing he said
1:27was there taking over everything now interestingly enough
1:33as algorithmic science advanced on wall street.
1:36people began taking these methods
1:41and peeling off and taking them to other places
1:44and starting many revolutions all sorts feels
1:47medicine sports music
1:50we talk about these things these algorithms we often
1:54talk about them like we know though but there are bodies some ill ask you
1:58now that the weather app or can you go just now for them
2:02what exactly is an algorithm another them
2:08quite simply is a set instructions written computer language
2:11that informs machine what to do with the piece of information
2:14over the steak input and they produce output
2:18often freshman and sophomore
2:22engineering students in the first computer science courses have to write
2:25an hour than the play tic-tac-toe that Albertans
2:29import are the moves that human the output other moves
2:33the computer but after that have gone
2:36far past this now they can learn
2:39they can adapt they can evolve they've evolved to the point in fact
2:42where we are not only shipping the algorithms they are shaping
2:46us the shaper culture
2:50shape but we see they say but we hear shape how we live
2:54I think we all match in there some wine
2:58wherever that can't get past the can't do
3:01those most human of tasks right but actually
3:06these places are now to the province the box
3:09well I mean by the most human of tasks I mean things like
3:14grading students written essays creating
3:17original art making crucial national security decisions
3:21reading legal documents music industry already employs algorithms to find you
3:27artists
3:27the very good at finding pop songs because
3:31they know the math behind the best pop hooks
3:34already deciding more more what plays
3:37on our radios we have to ask ourselves
3:41with the algorithms finer vana what they find the Beatles
3:45your doctor Sunday be an algorithm
3:49we already have a fully robotic pharmacy
3:53running and algorithms at the University of California San Francisco
3:56its doled out two million prescriptions without making a single mistake
4:00average human pharmacists would have made twenty thousand mistakes
4:06filling out the same prescriptions
4:09you will meets now for them in the emergency room Sunday
4:13the question of course is this Pat
4:17that answer has yet to be determined the story the next 20 years is a story a big
4:21data
4:22and algorithms we are at a giant
4:25fork in the archive humanity just how much
4:29we allow algorithms to take over
4:33but the better question for today the better question for right now
4:37is how much have the already taken over
4:41with my favorite examples of how far algorithms can go
4:44and how far they've already gone
4:47it's about algorithms that perform accurate psychological
4:50the valuations people this is something we normally reserved
4:54for humans who have been in medical school for 10 years
4:59how can a bot that doesn't know of itself no us
5:04the story starts fifty years ago in the late 1960s
5:07when nasa made the decision to begin sending
5:11scientists in outer space up until this point astronauts had been
5:15plucked from the Air Force they were test pilots
5:19these are the men who flew the planes that broke the speed of sound
5:22that road next to the stratosphere that spied
5:25on the Soviet Union these men were unflappable
5:29which is why they made such good astronauts nasa new
5:32would not crack under pressure but there's a test pilot program for
5:35scientists
5:37Howard nasa no which scientists would stand up to the pressure
5:40in which scientists would fall I wouldn't know which personalities would
5:44clash
5:45after being locked in a space capsule for 72 hours
5:50the russians nasa new had more than one mission compromised because a coup crew
5:54conflict
5:55so nasa set out to create a system a personality classification
6:00a predictive system that would foresee complex between people
6:04in Knowle who would perform well and who would crumble
6:09outer space the next 20 years
6:12next 20 years nasa created this system
6:15the most advanced psychoanalytic methods
6:18in the world they knew what you were thinking
6:22what your personality was what your capacity for stress was
6:25after a 10-minute conversation
6:29you could take humans years to master these methods
6:33but the important thing about these methods the reason we're talking about
6:37them today
6:38the quantifiable how they work
6:41works of the words that we speak the way in which we structure sentences
6:46the way in which you use pronouns and burt's all these little things
6:49offer clues to our inner personality how we work
6:52with other people in words of course
6:56just like the momentum behind our stock market for the math behind her music
7:00were to date did I can be organized stored
7:04parsed so it figures that somebody would come along to take mass is incredible
7:08science
7:09in Marion with the new algorithmic technologies
7:13that could employ everywhere so where do we run into the spot
7:18we're going to the nearly every day how do they know what we're thinking well
7:23when the pots know your personality
7:26they know the meaning behind your words for instance
7:30my personality something called thoughts based
7:34when I thought space person is being explained what somebody explaining
7:37something to a thought space person
7:39they might sometimes say that's interesting
7:43that's really interesting what I really mean when I say that
7:48unfortunately is that's not that interesting I don't
7:51I don't know what I think about this any more information
7:55now the pots no thats exactly
7:58what I thought space person means when he says that's interesting
8:02so we run into these pots that Noah so well
8:06well I think we're all we've all heard that refrain a customer service refrain
8:11this call may be recorded or monitored for quality assurance purposes
8:14we assume that means once in a while a boss is listening to the call
8:19and maybe yeah maybe that's what happens but often what happens
8:22is you just invited six million algorithms in for a listen
8:27nasa scientist was transferred to the set about us
8:30so how does it work well you call up customer service
8:34the over them pick no to your phone number than the route you to a random
8:38customer service agent then they settle in for a listen they listen very
8:41carefully
8:42within two minutes have you started speaking they sign you
8:45personality their credit the actor
8:49the next time you call the route you
8:52to an agent with the exact same personality is yourself and they can do
8:55this because a lot of these call centers have
8:57fifteen twenty thousand people on deck at any moment
9:01what happens when you get people with the same personnel your calls are half
9:04as long
9:07Indicom happy resolutions ninety percent of the time
9:11instead of 47 percent of the time
9:14the company behind all this is listen to one billion conversations
9:19it spent $60 million dollars over 10 years to create this library of six
9:23million algorithms
9:24to categorize the human language to build this mind reading
9:28by there's a warehouse
9:32Minnesota the size of a football field world these conversations
9:35all these data store waiting
9:39to recall that one file that matches perfectly
9:42your brain and how you talk now we can do this with data
9:45and algorithms if we can know who you are why you called
9:49and what you thinking doesn't seem to be much
9:53outside a bar reach
9:57the question isn't whether we will soar people this way the question is
10:00where and how often well we sort job applicants with these methods
10:06will we sort college applicants with these methods we even saw a potential
10:09spouse is
10:11with this of person
10:17we sort children just as with any tool there's limits
10:21utility in all of this we've seen the end utility
10:26automation on wall street it's become a place where humans have little insight
10:30as to what's going on
10:32in little control the people in charge Wall Street
10:35have had a difficult time drawing a line between utility
10:38and menace as we go forward
10:41data scientists programmers Inc wants all sorts of fields
10:46face the same dilemma where to draw the line between utility
10:50and menace the story in the next 20 years is
10:53the story big data algorithms
10:57that story will be determined by where these lines get drawn
11:01and who gets to draw them
11:05hope you found that interesting very interesting
11:08thank you