Nicholas Christakis: How social networks predict epidemics

93,295 views ・ 2010-09-16

TED


請雙擊下方英文字幕播放視頻。

譯者: Hsin Cheng Lin 審譯者: Adrienne Lin
00:15
For the last 10 years, I've been spending my time trying to figure out
0
15260
3000
過去10年來,我試著了解,
00:18
how and why human beings
1
18260
2000
人們為何形成社交網路,
00:20
assemble themselves into social networks.
2
20260
3000
以及這些網路是如何形成的。
00:23
And the kind of social network I'm talking about
3
23260
2000
我所要談的網路,
00:25
is not the recent online variety,
4
25260
2000
並非現在所謂的網路社群。
00:27
but rather, the kind of social networks
5
27260
2000
而是更原始的社交網路,
00:29
that human beings have been assembling for hundreds of thousands of years,
6
29260
3000
自從人類在非洲大草原出現以來,
00:32
ever since we emerged from the African savannah.
7
32260
3000
已經使用這種連結十幾萬年了。
00:35
So, I form friendships and co-worker
8
35260
2000
我和其他人分享友誼、同事、
00:37
and sibling and relative relationships with other people
9
37260
3000
手足和親戚等等人際關係,
00:40
who in turn have similar relationships with other people.
10
40260
2000
這些人也和其他人有相似的連結。
00:42
And this spreads on out endlessly into a distance.
11
42260
3000
這樣的連結向外擴散,
00:45
And you get a network that looks like this.
12
45260
2000
從而得到的網路看起來會像這樣。
00:47
Every dot is a person.
13
47260
2000
每點代表一個人,
00:49
Every line between them is a relationship between two people --
14
49260
2000
兩點間的線則代表兩個人之間的關係,
00:51
different kinds of relationships.
15
51260
2000
各種不同的關係。
00:53
And you can get this kind of vast fabric of humanity,
16
53260
3000
種種的關係交織成一幅巨大的網路,
00:56
in which we're all embedded.
17
56260
2000
而我們都位於其中。
00:58
And my colleague, James Fowler and I have been studying for quite sometime
18
58260
3000
我的同事James Fowler和我花了滿長時間研究,
01:01
what are the mathematical, social,
19
61260
2000
想找到一個基於數學、社會學、
01:03
biological and psychological rules
20
63260
3000
生物學或是心理學的規則,
01:06
that govern how these networks are assembled
21
66260
2000
能夠主導這些網路的形成。
01:08
and what are the similar rules
22
68260
2000
以及是否有類似的規則
01:10
that govern how they operate, how they affect our lives.
23
70260
3000
主導網路的運作,進而影響我們的生活。
01:13
But recently, we've been wondering
24
73260
2000
直到最近,我們開始思考,
01:15
whether it might be possible to take advantage of this insight,
25
75260
3000
是否有可能利用這些發現,
01:18
to actually find ways to improve the world,
26
78260
2000
來找出增進人類福祉的方法,
01:20
to do something better,
27
80260
2000
改善現況,
01:22
to actually fix things, not just understand things.
28
82260
3000
去導正,而非只是單純理解問題。
01:25
So one of the first things we thought we would tackle
29
85260
3000
我們最先著手研究的議題,
01:28
would be how we go about predicting epidemics.
30
88260
3000
是如何預測流行趨勢。
01:31
And the current state of the art in predicting an epidemic --
31
91260
2000
目前最先進的預測方法—
01:33
if you're the CDC or some other national body --
32
93260
3000
如果你在疾病管制中心(CDC)或類似的政府單位工作—
01:36
is to sit in the middle where you are
33
96260
2000
是待在中央枯等,
01:38
and collect data
34
98260
2000
並收集資料,
01:40
from physicians and laboratories in the field
35
100260
2000
第一線的醫生和實驗室把資料傳進來,
01:42
that report the prevalence or the incidence of certain conditions.
36
102260
3000
報告疾病的流行程度或發生機率。
01:45
So, so and so patients have been diagnosed with something,
37
105260
3000
這邊有某個病患被診斷出來,
01:48
or other patients have been diagnosed,
38
108260
2000
那邊又有別人得病。
01:50
and all these data are fed into a central repository, with some delay.
39
110260
3000
資訊經過一些延遲之後,傳進中央的資料庫裡。
01:53
And if everything goes smoothly,
40
113260
2000
如果一切順利,
01:55
one to two weeks from now
41
115260
2000
一到兩個禮拜之後,
01:57
you'll know where the epidemic was today.
42
117260
3000
我們才會得知當天流行病的狀況。
02:00
And actually, about a year or so ago,
43
120260
2000
事實上一年多以前,
02:02
there was this promulgation
44
122260
2000
有人發表了這樣的概念,
02:04
of the idea of Google Flu Trends, with respect to the flu,
45
124260
3000
使用Google流感趨勢(Flu Trends)來尋找流感。
02:07
where by looking at people's searching behavior today,
46
127260
3000
透過對搜尋行為的分析,
02:10
we could know where the flu --
47
130260
2000
我們能夠得知流感發生的區域,
02:12
what the status of the epidemic was today,
48
132260
2000
得知當天傳染病的狀態,
02:14
what's the prevalence of the epidemic today.
49
134260
3000
以及傳染病的影響程度。
02:17
But what I'd like to show you today
50
137260
2000
不過這次我要介紹的方法,
02:19
is a means by which we might get
51
139260
2000
讓我們不只能夠
02:21
not just rapid warning about an epidemic,
52
141260
3000
得到傳染病的快速預警,
02:24
but also actually
53
144260
2000
更能夠讓我們
02:26
early detection of an epidemic.
54
146260
2000
提早偵測到流行病的發生。
02:28
And, in fact, this idea can be used
55
148260
2000
事實上,這個概念不止能夠
02:30
not just to predict epidemics of germs,
56
150260
3000
用來預測病菌的流行,
02:33
but also to predict epidemics of all sorts of kinds.
57
153260
3000
也能夠應用來預測各種事物的趨勢。
02:37
For example, anything that spreads by a form of social contagion
58
157260
3000
例如,任何能透過社群的方式傳播的事物,
02:40
could be understood in this way,
59
160260
2000
都可以用這種方式理解。
02:42
from abstract ideas on the left
60
162260
2000
從左邊的抽象概念,
02:44
like patriotism, or altruism, or religion
61
164260
3000
像是愛國主義、利他精神,或是宗教,
02:47
to practices
62
167260
2000
到具體的事物,
02:49
like dieting behavior, or book purchasing,
63
169260
2000
像是飲食行為、購買書籍、
02:51
or drinking, or bicycle-helmet [and] other safety practices,
64
171260
3000
酗酒、使用腳踏車安全帽等安全措施,
02:54
or products that people might buy,
65
174260
2000
或是日常用品,
02:56
purchases of electronic goods,
66
176260
2000
電子產品,
02:58
anything in which there's kind of an interpersonal spread.
67
178260
3000
任何透過人與人之間傳遞的事物。
03:01
A kind of a diffusion of innovation
68
181260
2000
這種創新的擴散,
03:03
could be understood and predicted
69
183260
2000
可以透過接下來我將展示的機制,
03:05
by the mechanism I'm going to show you now.
70
185260
3000
來理解並且預測。
03:08
So, as all of you probably know,
71
188260
2000
你們或許知道,
03:10
the classic way of thinking about this
72
190260
2000
最經典的範例,
03:12
is the diffusion-of-innovation,
73
192260
2000
就是創新的擴散,
03:14
or the adoption curve.
74
194260
2000
或是所謂的「普及曲線」。
03:16
So here on the Y-axis, we have the percent of the people affected,
75
196260
2000
Y軸是受影響人數的百分比,
03:18
and on the X-axis, we have time.
76
198260
2000
X軸表示時間的推移。
03:20
And at the very beginning, not too many people are affected,
77
200260
3000
剛開始沒有太多人受到影響,
03:23
and you get this classic sigmoidal,
78
203260
2000
然後你會看到經典的反曲線,
03:25
or S-shaped, curve.
79
205260
2000
或是S型曲線。
03:27
And the reason for this shape is that at the very beginning,
80
207260
2000
形成這種曲線的原因是,
03:29
let's say one or two people
81
209260
2000
一開始只有一兩個人
03:31
are infected, or affected by the thing
82
211260
2000
被影響,或是被「感染」,
03:33
and then they affect, or infect, two people,
83
213260
2000
然後傳遞給另外兩個人,
03:35
who in turn affect four, eight, 16 and so forth,
84
215260
3000
接著4、8、16,以此類推,
03:38
and you get the epidemic growth phase of the curve.
85
218260
3000
這時進入迅速增長的階段。
03:41
And eventually, you saturate the population.
86
221260
2000
最終擴散到整個群體。
03:43
There are fewer and fewer people
87
223260
2000
於是越來越難找到
03:45
who are still available that you might infect,
88
225260
2000
尚未被影響的人,
03:47
and then you get the plateau of the curve,
89
227260
2000
這時候曲線進入高原期,
03:49
and you get this classic sigmoidal curve.
90
229260
3000
形成整條反曲線。
03:52
And this holds for germs, ideas,
91
232260
2000
這個模式在病菌、創意、
03:54
product adoption, behaviors,
92
234260
2000
新產品的普及、行為,
03:56
and the like.
93
236260
2000
以及類似情況都適用。
03:58
But things don't just diffuse in human populations at random.
94
238260
3000
要注意的是,事物並不是隨機在人群中蔓延,
04:01
They actually diffuse through networks.
95
241260
2000
而是隨著網路分布來擴散。
04:03
Because, as I said, we live our lives in networks,
96
243260
3000
因為我們活在網路的世界,
04:06
and these networks have a particular kind of a structure.
97
246260
3000
而這種網路有特定的結構。
04:09
Now if you look at a network like this --
98
249260
2000
觀察這個網路,
04:11
this is 105 people.
99
251260
2000
裡面有105人。
04:13
And the lines represent -- the dots are the people,
100
253260
2000
每個點代表一個人
04:15
and the lines represent friendship relationships.
101
255260
2000
每條線代表彼此間的友誼關係。
04:17
You might see that people occupy
102
257260
2000
人們在這個網路中
04:19
different locations within the network.
103
259260
2000
佔據不同的位置,
04:21
And there are different kinds of relationships between the people.
104
261260
2000
彼此間有不同類型的關係。
04:23
You could have friendship relationships, sibling relationships,
105
263260
3000
可能是朋友、手足、
04:26
spousal relationships, co-worker relationships,
106
266260
3000
配偶、同事、
04:29
neighbor relationships and the like.
107
269260
3000
鄰居等等。
04:32
And different sorts of things
108
272260
2000
不同的事物會
04:34
spread across different sorts of ties.
109
274260
2000
透過不同的關係來傳播。
04:36
For instance, sexually transmitted diseases
110
276260
2000
例如,性傳染病,
04:38
will spread across sexual ties.
111
278260
2000
會藉由性伴侶的聯繫來散佈。
04:40
Or, for instance, people's smoking behavior
112
280260
2000
或者像人們吸菸,
04:42
might be influenced by their friends.
113
282260
2000
可能是受到朋友的影響。
04:44
Or their altruistic or their charitable giving behavior
114
284260
2000
人們的善行或捐助,
04:46
might be influenced by their coworkers,
115
286260
2000
可能是出自同事間的影響,
04:48
or by their neighbors.
116
288260
2000
或是他們鄰居的行為。
04:50
But not all positions in the network are the same.
117
290260
3000
但是網路中的位置並非都一樣。
04:53
So if you look at this, you might immediately grasp
118
293260
2000
這張圖或許能讓你了解,
04:55
that different people have different numbers of connections.
119
295260
3000
不同人有不同數量的連結。
04:58
Some people have one connection, some have two,
120
298260
2000
有的人一個,有人兩個,
05:00
some have six, some have 10 connections.
121
300260
3000
有人六個,有的人擁有十個連結。
05:03
And this is called the "degree" of a node,
122
303260
2000
也就是一個節點的「度數」,
05:05
or the number of connections that a node has.
123
305260
2000
或是一個節點所擁有的連結數。
05:07
But in addition, there's something else.
124
307260
2000
除此之外,
05:09
So, if you look at nodes A and B,
125
309260
2000
如果觀察節點A與B,
05:11
they both have six connections.
126
311260
2000
兩者都擁有六個連結。
05:13
But if you can see this image [of the network] from a bird's eye view,
127
313260
3000
但是如果鳥瞰整個圖像,
05:16
you can appreciate that there's something very different
128
316260
2000
你就會發現兩者之間,
05:18
about nodes A and B.
129
318260
2000
A與B的不同之處
05:20
So, let me ask you this -- I can cultivate this intuition by asking a question --
130
320260
3000
問題來了 -請用直覺回答-
05:23
who would you rather be
131
323260
2000
你比較想當誰:
05:25
if a deadly germ was spreading through the network, A or B?
132
325260
3000
如果致命病菌正在網路中散佈,A或是B?
05:28
(Audience: B.) Nicholas Christakis: B, it's obvious.
133
328260
2000
(觀眾:B)很明顯的是B。
05:30
B is located on the edge of the network.
134
330260
2000
B處在網路的邊緣。
05:32
Now, who would you rather be
135
332260
2000
現在,你比較想當誰:
05:34
if a juicy piece of gossip were spreading through the network?
136
334260
3000
如果網路中流傳著一個天大的八卦?
05:37
A. And you have an immediate appreciation
137
337260
3000
A。而且你馬上能夠理解到,
05:40
that A is going to be more likely
138
340260
2000
A會有更高的機率
05:42
to get the thing that's spreading and to get it sooner
139
342260
3000
趕上流行,而且早先一步。
05:45
by virtue of their structural location within the network.
140
345260
3000
這要歸功於他們在網路中的位置。
05:48
A, in fact, is more central,
141
348260
2000
A比較靠近中央,
05:50
and this can be formalized mathematically.
142
350260
3000
這可以用數學形式來描述。
05:53
So, if we want to track something
143
353260
2000
因此,如果我們希望追蹤某些事物
05:55
that was spreading through a network,
144
355260
3000
在網路中散佈的狀態,
05:58
what we ideally would like to do is to set up sensors
145
358260
2000
理想狀況是佈置感測器,
06:00
on the central individuals within the network,
146
360260
2000
對準網路裡的中央個體,
06:02
including node A,
147
362260
2000
包括節點A。
06:04
monitor those people that are right there in the middle of the network,
148
364260
3000
監視這些位於中心位置的人們,
06:07
and somehow get an early detection
149
367260
2000
以早期的預警到
06:09
of whatever it is that is spreading through the network.
150
369260
3000
正在網路上傳播的事物。
06:12
So if you saw them contract a germ or a piece of information,
151
372260
3000
亦即,如果這些人染病或是獲悉某些資訊,
06:15
you would know that, soon enough,
152
375260
2000
你就可以推斷,要不了多久,
06:17
everybody was about to contract this germ
153
377260
2000
所有人都會被波及,不管是染病,
06:19
or this piece of information.
154
379260
2000
或是得到資訊。
06:21
And this would be much better
155
381260
2000
這樣的作法遠勝於
06:23
than monitoring six randomly chosen people,
156
383260
2000
隨機挑選六個人來監控,
06:25
without reference to the structure of the population.
157
385260
3000
因為該做法並未考慮到群體的結構。
06:28
And in fact, if you could do that,
158
388260
2000
若是真的能夠實行,
06:30
what you would see is something like this.
159
390260
2000
我們會得到類似這樣的情況:
06:32
On the left-hand panel, again, we have the S-shaped curve of adoption.
160
392260
3000
左邊的圖表,是S型的普及曲線。
06:35
In the dotted red line, we show
161
395260
2000
我們用紅色虛線標示出,
06:37
what the adoption would be in the random people,
162
397260
2000
一般人的普及情形,
06:39
and in the left-hand line, shifted to the left,
163
399260
3000
左邊的線段,則向左偏移,
06:42
we show what the adoption would be
164
402260
2000
顯示出網路中的核心個體,
06:44
in the central individuals within the network.
165
404260
2000
他們的普及情形。
06:46
On the Y-axis is the cumulative instances of contagion,
166
406260
2000
Y軸是受到傳染「病例」的累積數量,
06:48
and on the X-axis is the time.
167
408260
2000
X軸則是時間。
06:50
And on the right-hand side, we show the same data,
168
410260
2000
右邊的圖表是相同的資料,
06:52
but here with daily incidence.
169
412260
2000
呈現的是每日的「感染」數字。
06:54
And what we show here is -- like, here --
170
414260
2000
我們想要傳達的是,
06:56
very few people are affected, more and more and more and up to here,
171
416260
2000
一開始少數人受到影響,然後越來越多直到這裡,
06:58
and here's the peak of the epidemic.
172
418260
2000
這裡就是傳播的高峰期。
07:00
But shifted to the left is what's occurring in the central individuals.
173
420260
2000
向左偏的則是在核心個體發生的情形,
07:02
And this difference in time between the two
174
422260
3000
這兩條曲線間的時間差,
07:05
is the early detection, the early warning we can get,
175
425260
3000
就是預測時差,我們可以從中得到預警,
07:08
about an impending epidemic
176
428260
2000
人群中是否有
07:10
in the human population.
177
430260
2000
即將爆發的疫情。
07:12
The problem, however,
178
432260
2000
然而問題在於,
07:14
is that mapping human social networks
179
434260
2000
人際間的社交網路,
07:16
is not always possible.
180
436260
2000
並不容易繪測。
07:18
It can be expensive, not feasible,
181
438260
2000
這樣的計畫可能所費不貲、非常困難、
07:20
unethical,
182
440260
2000
具有道德爭議
07:22
or, frankly, just not possible to do such a thing.
183
442260
3000
說實話,就是不可能。
07:25
So, how can we figure out
184
445260
2000
所以,我們要如何找出,
07:27
who the central people are in a network
185
447260
2000
網路中的核心個體在哪,
07:29
without actually mapping the network?
186
449260
3000
而無需繪出整個網路?
07:32
What we came up with
187
452260
2000
我們所想到的,
07:34
was an idea to exploit an old fact,
188
454260
2000
是利用一個既有的事實
07:36
or a known fact, about social networks,
189
456260
2000
關於社交網路,眾所皆知的事實。
07:38
which goes like this:
190
458260
2000
也就是:
07:40
Do you know that your friends
191
460260
2000
你知道你的朋友,
07:42
have more friends than you do?
192
462260
3000
所擁有的友人數目比你還多嗎?
07:45
Your friends have more friends than you do,
193
465260
3000
朋友的友人數目比自己擁有的還多,
07:48
and this is known as the friendship paradox.
194
468260
2000
通常這種情況被稱做「友誼悖論」。
07:50
Imagine a very popular person in the social network --
195
470260
2000
試想社交網路中的人氣王 -
07:52
like a party host who has hundreds of friends --
196
472260
3000
例如派對的主人,身邊有上百個朋友 --
07:55
and a misanthrope who has just one friend,
197
475260
2000
和孤僻成性,只有一個朋友的人。
07:57
and you pick someone at random from the population;
198
477260
3000
若是你隨便從人群中挑出一位,
08:00
they were much more likely to know the party host.
199
480260
2000
他們就非常有可能認識這位派對主人,
08:02
And if they nominate the party host as their friend,
200
482260
2000
而當他們舉出派對主人是自己的朋友,
08:04
that party host has a hundred friends,
201
484260
2000
由於他有上百個朋友,
08:06
therefore, has more friends than they do.
202
486260
3000
因此遠比自己的朋友數目還多。
08:09
And this, in essence, is what's known as the friendship paradox.
203
489260
3000
在本質上,這就是友誼悖論:
08:12
The friends of randomly chosen people
204
492260
3000
隨機挑選的人,他的朋友,
08:15
have higher degree, and are more central
205
495260
2000
會有較高的連結數目,也較為趨近核心,
08:17
than the random people themselves.
206
497260
2000
因而優於那些隨機挑選的人。
08:19
And you can get an intuitive appreciation for this
207
499260
2000
因此,你可以憑直覺想像,
08:21
if you imagine just the people at the perimeter of the network.
208
501260
3000
如果是那些位於網路邊緣的人,
08:24
If you pick this person,
209
504260
2000
這樣的人,
08:26
the only friend they have to nominate is this person,
210
506260
3000
他的朋友只會有這個人,
08:29
who, by construction, must have at least two
211
509260
2000
而結構上來說,這個人至少會有兩位、
08:31
and typically more friends.
212
511260
2000
甚至更多的朋友。
08:33
And that happens at every peripheral node.
213
513260
2000
在每個外圍的節點都是這樣。
08:35
And in fact, it happens throughout the network as you move in,
214
515260
3000
當你越往網路的中心移動時就越常見,
08:38
everyone you pick, when they nominate a random --
215
518260
2000
每個被你挑到的人,當他們隨意提出一個...
08:40
when a random person nominates a friend of theirs,
216
520260
3000
每當提出一個他們的朋友,
08:43
you move closer to the center of the network.
217
523260
3000
你就越靠近網路的中心。
08:46
So, we thought we would exploit this idea
218
526260
3000
於是我們認為可以利用這個概念,
08:49
in order to study whether we could predict phenomena within networks.
219
529260
3000
來研究我們是否能預測網路中所發生的現象。
08:52
Because now, with this idea
220
532260
2000
因為有了這樣的發現,
08:54
we can take a random sample of people,
221
534260
2000
我們可以從人群中隨機挑選樣本,
08:56
have them nominate their friends,
222
536260
2000
請他們指出他們的朋友,
08:58
those friends would be more central,
223
538260
2000
這些朋友會比較靠近中心,
09:00
and we could do this without having to map the network.
224
540260
3000
而我們就無須標出整個網路的圖像。
09:03
And we tested this idea with an outbreak of H1N1 flu
225
543260
3000
在哈佛大學,我們利用H1N1流感的爆發
09:06
at Harvard College
226
546260
2000
來測試這個概念。
09:08
in the fall and winter of 2009, just a few months ago.
227
548260
3000
在2009年秋冬,只有幾個月前,
09:11
We took 1,300 randomly selected undergraduates,
228
551260
3000
我們隨機挑選了1300位大學生,
09:14
we had them nominate their friends,
229
554260
2000
請這些人提供他們的朋友名單,
09:16
and we followed both the random students and their friends
230
556260
2000
我們同時追蹤了這些人和他們的朋友,
09:18
daily in time
231
558260
2000
每天為間隔,
09:20
to see whether or not they had the flu epidemic.
232
560260
3000
確認他們是否染上流感。
09:23
And we did this passively by looking at whether or not they'd gone to university health services.
233
563260
3000
除了被動觀察他們是否去健康中心報到,
09:26
And also, we had them [actively] email us a couple of times a week.
234
566260
3000
同時也要求每個禮拜Email給我們。
09:29
Exactly what we predicted happened.
235
569260
3000
結果一如我們所預期。
09:32
So the random group is in the red line.
236
572260
3000
隨機挑選的群體用紅線標示,
09:35
The epidemic in the friends group has shifted to the left, over here.
237
575260
3000
他們的朋友則向左邊偏移,在這邊。
09:38
And the difference in the two is 16 days.
238
578260
3000
兩者間的差距是16天。
09:41
By monitoring the friends group,
239
581260
2000
觀察朋友的群體,
09:43
we could get 16 days advance warning
240
583260
2000
能夠讓我們提早16天得到警示,
09:45
of an impending epidemic in this human population.
241
585260
3000
警告人群中即將爆發的傳染病。
09:48
Now, in addition to that,
242
588260
2000
除此之外,
09:50
if you were an analyst who was trying to study an epidemic
243
590260
3000
如果你是研究傳染病的分析師,
09:53
or to predict the adoption of a product, for example,
244
593260
3000
或者想要預測產品的普及情形。
09:56
what you could do is you could pick a random sample of the population,
245
596260
3000
你可以從人群中挑選隨機樣本,
09:59
also have them nominate their friends and follow the friends
246
599260
3000
請他們指出自己的朋友,
10:02
and follow both the randoms and the friends.
247
602260
3000
並且同時追蹤這兩群樣本("隨機群"和"朋友群")。
10:05
Among the friends, the first evidence you saw of a blip above zero
248
605260
3000
在朋友群中,當曲線首次開始上升...
10:08
in adoption of the innovation, for example,
249
608260
3000
...例如創新概念的普及,
10:11
would be evidence of an impending epidemic.
250
611260
2000
這個轉折便能標示出即將發生的流行趨勢。
10:13
Or you could see the first time the two curves diverged,
251
613260
3000
另一種情況是當兩條曲線首次出現分歧時,
10:16
as shown on the left.
252
616260
2000
如左圖所示。
10:18
When did the randoms -- when did the friends take off
253
618260
3000
隨機群...他們的朋友群是何時起頭,
10:21
and leave the randoms,
254
621260
2000
離開隨機群的曲線,
10:23
and [when did] their curve start shifting?
255
623260
2000
使得這條曲線開始偏移?
10:25
And that, as indicated by the white line,
256
625260
2000
從白線上可以發現,
10:27
occurred 46 days
257
627260
2000
在整體趨勢達到高峰之前,
10:29
before the peak of the epidemic.
258
629260
2000
提早了46天。
10:31
So this would be a technique
259
631260
2000
這樣的技術,
10:33
whereby we could get more than a month-and-a-half warning
260
633260
2000
可以讓我們提早一個半月得到預警,
10:35
about a flu epidemic in a particular population.
261
635260
3000
得知特定群體中感冒的流行。
10:38
I should say that
262
638260
2000
應該這樣說,
10:40
how far advanced a notice one might get about something
263
640260
2000
我們能夠多早預知事件的發生,
10:42
depends on a host of factors.
264
642260
2000
取決於幾個主要的因素。
10:44
It could depend on the nature of the pathogen --
265
644260
2000
可能由於病原的性質 -
10:46
different pathogens,
266
646260
2000
不同的病原體,
10:48
using this technique, you'd get different warning --
267
648260
2000
利用這個技術,可以得到不同的警示 -
10:50
or other phenomena that are spreading,
268
650260
2000
或是可以說,在人際網路的結構裡
10:52
or frankly, on the structure of the human network.
269
652260
3000
某些正在傳播中的現象。
10:55
Now in our case, although it wasn't necessary,
270
655260
3000
雖然並非必要,不過在這個案例中,
10:58
we could also actually map the network of the students.
271
658260
2000
我們可以將學生的網路完整描繪出來,
11:00
So, this is a map of 714 students
272
660260
2000
所以,這幅圖包含了714個學生,
11:02
and their friendship ties.
273
662260
2000
以及他們的人際關係。
11:04
And in a minute now, I'm going to put this map into motion.
274
664260
2000
接下來我會用動畫呈現這幅圖,
11:06
We're going to take daily cuts through the network
275
666260
2000
逐日推進,
11:08
for 120 days.
276
668260
2000
一共120天。
11:10
The red dots are going to be cases of the flu,
277
670260
3000
紅點代表受到感染的案例,
11:13
and the yellow dots are going to be friends of the people with the flu.
278
673260
3000
黃點則代表受感染學生的朋友,
11:16
And the size of the dots is going to be proportional
279
676260
2000
而點的大小則以比例的方式,
11:18
to how many of their friends have the flu.
280
678260
2000
呈現它周遭朋友受到傳染的數量,
11:20
So bigger dots mean more of your friends have the flu.
281
680260
3000
也就是說,越大的點代表你有越多的朋友感冒。
11:23
And if you look at this image -- here we are now in September the 13th --
282
683260
3000
觀察這張圖 -現在是9月13號-
11:26
you're going to see a few cases light up.
283
686260
2000
你會看到幾個病例亮起來。
11:28
You're going to see kind of blooming of the flu in the middle.
284
688260
2000
中心區域裡,傳染就像開花一樣向外散布。
11:30
Here we are on October the 19th.
285
690260
3000
接下來到了10月19號,
11:33
The slope of the epidemic curve is approaching now, in November.
286
693260
2000
傳染曲線開始上升,到了11月,
11:35
Bang, bang, bang, bang, bang -- you're going to see lots of blooming in the middle,
287
695260
3000
砰,砰,砰,越來越多病例在中央區域發生。
11:38
and then you're going to see a sort of leveling off,
288
698260
2000
接著情勢開始趨緩,
11:40
fewer and fewer cases towards the end of December.
289
700260
3000
越來越少人受到感染,直到十二月底。
11:43
And this type of a visualization
290
703260
2000
這種類型的圖像化資訊,
11:45
can show that epidemics like this take root
291
705260
2000
可以呈現出流行事件開始扎根,
11:47
and affect central individuals first,
292
707260
2000
先影響中心的個體,
11:49
before they affect others.
293
709260
2000
再向外擴散的全貌。
11:51
Now, as I've been suggesting,
294
711260
2000
如我之前所說的,
11:53
this method is not restricted to germs,
295
713260
3000
這套方法並不局限於病菌,
11:56
but actually to anything that spreads in populations.
296
716260
2000
可以是透過人群傳播的任何事物。
11:58
Information spreads in populations,
297
718260
2000
資訊透過人群傳遞,
12:00
norms can spread in populations,
298
720260
2000
規則能透過人群來散佈,
12:02
behaviors can spread in populations.
299
722260
2000
行為也能夠透過人群傳播
12:04
And by behaviors, I can mean things like criminal behavior,
300
724260
3000
談到行為,像是犯罪,
12:07
or voting behavior, or health care behavior,
301
727260
3000
投票,衛生習慣-
12:10
like smoking, or vaccination,
302
730260
2000
像是吸菸或是疫苗接種,
12:12
or product adoption, or other kinds of behaviors
303
732260
2000
新產品的採用,或是其他種類的行為,
12:14
that relate to interpersonal influence.
304
734260
2000
與人們之間的相互影響有關。
12:16
If I'm likely to do something that affects others around me,
305
736260
3000
如果我打算做某些事來影響周圍的人,
12:19
this technique can get early warning or early detection
306
739260
3000
這套技巧就可以提前預警,或是偵測,
12:22
about the adoption within the population.
307
742260
3000
事物在人群中的普及程度。
12:25
The key thing is that for it to work,
308
745260
2000
讓它管用的關鍵在於,
12:27
there has to be interpersonal influence.
309
747260
2000
人們之間要能互相影響,
12:29
It cannot be because of some broadcast mechanism
310
749260
2000
而非因為某種廣播機制,
12:31
affecting everyone uniformly.
311
751260
3000
使得每個人都受到相同的影響。
12:35
Now the same insights
312
755260
2000
同樣的發現,
12:37
can also be exploited -- with respect to networks --
313
757260
3000
透過網路的傳播,也能夠有
12:40
can also be exploited in other ways,
314
760260
3000
各式各樣的應用,
12:43
for example, in the use of targeting
315
763260
2000
例如,用來標示出,
12:45
specific people for interventions.
316
765260
2000
特定的目標以進行干預。
12:47
So, for example, most of you are probably familiar
317
767260
2000
舉例來說,大部分的人可能對
12:49
with the notion of herd immunity.
318
769260
2000
"群體免疫力"感到熟悉。
12:51
So, if we have a population of a thousand people,
319
771260
3000
如果這裡有一千人的群體,
12:54
and we want to make the population immune to a pathogen,
320
774260
3000
我們希望讓群體對某個病原體免疫,
12:57
we don't have to immunize every single person.
321
777260
2000
我們不需要對每個人施打疫苗。
12:59
If we immunize 960 of them,
322
779260
2000
若是讓其中960人免疫,
13:01
it's as if we had immunized a hundred [percent] of them.
323
781260
3000
效果就相當於整個群體都免疫,
13:04
Because even if one or two of the non-immune people gets infected,
324
784260
3000
因為即使一兩個沒有免疫能力的人受到感染,
13:07
there's no one for them to infect.
325
787260
2000
他也沒有人能夠傳染,
13:09
They are surrounded by immunized people.
326
789260
2000
感染者被免疫的人所圍繞。
13:11
So 96 percent is as good as 100 percent.
327
791260
3000
所以百分之96的效果相當於百分之百。
13:14
Well, some other scientists have estimated
328
794260
2000
其他的科學家估計,
13:16
what would happen if you took a 30 percent random sample
329
796260
2000
如果只靠30%的隨機樣本,
13:18
of these 1000 people, 300 people and immunized them.
330
798260
3000
30%在1000人中,也就是讓300個人免疫,
13:21
Would you get any population-level immunity?
331
801260
2000
是否能夠達到群體層次的免疫?
13:23
And the answer is no.
332
803260
3000
答案是"不能"。
13:26
But if you took this 30 percent, these 300 people
333
806260
2000
但是,如果對這30%,要300個人
13:28
and had them nominate their friends
334
808260
2000
舉出他們的朋友,
13:30
and took the same number of vaccine doses
335
810260
3000
然後用同樣數量的疫苗藥劑,
13:33
and vaccinated the friends of the 300 --
336
813260
2000
為這群300人的朋友接種,
13:35
the 300 friends --
337
815260
2000
300位朋友,
13:37
you can get the same level of herd immunity
338
817260
2000
就能夠得到相同於,讓96%的人免疫
13:39
as if you had vaccinated 96 percent of the population
339
819260
3000
所達到的群體免疫程度。
13:42
at a much greater efficiency, with a strict budget constraint.
340
822260
3000
更有效率,也節省預算。
13:45
And similar ideas can be used, for instance,
341
825260
2000
類似的概念也能用於
13:47
to target distribution of things like bed nets
342
827260
2000
物資的分配標的,例如在發展中國家
13:49
in the developing world.
343
829260
2000
蚊帳的分發方式。
13:51
If we could understand the structure of networks in villages,
344
831260
3000
若是能夠了解村落中的網路架構,
13:54
we could target to whom to give the interventions
345
834260
2000
我們就能影響關鍵的節點,
13:56
to foster these kinds of spreads.
346
836260
2000
以增進這種形式的散佈。
13:58
Or, frankly, for advertising with all kinds of products.
347
838260
3000
或是老實說,用來宣傳各式各樣的產品。
14:01
If we could understand how to target,
348
841260
2000
如果能夠了解
14:03
it could affect the efficiency
349
843260
2000
如何鎖定焦點,
14:05
of what we're trying to achieve.
350
845260
2000
就可以提高成功的效率。
14:07
And in fact, we can use data
351
847260
2000
事實上現在有數不清的來源。
14:09
from all kinds of sources nowadays [to do this].
352
849260
2000
能夠提供我們所需的資料。
14:11
This is a map of eight million phone users
353
851260
2000
這是一份歐洲國家中,
14:13
in a European country.
354
853260
2000
八百萬電話用戶的分布圖。
14:15
Every dot is a person, and every line represents
355
855260
2000
每個點代表一個用戶,每條線
14:17
a volume of calls between the people.
356
857260
2000
代表人們之間的通話量。
14:19
And we can use such data, that's being passively obtained,
357
859260
3000
我們可以利用這份被動獲得的資料,
14:22
to map these whole countries
358
862260
2000
描繪出整個國家的全貌,
14:24
and understand who is located where within the network.
359
864260
3000
並且定位每個人在網路中的位置,
14:27
Without actually having to query them at all,
360
867260
2000
而無須一個個去問,
14:29
we can get this kind of a structural insight.
361
869260
2000
從而得到對整體架構的瞭解。
14:31
And other sources of information, as you're no doubt aware
362
871260
3000
你一定也知道,其他來源的資訊
14:34
are available about such features, from email interactions,
363
874260
3000
也能提供類似的特徵,從email互動,
14:37
online interactions,
364
877260
2000
線上互動,
14:39
online social networks and so forth.
365
879260
3000
線上社群網站等等。
14:42
And in fact, we are in the era of what I would call
366
882260
2000
而事實上我們正在這樣的一個世界,
14:44
"massive-passive" data collection efforts.
367
884260
3000
「巨量-被動」的資料被收集起來。
14:47
They're all kinds of ways we can use massively collected data
368
887260
3000
我們有一大堆方法可以使用這些廣泛收集的資料,
14:50
to create sensor networks
369
890260
3000
用來建立偵測網路,
14:53
to follow the population,
370
893260
2000
用來追蹤人群,
14:55
understand what's happening in the population,
371
895260
2000
找出群體中正在發生的事件,
14:57
and intervene in the population for the better.
372
897260
3000
並且適時介入以改善情況。
15:00
Because these new technologies tell us
373
900260
2000
因為這些新的科技讓我們理解,
15:02
not just who is talking to whom,
374
902260
2000
不只是誰正和誰溝通,
15:04
but where everyone is,
375
904260
2000
還有每個人的位置所在。
15:06
and what they're thinking based on what they're uploading on the Internet,
376
906260
3000
人們在想什麼,是看他們上傳了什麼到網路上,
15:09
and what they're buying based on their purchases.
377
909260
2000
現在的購買決策受到過去購物的影響。
15:11
And all this administrative data can be pulled together
378
911260
3000
所有這樣的資料可以組織起來,
15:14
and processed to understand human behavior
379
914260
2000
經過處理以了解人類的行為,
15:16
in a way we never could before.
380
916260
3000
以一種前所未見的方式。
15:19
So, for example, we could use truckers' purchases of fuel.
381
919260
3000
舉例來說,我們可以觀察卡車司機加油,
15:22
So the truckers are just going about their business,
382
922260
2000
司機們正準備開工,
15:24
and they're buying fuel.
383
924260
2000
他們買入了汽油,
15:26
And we see a blip up in the truckers' purchases of fuel,
384
926260
3000
我們觀察到卡車司機加油的曲線開始上升,
15:29
and we know that a recession is about to end.
385
929260
2000
而能夠推估景氣即將好轉了。
15:31
Or we can monitor the velocity
386
931260
2000
或是可以透過手機,
15:33
with which people are moving with their phones on a highway,
387
933260
3000
監視高速公路上人們的移動速度,
15:36
and the phone company can see,
388
936260
2000
電信公司便能夠得知,
15:38
as the velocity is slowing down,
389
938260
2000
當移動速度下降的時候,
15:40
that there's a traffic jam.
390
940260
2000
代表可能有交通堵塞。
15:42
And they can feed that information back to their subscribers,
391
942260
3000
這些資訊便回傳給電信公司的用戶,
15:45
but only to their subscribers on the same highway
392
945260
2000
並且針對那些在同一條高速公路上,
15:47
located behind the traffic jam!
393
947260
2000
位於車陣後方的用戶!
15:49
Or we can monitor doctors prescribing behaviors, passively,
394
949260
3000
我們也可以被動監測醫生開藥的行為,
15:52
and see how the diffusion of innovation with pharmaceuticals
395
952260
3000
以了解對藥品的接受度,
15:55
occurs within [networks of] doctors.
396
955260
2000
是如何在醫生之間擴散的。
15:57
Or again, we can monitor purchasing behavior in people
397
957260
2000
我們也可以監測人們的購買行為,
15:59
and watch how these types of phenomena
398
959260
2000
觀察購買現象是如何
16:01
can diffuse within human populations.
399
961260
3000
在人群中散播的。
16:04
And there are three ways, I think,
400
964260
2000
我認為這些巨量-被動收集所得的資料,
16:06
that these massive-passive data can be used.
401
966260
2000
有三種方式可以利用。
16:08
One is fully passive,
402
968260
2000
一種是完全的被動,
16:10
like I just described --
403
970260
2000
像我剛剛所描述的 -
16:12
as in, for instance, the trucker example,
404
972260
2000
例如卡車司機的例子,
16:14
where we don't actually intervene in the population in any way.
405
974260
2000
我們並不對群體做任何形式的干預。
16:16
One is quasi-active,
406
976260
2000
一種是半主動,
16:18
like the flu example I gave,
407
978260
2000
像是之前流感的例子,
16:20
where we get some people to nominate their friends
408
980260
3000
我們讓某些人舉出他們的朋友,
16:23
and then passively monitor their friends --
409
983260
2000
然後被動的觀察他們的朋友 -
16:25
do they have the flu, or not? -- and then get warning.
410
985260
2000
他們感冒了沒?- 並據此取得預警。
16:27
Or another example would be,
411
987260
2000
另外一個例子是,
16:29
if you're a phone company, you figure out who's central in the network
412
989260
3000
電信公司可以想辦法找出網路的中心群,
16:32
and you ask those people, "Look, will you just text us your fever every day?
413
992260
3000
問他們,"你能不能每天用簡訊,讓我們知道你發燒了沒?
16:35
Just text us your temperature."
414
995260
2000
只要傳送體溫即可"
16:37
And collect vast amounts of information about people's temperature,
415
997260
3000
然後從中心群體裡,
16:40
but from centrally located individuals.
416
1000260
2000
大量收集體溫資料,
16:42
And be able, on a large scale,
417
1002260
2000
便能夠用少量的資料輸入,
16:44
to monitor an impending epidemic
418
1004260
2000
來進行大規模的監控,
16:46
with very minimal input from people.
419
1006260
2000
以預測流感的爆發。
16:48
Or, finally, it can be more fully active --
420
1008260
2000
最後是完全主動的方式 -
16:50
as I know subsequent speakers will also talk about today --
421
1010260
2000
就我所知下位演講者也會談到-
16:52
where people might globally participate in wikis,
422
1012260
2000
現在全世界的人都參與維基百科的編寫、
16:54
or photographing, or monitoring elections,
423
1014260
3000
拍攝照片、或是監視選舉,
16:57
and upload information in a way that allows us to pool
424
1017260
2000
人們將資訊上傳,使得我們能夠匯集
16:59
information in order to understand social processes
425
1019260
2000
資訊以了解社會進程,
17:01
and social phenomena.
426
1021260
2000
以及社會現象的產生。
17:03
In fact, the availability of these data, I think,
427
1023260
2000
我認為這些資料的垂手可得,
17:05
heralds a kind of new era
428
1025260
2000
揭示了一個新時代的來臨,
17:07
of what I and others would like to call
429
1027260
2000
我們將之稱作
17:09
"computational social science."
430
1029260
2000
"計算社會科學"。
17:11
It's sort of like when Galileo invented -- or, didn't invent --
431
1031260
3000
有點類似伽利略發明 -或許沒有發明-
17:14
came to use a telescope
432
1034260
2000
望遠鏡的誕生,
17:16
and could see the heavens in a new way,
433
1036260
2000
而可以從全新的角度來觀看天空。
17:18
or Leeuwenhoek became aware of the microscope --
434
1038260
2000
或是雷文霍克發現顯微鏡 -
17:20
or actually invented --
435
1040260
2000
或許是他發明的-
17:22
and could see biology in a new way.
436
1042260
2000
而能夠用新的方式看待生物學。
17:24
But now we have access to these kinds of data
437
1044260
2000
現在我們能夠取得的資料,
17:26
that allow us to understand social processes
438
1046260
2000
能夠讓我們用過去未見的嶄新角度
17:28
and social phenomena
439
1048260
2000
了解社會的進程,
17:30
in an entirely new way that was never before possible.
440
1050260
3000
以及其中發生的現象。
17:33
And with this science, we can
441
1053260
2000
有了這樣的科學,
17:35
understand how exactly
442
1055260
2000
我們就能夠了解
17:37
the whole comes to be greater
443
1057260
2000
群體的綜效,是如何優於
17:39
than the sum of its parts.
444
1059260
2000
單純個體的加總。
17:41
And actually, we can use these insights
445
1061260
2000
我們也能運用這些理解,
17:43
to improve society and improve human well-being.
446
1063260
3000
來增進社會以及人類的福祉。
17:46
Thank you.
447
1066260
2000
謝謝。
關於本網站

本網站將向您介紹對學習英語有用的 YouTube 視頻。 您將看到來自世界各地的一流教師教授的英語課程。 雙擊每個視頻頁面上顯示的英文字幕,從那裡播放視頻。 字幕與視頻播放同步滾動。 如果您有任何意見或要求,請使用此聯繫表與我們聯繫。

https://forms.gle/WvT1wiN1qDtmnspy7