TED2015
Nick Bostrom: What happens when our computers get smarter than we are?
尼克．波斯特羅姆: 當電腦比人類聰明時會發生什麼?
Filmed:
Readability: 4.4
4,632,705 views
人工智慧科技正在突飛猛進──研究顯示在這個世紀以內，一部電腦的人工智慧有可能和人類一樣聰明。在這之後，尼克．波斯特羅姆 指出，它們將會超越人類 : 「機械智慧將是人類所需要的最後一個發明。」作為一個哲學家及科學家，尼克．波斯特羅姆讓我們反思我們正在創造的世界，一個由有思考能力的機械所組成的世界。這些具有智能的機械會幫我們維持人性以及我們的價值，還是會創造出它們自己的價值觀？
Nick Bostrom - Philosopher
Nick Bostrom asks big questions: What should we do, as individuals and as a species, to optimize our long-term prospects? Will humanity’s technological advancements ultimately destroy us? Full bio
Nick Bostrom asks big questions: What should we do, as individuals and as a species, to optimize our long-term prospects? Will humanity’s technological advancements ultimately destroy us? Full bio
Double-click the English transcript below to play the video.
00:12
I work with a bunch of mathematicians,
philosophers and computer scientists,
philosophers and computer scientists,
0
570
4207
我和一群數學家、 哲學家、
及電腦科學家一起工作。
及電腦科學家一起工作。
00:16
and we sit around and think about
the future of machine intelligence,
the future of machine intelligence,
1
4777
5209
我們坐在一起
思考機器智慧的未來。
思考機器智慧的未來。
00:21
among other things.
2
9986
2044
以及其他問題。
00:24
Some people think that some of these
things are sort of science fiction-y,
things are sort of science fiction-y,
3
12030
4725
有些人可能認為
這是科幻小說的範疇，
這是科幻小說的範疇，
00:28
far out there, crazy.
4
16755
3101
離我們很遙遠，很瘋狂。
00:31
But I like to say,
5
19856
1470
但是我要說，
00:33
okay, let's look at the modern
human condition.
human condition.
6
21326
3604
好，我們來看看
現代人類的狀況....
現代人類的狀況....
00:36
(Laughter)
7
24930
1692
(觀眾笑聲)
00:38
This is the normal way for things to be.
8
26622
2402
這是人類的常態。
00:41
But if we think about it,
9
29024
2285
但如果我們仔細想想，
00:43
we are actually recently arrived
guests on this planet,
guests on this planet,
10
31309
3293
其實人類是剛剛才抵達地球的訪客
00:46
the human species.
11
34602
2082
假設地球在一年前誕生，
00:48
Think about if Earth
was created one year ago,
was created one year ago,
12
36684
4746
00:53
the human species, then,
would be 10 minutes old.
would be 10 minutes old.
13
41430
3548
人類這個物種則僅存在了10分鐘。
00:56
The industrial era started
two seconds ago.
two seconds ago.
14
44978
3168
工業革命在2秒鐘前開始。
另外一個角度是
看看這一萬年來的GDP增長
看看這一萬年來的GDP增長
01:01
Another way to look at this is to think of
world GDP over the last 10,000 years,
world GDP over the last 10,000 years,
15
49276
5225
01:06
I've actually taken the trouble
to plot this for you in a graph.
to plot this for you in a graph.
16
54501
3029
我花了時間作了張圖表，
01:09
It looks like this.
17
57530
1774
它長這個樣子
01:11
(Laughter)
18
59304
1363
(觀眾笑聲)
01:12
It's a curious shape
for a normal condition.
for a normal condition.
19
60667
2151
對一個正常的狀態來說
這是個很有趣的形狀。
這是個很有趣的形狀。
01:14
I sure wouldn't want to sit on it.
20
62818
1698
我可不想要坐在上面。
01:16
(Laughter)
21
64516
2551
(觀眾笑聲)
01:19
Let's ask ourselves, what is the cause
of this current anomaly?
of this current anomaly?
22
67067
4774
我們不禁問自己：
「是什麼造成了這種異態呢?」
「是什麼造成了這種異態呢?」
01:23
Some people would say it's technology.
23
71841
2552
有些人會說是科技
01:26
Now it's true, technology has accumulated
through human history,
through human history,
24
74393
4668
這是對的，
科技在人類歷史上不斷累積，
科技在人類歷史上不斷累積，
01:31
and right now, technology
advances extremely rapidly --
advances extremely rapidly --
25
79061
4652
而現在科技正以飛快的速度進步。
01:35
that is the proximate cause,
26
83713
1565
這個是近因，
01:37
that's why we are currently
so very productive.
so very productive.
27
85278
2565
這也是為什麼
我們現在的生產力很高。
我們現在的生產力很高。
01:40
But I like to think back further
to the ultimate cause.
to the ultimate cause.
28
88473
3661
但是我想要進一步
回想到最終的原因
回想到最終的原因
01:45
Look at these two highly
distinguished gentlemen:
distinguished gentlemen:
29
93114
3766
看看這兩位非常傑出的紳士：
01:48
We have Kanzi --
30
96880
1600
這位是坎茲先生
01:50
he's mastered 200 lexical
tokens, an incredible feat.
tokens, an incredible feat.
31
98480
4643
他掌握了200個詞彙，
這是一個難以置信的壯舉。
這是一個難以置信的壯舉。
01:55
And Ed Witten unleashed the second
superstring revolution.
superstring revolution.
32
103123
3694
以及 愛德 維騰，
他掀起了第二次超弦革命。
他掀起了第二次超弦革命。
01:58
If we look under the hood,
this is what we find:
this is what we find:
33
106817
2324
如果我們往腦袋瓜裡面看，
這是我們看到的：
這是我們看到的：
02:01
basically the same thing.
34
109141
1570
基本上是一樣的東西。
02:02
One is a little larger,
35
110711
1813
一個稍微大一點，
02:04
it maybe also has a few tricks
in the exact way it's wired.
in the exact way it's wired.
36
112524
2758
它可能有一些特別的連結方法。
02:07
These invisible differences cannot
be too complicated, however,
be too complicated, however,
37
115282
3812
但是這些無形的差異不會太複雜，
02:11
because there have only
been 250,000 generations
been 250,000 generations
38
119094
4285
因為從我們共同的祖先以來，
02:15
since our last common ancestor.
39
123379
1732
只經過了25萬代。
02:17
We know that complicated mechanisms
take a long time to evolve.
take a long time to evolve.
40
125111
3849
我們知道複雜的機制
需要很長的時間演化。
需要很長的時間演化。
02:22
So a bunch of relatively minor changes
41
130000
2499
因此 一些相對微小的變化
02:24
take us from Kanzi to Witten,
42
132499
3067
將我們從坎茲先生變成了維騰，
02:27
from broken-off tree branches
to intercontinental ballistic missiles.
to intercontinental ballistic missiles.
43
135566
4543
從撿起掉落的樹枝當武器
到發射洲際彈道飛彈
到發射洲際彈道飛彈
因此，顯而易見的是
至今我們所實現的所有事
至今我們所實現的所有事
02:32
So this then seems pretty obvious
that everything we've achieved,
that everything we've achieved,
44
140839
3935
02:36
and everything we care about,
45
144774
1378
以及我們關心的所有事物，
02:38
depends crucially on some relatively minor
changes that made the human mind.
changes that made the human mind.
46
146152
5228
都取決於人腦中相對微小的改變。
02:44
And the corollary, of course,
is that any further changes
is that any further changes
47
152650
3662
由此而來的推論就是：在未來，
02:48
that could significantly change
the substrate of thinking
the substrate of thinking
48
156312
3477
任何能顯著地改變思想基體的變化
02:51
could have potentially
enormous consequences.
enormous consequences.
49
159789
3202
都有可能會帶來巨大的後果。
02:56
Some of my colleagues
think we're on the verge
think we're on the verge
50
164321
2905
我的一些同事覺得我們即將發現
02:59
of something that could cause
a profound change in that substrate,
a profound change in that substrate,
51
167226
3908
足以深刻的改變思想基體的科技
03:03
and that is machine superintelligence.
52
171134
3213
那就是超級機器智慧
03:06
Artificial intelligence used to be
about putting commands in a box.
about putting commands in a box.
53
174347
4739
以前的人工智慧是
將指令輸入到一個箱子裡。
將指令輸入到一個箱子裡。
03:11
You would have human programmers
54
179086
1665
你需要程式設計師
03:12
that would painstakingly
handcraft knowledge items.
handcraft knowledge items.
55
180751
3135
精心的將知識設計成指令。
03:15
You build up these expert systems,
56
183886
2086
你建立這些專門系統，
03:17
and they were kind of useful
for some purposes,
for some purposes,
57
185972
2324
這些系統在某些特定的領域中有點用，
03:20
but they were very brittle,
you couldn't scale them.
you couldn't scale them.
58
188296
2681
但是它們很生硬，
你無法延展這些系統。
你無法延展這些系統。
03:22
Basically, you got out only
what you put in.
what you put in.
59
190977
3433
基本上這些系統所輸出的東西
僅限於你事先輸入的範圍。
僅限於你事先輸入的範圍。
03:26
But since then,
60
194410
997
但是從那時起，
03:27
a paradigm shift has taken place
in the field of artificial intelligence.
in the field of artificial intelligence.
61
195407
3467
人工智慧的領域裡發生了模式的轉變。
03:30
Today, the action is really
around machine learning.
around machine learning.
62
198874
2770
現在主要的課題是機器的學習。
03:34
So rather than handcrafting knowledge
representations and features,
representations and features,
63
202394
5387
因此，與其設計知識的表現及特點，
03:40
we create algorithms that learn,
often from raw perceptual data.
often from raw perceptual data.
64
208511
5554
我們寫出具有學習原始感官數據
的能力的程式碼。
的能力的程式碼。
03:46
Basically the same thing
that the human infant does.
that the human infant does.
65
214065
4998
基本上和嬰兒所做的是一樣的。
03:51
The result is A.I. that is not
limited to one domain --
limited to one domain --
66
219063
4207
結果就是不侷限於
某個領域的人工智慧 --
某個領域的人工智慧 --
03:55
the same system can learn to translate
between any pairs of languages,
between any pairs of languages,
67
223270
4631
同一個系統可以學習
在任何兩種語言之間翻譯
在任何兩種語言之間翻譯
03:59
or learn to play any computer game
on the Atari console.
on the Atari console.
68
227901
5437
或者學著玩雅達利系統上
的任何一款遊戲。
的任何一款遊戲。
04:05
Now of course,
69
233338
1779
現在當然
04:07
A.I. is still nowhere near having
the same powerful, cross-domain
the same powerful, cross-domain
70
235117
3999
人工智慧到現在
還未能達到像人類一樣
還未能達到像人類一樣
04:11
ability to learn and plan
as a human being has.
as a human being has.
71
239116
3219
具有強大的跨領域的學習能力。
04:14
The cortex still has some
algorithmic tricks
algorithmic tricks
72
242335
2126
人類大腦還具有一些運算技巧
04:16
that we don't yet know
how to match in machines.
how to match in machines.
73
244461
2355
我們不知道如何將這些技巧
複製到機器中。
複製到機器中。
04:19
So the question is,
74
247886
1899
所以現在需要問的是：
04:21
how far are we from being able
to match those tricks?
to match those tricks?
75
249785
3500
我們還要多久才能
在機器裡面複製這些技巧?
在機器裡面複製這些技巧?
04:26
A couple of years ago,
76
254245
1083
在幾年前，
04:27
we did a survey of some of the world's
leading A.I. experts,
leading A.I. experts,
77
255328
2888
我們對世界頂尖的人工智慧專家
做了一次問卷調查，
做了一次問卷調查，
04:30
to see what they think,
and one of the questions we asked was,
and one of the questions we asked was,
78
258216
3224
想要看看他們的想法，
其中的一個題目是：
其中的一個題目是：
04:33
"By which year do you think
there is a 50 percent probability
there is a 50 percent probability
79
261440
3353
"到哪一年你覺得
人類會有50%的機率
人類會有50%的機率
04:36
that we will have achieved
human-level machine intelligence?"
human-level machine intelligence?"
80
264793
3482
能夠達成人類級的人工智慧？"
04:40
We defined human-level here
as the ability to perform
as the ability to perform
81
268785
4183
我們把人類級的人工智慧
定義為有能力
定義為有能力
04:44
almost any job at least as well
as an adult human,
as an adult human,
82
272968
2871
將任何任務至少執行得
像一名成年人一樣好，
像一名成年人一樣好，
04:47
so real human-level, not just
within some limited domain.
within some limited domain.
83
275839
4005
所以是真正的人類級別,
而不是僅限於某些領域。
而不是僅限於某些領域。
04:51
And the median answer was 2040 or 2050,
84
279844
3650
而答案的中位數是2040或2050年
04:55
depending on precisely which
group of experts we asked.
group of experts we asked.
85
283494
2806
取決於我們問的專家屬於什麼群體。
04:58
Now, it could happen much,
much later, or sooner,
much later, or sooner,
86
286300
4039
當然，這個有可能過很久才實現，
也有可能提早實現
也有可能提早實現
05:02
the truth is nobody really knows.
87
290339
1940
沒有人知道確切的時間。
05:05
What we do know is that the ultimate
limit to information processing
limit to information processing
88
293259
4412
我們知道的是，
機器基體處理資訊能力的最終界限
機器基體處理資訊能力的最終界限
05:09
in a machine substrate lies far outside
the limits in biological tissue.
the limits in biological tissue.
89
297671
4871
比生物組織的界限要大的多。
05:15
This comes down to physics.
90
303241
2378
這取決於物理原理。
05:17
A biological neuron fires, maybe,
at 200 hertz, 200 times a second.
at 200 hertz, 200 times a second.
91
305619
4718
一個生物神經元發出脈衝的頻率
可能在200赫茲，每秒200次。
可能在200赫茲，每秒200次。
05:22
But even a present-day transistor
operates at the Gigahertz.
operates at the Gigahertz.
92
310337
3594
但就算是現在的電晶體
都以千兆赫(GHz)的頻率運轉。
都以千兆赫(GHz)的頻率運轉。
05:25
Neurons propagate slowly in axons,
100 meters per second, tops.
100 meters per second, tops.
93
313931
5297
神經元在軸突中傳輸的速度
比較慢，頂多是每秒100公尺。
比較慢，頂多是每秒100公尺。
05:31
But in computers, signals can travel
at the speed of light.
at the speed of light.
94
319228
3111
但在電腦裡面，信號是以光速傳播的。
05:35
There are also size limitations,
95
323079
1869
另外還有尺寸的限制
05:36
like a human brain has
to fit inside a cranium,
to fit inside a cranium,
96
324948
3027
就像人類的大腦
必需要放得進顱骨內。
必需要放得進顱骨內。
05:39
but a computer can be the size
of a warehouse or larger.
of a warehouse or larger.
97
327975
4761
但是一部電腦可以
跟倉庫一樣大，甚至更大。
跟倉庫一樣大，甚至更大。
05:44
So the potential for superintelligence
lies dormant in matter,
lies dormant in matter,
98
332736
5599
因此超級智慧的潛能
現在正潛伏在物質裡面，
現在正潛伏在物質裡面，
05:50
much like the power of the atom
lay dormant throughout human history,
lay dormant throughout human history,
99
338335
5712
就像是原子能
在人類的歷史中一直潛伏著，
在人類的歷史中一直潛伏著，
05:56
patiently waiting there until 1945.
100
344047
4405
耐心的等著，一直到1945年。
06:00
In this century,
101
348452
1248
在這個世紀內，
06:01
scientists may learn to awaken
the power of artificial intelligence.
the power of artificial intelligence.
102
349700
4118
科學家有可能會
將人工智慧的力量喚醒。
將人工智慧的力量喚醒。
06:05
And I think we might then see
an intelligence explosion.
an intelligence explosion.
103
353818
4008
屆時我覺得我們會
見證到智慧的大爆發。
見證到智慧的大爆發。
06:10
Now most people, when they think
about what is smart and what is dumb,
about what is smart and what is dumb,
104
358406
3957
大部分的人，當他們在想
什麼是聰明什麼是愚笨的時候，
什麼是聰明什麼是愚笨的時候，
06:14
I think have in mind a picture
roughly like this.
roughly like this.
105
362363
3023
我想他們腦中浮現出的畫面
會是這樣的：
會是這樣的：
06:17
So at one end we have the village idiot,
106
365386
2598
在一邊是村裡的傻子，
06:19
and then far over at the other side
107
367984
2483
然後在另外一邊
06:22
we have Ed Witten, or Albert Einstein,
or whoever your favorite guru is.
or whoever your favorite guru is.
108
370467
4756
是 愛德 維騰 或愛因斯坦，
或你喜歡的某位大師。
或你喜歡的某位大師。
06:27
But I think that from the point of view
of artificial intelligence,
of artificial intelligence,
109
375223
3834
但是我覺得從人工智慧的觀點來看，
06:31
the true picture is actually
probably more like this:
probably more like this:
110
379057
3681
真正的畫面應該比較像這樣子：
06:35
AI starts out at this point here,
at zero intelligence,
at zero intelligence,
111
383258
3378
人工智慧從這一點開始，零智慧
06:38
and then, after many, many
years of really hard work,
years of really hard work,
112
386636
3011
然後，在許多許多年
辛苦的研究以後，
辛苦的研究以後，
06:41
maybe eventually we get to
mouse-level artificial intelligence,
mouse-level artificial intelligence,
113
389647
3844
我們可能可以達到
老鼠級的人工智慧，
老鼠級的人工智慧，
06:45
something that can navigate
cluttered environments
cluttered environments
114
393491
2430
它可以在凌亂的環境中找到路
06:47
as well as a mouse can.
115
395921
1987
就像一隻老鼠一樣。
06:49
And then, after many, many more years
of really hard work, lots of investment,
of really hard work, lots of investment,
116
397908
4313
然後，在更多年的辛苦研究
及投資了很多資源之後，
及投資了很多資源之後，
06:54
maybe eventually we get to
chimpanzee-level artificial intelligence.
chimpanzee-level artificial intelligence.
117
402221
4639
我們可能可以達到
黑猩猩級的人工智慧。
黑猩猩級的人工智慧。
06:58
And then, after even more years
of really, really hard work,
of really, really hard work,
118
406860
3210
然後，在更加多年的辛苦研究之後，
07:02
we get to village idiot
artificial intelligence.
artificial intelligence.
119
410070
2913
我們達到村莊傻子級別的人工智慧。
07:04
And a few moments later,
we are beyond Ed Witten.
we are beyond Ed Witten.
120
412983
3272
然後過一小會兒後，
我們就超越了愛德維騰。
我們就超越了愛德維騰。
07:08
The train doesn't stop
at Humanville Station.
at Humanville Station.
121
416255
2970
這列火車並不會
在人類村這一站就停車。
在人類村這一站就停車。
07:11
It's likely, rather, to swoosh right by.
122
419225
3022
它比較可能會直接呼嘯而過。
07:14
Now this has profound implications,
123
422247
1984
這個具有深遠的寓意，
07:16
particularly when it comes
to questions of power.
to questions of power.
124
424231
3862
特別是在談到權力的問題。
07:20
For example, chimpanzees are strong --
125
428093
1899
舉例來說，黑猩猩很強壯 --
07:21
pound for pound, a chimpanzee is about
twice as strong as a fit human male.
twice as strong as a fit human male.
126
429992
5222
以體重比例來說， 一隻黑猩猩
比一個健康的男性人類要強壯兩倍。
比一個健康的男性人類要強壯兩倍。
07:27
And yet, the fate of Kanzi
and his pals depends a lot more
and his pals depends a lot more
127
435214
4614
然而，坎茲和他朋友們的命運
則很大的部分取決於
則很大的部分取決於
07:31
on what we humans do than on
what the chimpanzees do themselves.
what the chimpanzees do themselves.
128
439828
4140
人類的作為，
而非黑猩猩們自己的作為。
而非黑猩猩們自己的作為。
07:37
Once there is superintelligence,
129
445228
2314
當超級智慧出現後，
07:39
the fate of humanity may depend
on what the superintelligence does.
on what the superintelligence does.
130
447542
3839
人類的命運可能會取決於
超級智慧的作為。
超級智慧的作為。
07:44
Think about it:
131
452451
1057
想想看：
07:45
Machine intelligence is the last invention
that humanity will ever need to make.
that humanity will ever need to make.
132
453508
5044
機器智慧將會是人類所需要作出的
最後一個發明。
最後一個發明。
07:50
Machines will then be better
at inventing than we are,
at inventing than we are,
133
458552
2973
從那之後機器將會比人類更會發明，
07:53
and they'll be doing so
on digital timescales.
on digital timescales.
134
461525
2540
他們也將會在"數位時間"裡
做出這些事。
做出這些事。
07:56
What this means is basically
a telescoping of the future.
a telescoping of the future.
135
464065
4901
這意味著未來到來的時間將被縮短。
08:00
Think of all the crazy technologies
that you could have imagined
that you could have imagined
136
468966
3558
想想那些我們曾經想像過的瘋狂科技
08:04
maybe humans could have developed
in the fullness of time:
in the fullness of time:
137
472524
2798
人類可能在有足夠的時間下
可以發明出來：
可以發明出來：
08:07
cures for aging, space colonization,
138
475322
3258
防止衰老、殖民太空、
08:10
self-replicating nanobots or uploading
of minds into computers,
of minds into computers,
139
478580
3731
自行複製的奈米機器人，
或將我們的頭腦上載到電腦裡，
或將我們的頭腦上載到電腦裡，
08:14
all kinds of science fiction-y stuff
140
482311
2159
這一些僅存在科幻小說範疇，
08:16
that's nevertheless consistent
with the laws of physics.
with the laws of physics.
141
484470
2737
但同時還是符合物理法則的東西
08:19
All of this superintelligence could
develop, and possibly quite rapidly.
develop, and possibly quite rapidly.
142
487207
4212
超級智慧有辦法開發出這些東西，
而且速度可能很快。
而且速度可能很快。
08:24
Now, a superintelligence with such
technological maturity
technological maturity
143
492449
3558
這麼成熟的超級智慧
08:28
would be extremely powerful,
144
496007
2179
將會非常的強大，
08:30
and at least in some scenarios,
it would be able to get what it wants.
it would be able to get what it wants.
145
498186
4546
最少在某些場景它將有辦法
得到它想要的東西。
得到它想要的東西。
08:34
We would then have a future that would
be shaped by the preferences of this A.I.
be shaped by the preferences of this A.I.
146
502732
5661
這樣以來我們的未來就將會
被這個超級智慧的偏好所影響。
被這個超級智慧的偏好所影響。
08:41
Now a good question is,
what are those preferences?
what are those preferences?
147
509855
3749
現在出現了一個好問題，
這些偏好是什麼呢？
這些偏好是什麼呢？
08:46
Here it gets trickier.
148
514244
1769
這個問題更棘手。
08:48
To make any headway with this,
149
516013
1435
要在這個領域往前走，
08:49
we must first of all
avoid anthropomorphizing.
avoid anthropomorphizing.
150
517448
3276
我們必須避免
將機器智慧擬人化(人格化)。
將機器智慧擬人化(人格化)。
08:53
And this is ironic because
every newspaper article
every newspaper article
151
521934
3301
這一點很諷刺因為
每一篇關於未來的人工智慧
每一篇關於未來的人工智慧
08:57
about the future of A.I.
has a picture of this:
has a picture of this:
152
525235
3855
的報導都會有這張照片：
09:02
So I think what we need to do is
to conceive of the issue more abstractly,
to conceive of the issue more abstractly,
153
530280
4134
所以我覺得我們必須要
更抽象的來想像這個議題，
更抽象的來想像這個議題，
09:06
not in terms of vivid Hollywood scenarios.
154
534414
2790
而非以好萊塢的鮮明場景來想像。
09:09
We need to think of intelligence
as an optimization process,
as an optimization process,
155
537204
3617
我們需要把智慧看做是
一個優化的過程，
一個優化的過程，
09:12
a process that steers the future
into a particular set of configurations.
into a particular set of configurations.
156
540821
5649
一個將未來指引到
特定的組態的過程。
特定的組態的過程。
09:18
A superintelligence is
a really strong optimization process.
a really strong optimization process.
157
546470
3511
一個超級智慧
是一個很強大的優化過程。
是一個很強大的優化過程。
09:21
It's extremely good at using
available means to achieve a state
available means to achieve a state
158
549981
4117
它將很會利用現有資源
09:26
in which its goal is realized.
159
554098
1909
去達到達成目標的狀態。
09:28
This means that there is no necessary
conenction between
conenction between
160
556447
2672
這意味著有著高智慧以及
09:31
being highly intelligent in this sense,
161
559119
2734
擁有一個對人類來說
是有意義的目標之間
是有意義的目標之間
09:33
and having an objective that we humans
would find worthwhile or meaningful.
would find worthwhile or meaningful.
162
561853
4662
並沒有必然的聯繫。
09:39
Suppose we give an A.I. the goal
to make humans smile.
to make humans smile.
163
567321
3794
假設我們給予人工智慧的目標
是讓人類笑。
是讓人類笑。
09:43
When the A.I. is weak, it performs useful
or amusing actions
or amusing actions
164
571115
2982
當人工智慧比較弱時，
它會做出有用的或是好笑的動作
它會做出有用的或是好笑的動作
09:46
that cause its user to smile.
165
574097
2517
以讓使用者笑出來。
09:48
When the A.I. becomes superintelligent,
166
576614
2417
當人工智慧演化成超級智慧的時後，
09:51
it realizes that there is a more
effective way to achieve this goal:
effective way to achieve this goal:
167
579031
3523
它會體認到有更有效的方法
可以達到這個目標：
可以達到這個目標：
09:54
take control of the world
168
582554
1922
控制這個世界
09:56
and stick electrodes into the facial
muscles of humans
muscles of humans
169
584476
3162
然後在人類的臉部肌肉上連接電級
09:59
to cause constant, beaming grins.
170
587638
2941
以使這個人不斷的微笑。
10:02
Another example,
171
590579
1035
另外一個例子，
10:03
suppose we give A.I. the goal to solve
a difficult mathematical problem.
a difficult mathematical problem.
172
591614
3383
假設我們給人工智慧的目標是
解出一個非常困難的數學問題。
解出一個非常困難的數學問題。
10:06
When the A.I. becomes superintelligent,
173
594997
1937
當人工智慧變成超級智慧時，
10:08
it realizes that the most effective way
to get the solution to this problem
to get the solution to this problem
174
596934
4171
它會體認到最有效的方法是
10:13
is by transforming the planet
into a giant computer,
into a giant computer,
175
601105
2930
把整個地球轉化成
一部超大號的電腦，
一部超大號的電腦，
10:16
so as to increase its thinking capacity.
176
604035
2246
進而增加它自己的運算能力。
10:18
And notice that this gives the A.I.s
an instrumental reason
an instrumental reason
177
606281
2764
注意到這個模式
會給人工智慧理由去做
會給人工智慧理由去做
10:21
to do things to us that we
might not approve of.
might not approve of.
178
609045
2516
我們可能不認可的事情。
10:23
Human beings in this model are threats,
179
611561
1935
在這個模型裡面人類是威脅，
10:25
we could prevent the mathematical
problem from being solved.
problem from being solved.
180
613496
2921
我們可能會在解開數學問題
的過程中成為阻礙。
的過程中成為阻礙。
10:29
Of course, perceivably things won't
go wrong in these particular ways;
go wrong in these particular ways;
181
617207
3494
當然，在我們可預見的範圍內，
事情不會以這種方式出錯；
事情不會以這種方式出錯；
10:32
these are cartoon examples.
182
620701
1753
這些是誇大的例子。
10:34
But the general point here is important:
183
622454
1939
但是它指出的概念很重要：
10:36
if you create a really powerful
optimization process
optimization process
184
624393
2873
如果你創造了一個
非常強大的優化流程
非常強大的優化流程
10:39
to maximize for objective x,
185
627266
2234
要最大化目標X，
10:41
you better make sure
that your definition of x
that your definition of x
186
629500
2276
你最好確保你對目標X的定義
10:43
incorporates everything you care about.
187
631776
2469
包含了所有你所在意的事情。
10:46
This is a lesson that's also taught
in many a myth.
in many a myth.
188
634835
4384
這也是在很多神話故事中
教導的寓意。
教導的寓意。
10:51
King Midas wishes that everything
he touches be turned into gold.
he touches be turned into gold.
189
639219
5298
希臘神話中的米達斯國王希望
他碰到的所有東西都可以變成金子。
他碰到的所有東西都可以變成金子。
10:56
He touches his daughter,
she turns into gold.
she turns into gold.
190
644517
2861
他碰到了他的女兒，
她變成了黃金。
她變成了黃金。
10:59
He touches his food, it turns into gold.
191
647378
2553
他碰到了他的食物，
他的食物也變成了黃金。
他的食物也變成了黃金。
11:01
This could become practically relevant,
192
649931
2589
這實際上跟我們的題目有關，
11:04
not just as a metaphor for greed,
193
652520
2070
不僅僅是對貪婪的隱喻，
11:06
but as an illustration of what happens
194
654590
1895
但也指出了如果你創造了
一個強大的優化流程
一個強大的優化流程
11:08
if you create a powerful
optimization process
optimization process
195
656485
2837
但同時給了它
不正確或不精確的目標後
不正確或不精確的目標後
11:11
and give it misconceived
or poorly specified goals.
or poorly specified goals.
196
659322
4789
會發生什麼事。
11:16
Now you might say, if a computer starts
sticking electrodes into people's faces,
sticking electrodes into people's faces,
197
664111
5189
你可能會說，如果電腦系統
開始在人臉上安裝電極，
開始在人臉上安裝電極，
11:21
we'd just shut it off.
198
669300
2265
我們可以直接把他關掉就好了。
11:24
A, this is not necessarily so easy to do
if we've grown dependent on the system --
if we've grown dependent on the system --
199
672555
5340
一、這並不一定容易做到，如果我們
已經對這個系統產生依賴性 ——
已經對這個系統產生依賴性 ——
11:29
like, where is the off switch
to the Internet?
to the Internet?
200
677895
2732
比如：你知道網際網路的開關在哪裡嗎？
11:32
B, why haven't the chimpanzees
flicked the off switch to humanity,
flicked the off switch to humanity,
201
680627
5120
二、為什麼黑猩猩當初
沒有把人類的開關關掉？
沒有把人類的開關關掉？
11:37
or the Neanderthals?
202
685747
1551
或是尼安德特人？
11:39
They certainly had reasons.
203
687298
2666
他們有很明顯的理由要這麼做，
11:41
We have an off switch,
for example, right here.
for example, right here.
204
689964
2795
而我們的開關就在這裡：
11:44
(Choking)
205
692759
1554
(窒息聲)
11:46
The reason is that we are
an intelligent adversary;
an intelligent adversary;
206
694313
2925
原因是人類是很聰明的敵人；
11:49
we can anticipate threats
and plan around them.
and plan around them.
207
697238
2728
我們可以預見威脅
並為其做出準備。
並為其做出準備。
11:51
But so could a superintelligent agent,
208
699966
2504
但一個超級智慧也會，
11:54
and it would be much better
at that than we are.
at that than we are.
209
702470
3254
而且它的能力將比我們強大的多。
11:57
The point is, we should not be confident
that we have this under control here.
that we have this under control here.
210
705724
7187
我想要說的一點是，我們不應該
覺得一切都在我們的掌握之中。
覺得一切都在我們的掌握之中。
12:04
And we could try to make our job
a little bit easier by, say,
a little bit easier by, say,
211
712911
3447
我們可能可以藉由
把AI放到一個盒子裡面
把AI放到一個盒子裡面
12:08
putting the A.I. in a box,
212
716358
1590
來給我們更多的掌握，
12:09
like a secure software environment,
213
717948
1796
就像是一個獨立的軟體環境，
12:11
a virtual reality simulation
from which it cannot escape.
from which it cannot escape.
214
719744
3022
一個AI無法逃脫的虛擬實境。
12:14
But how confident can we be that
the A.I. couldn't find a bug.
the A.I. couldn't find a bug.
215
722766
4146
但是我們有多大的信心
這個AI不會找到漏洞？
這個AI不會找到漏洞？
12:18
Given that merely human hackers
find bugs all the time,
find bugs all the time,
216
726912
3169
就算只是人類駭客，
他們還經常找出漏洞。
他們還經常找出漏洞。
12:22
I'd say, probably not very confident.
217
730081
3036
我想我們不是很有信心。
12:26
So we disconnect the ethernet cable
to create an air gap,
to create an air gap,
218
734237
4548
那所以我們把網路線拔掉，
製造一個物理間隙，
製造一個物理間隙，
12:30
but again, like merely human hackers
219
738785
2668
但同樣的，就算只是人類駭客
12:33
routinely transgress air gaps
using social engineering.
using social engineering.
220
741453
3381
也經常可以利用社交工程陷阱
來突破物理間隙。
來突破物理間隙。
12:36
Right now, as I speak,
221
744834
1259
現在，在我在台上說話的同時
12:38
I'm sure there is some employee
out there somewhere
out there somewhere
222
746093
2389
我確定在世界的某一個角落裡
有一名公司職員
有一名公司職員
12:40
who has been talked into handing out
her account details
her account details
223
748482
3346
才剛剛被自稱來自IT部門
的人士說服(詐騙)
的人士說服(詐騙)
12:43
by somebody claiming to be
from the I.T. department.
from the I.T. department.
224
751828
2746
並交出了她的帳戶信息。
12:46
More creative scenarios are also possible,
225
754574
2127
更天馬行空的狀況也可能會發生，
12:48
like if you're the A.I.,
226
756701
1315
就像是如果你是AI，
12:50
you can imagine wiggling electrodes
around in your internal circuitry
around in your internal circuitry
227
758016
3532
你可以想像藉由擺動你體內的電路
12:53
to create radio waves that you
can use to communicate.
can use to communicate.
228
761548
3462
然後創造出無線電波，
用以與外界溝通。
用以與外界溝通。
12:57
Or maybe you could pretend to malfunction,
229
765010
2424
或這你可以假裝有故障，
12:59
and then when the programmers open
you up to see what went wrong with you,
you up to see what went wrong with you,
230
767434
3497
然後當程式設計師
把你打開檢查哪裡出錯時，
把你打開檢查哪裡出錯時，
13:02
they look at the source code -- Bam! --
231
770931
1936
他們找出了原始碼 --梆--
13:04
the manipulation can take place.
232
772867
2447
你可以在此做出操控。
13:07
Or it could output the blueprint
to a really nifty technology,
to a really nifty technology,
233
775314
3430
或這它可以做出一個
很巧妙的科技藍圖，
很巧妙的科技藍圖，
13:10
and when we implement it,
234
778744
1398
當我們實施這個藍圖後，
13:12
it has some surreptitious side effect
that the A.I. had planned.
that the A.I. had planned.
235
780142
4397
它會產生一些AI計劃好的
秘密副作用。
秘密副作用。
13:16
The point here is that we should
not be confident in our ability
not be confident in our ability
236
784539
3463
寓意是我們不能
對我們控制人工智慧的能力
對我們控制人工智慧的能力
13:20
to keep a superintelligent genie
locked up in its bottle forever.
locked up in its bottle forever.
237
788002
3808
具有太大的信心
13:23
Sooner or later, it will out.
238
791810
2254
它終究會逃脫出來，
只是時間問題而已。
只是時間問題而已。
13:27
I believe that the answer here
is to figure out
is to figure out
239
795034
3103
我覺得解方是我們需要弄清楚
13:30
how to create superintelligent A.I.
such that even if -- when -- it escapes,
such that even if -- when -- it escapes,
240
798137
5024
如何創造出一個超級智慧，
哪怕是它逃出來了，
哪怕是它逃出來了，
13:35
it is still safe because it is
fundamentally on our side
fundamentally on our side
241
803161
3277
它還是安全的，
因為它是站在我們這一邊的
因為它是站在我們這一邊的
13:38
because it shares our values.
242
806438
1899
因為它擁有了我們的價值觀。
13:40
I see no way around
this difficult problem.
this difficult problem.
243
808337
3210
我們沒有辦法避免這個艱難的問題。
13:44
Now, I'm actually fairly optimistic
that this problem can be solved.
that this problem can be solved.
244
812557
3834
但是我覺得我們可以解決這個問題。
13:48
We wouldn't have to write down
a long list of everything we care about,
a long list of everything we care about,
245
816391
3903
我們並不需要把
我們在乎的所有事物寫下來，
我們在乎的所有事物寫下來，
13:52
or worse yet, spell it out
in some computer language
in some computer language
246
820294
3643
或更麻煩的把這些事物
寫成電腦程式語言
寫成電腦程式語言
13:55
like C++ or Python,
247
823937
1454
像是 C++或 Python，
13:57
that would be a task beyond hopeless.
248
825391
2767
這是個不可能完成的任務。
14:00
Instead, we would create an A.I.
that uses its intelligence
that uses its intelligence
249
828158
4297
與其，我們可以創造出
一個人工智慧，它用它自己的智慧
一個人工智慧，它用它自己的智慧
14:04
to learn what we value,
250
832455
2771
來學習我們的價值觀，
14:07
and its motivation system is constructed
in such a way that it is motivated
in such a way that it is motivated
251
835226
5280
它的激勵機制要設計成
會讓它想要
會讓它想要
14:12
to pursue our values or to perform actions
that it predicts we would approve of.
that it predicts we would approve of.
252
840506
5232
來追求我們的價值觀或者
去做它認為我們會贊成的事情。
去做它認為我們會贊成的事情。
14:17
We would thus leverage
its intelligence as much as possible
its intelligence as much as possible
253
845738
3414
藉此我們可以最大化地
利用到它們的智慧
利用到它們的智慧
14:21
to solve the problem of value-loading.
254
849152
2745
來解決這個價值觀的問題。
14:24
This can happen,
255
852727
1512
這個是有可能的，
14:26
and the outcome could be
very good for humanity.
very good for humanity.
256
854239
3596
而且這個的結果
可對人類是非常有益的。
可對人類是非常有益的。
14:29
But it doesn't happen automatically.
257
857835
3957
但是它不會自動發生。
14:33
The initial conditions
for the intelligence explosion
for the intelligence explosion
258
861792
2998
如果我們需要控制
這個智慧的大爆炸，
這個智慧的大爆炸，
14:36
might need to be set up
in just the right way
in just the right way
259
864790
2863
那智慧大爆炸的初始條件
14:39
if we are to have a controlled detonation.
260
867653
3530
需要被正確的建立起來。
14:43
The values that the A.I. has
need to match ours,
need to match ours,
261
871183
2618
人工智慧的價值觀
要和我們的一致，
要和我們的一致，
14:45
not just in the familiar context,
262
873801
1760
並不只是在常見的狀況下，
14:47
like where we can easily check
how the A.I. behaves,
how the A.I. behaves,
263
875561
2438
比如我們可以
很簡單低檢查它的行為，
很簡單低檢查它的行為，
14:49
but also in all novel contexts
that the A.I. might encounter
that the A.I. might encounter
264
877999
3234
但也要在未來所有人工智慧
可能會遇到的情況下
可能會遇到的情況下
14:53
in the indefinite future.
265
881233
1557
保持價值觀的一致。
14:54
And there are also some esoteric issues
that would need to be solved, sorted out:
that would need to be solved, sorted out:
266
882790
4737
還有很多深奧的問題需要被解決：
14:59
the exact details of its decision theory,
267
887527
2089
它們決策概念的所有細節，
15:01
how to deal with logical
uncertainty and so forth.
uncertainty and so forth.
268
889616
2864
它如何面對解決
邏輯不確定性的情況等問題。
邏輯不確定性的情況等問題。
15:05
So the technical problems that need
to be solved to make this work
to be solved to make this work
269
893330
3102
所以技術上待解決的問題
15:08
look quite difficult --
270
896432
1113
讓這個任務看起來蠻難的 --
15:09
not as difficult as making
a superintelligent A.I.,
a superintelligent A.I.,
271
897545
3380
還沒有像做出一個超級智慧
那樣的難，
那樣的難，
15:12
but fairly difficult.
272
900925
2868
但還是挺難的。
15:15
Here is the worry:
273
903793
1695
我們所擔心的是：
15:17
Making superintelligent A.I.
is a really hard challenge.
is a really hard challenge.
274
905488
4684
創造出一個超級智慧
是一個很難的挑戰。
是一個很難的挑戰。
15:22
Making superintelligent A.I. that is safe
275
910172
2548
創造出一個安全的超級智慧
15:24
involves some additional
challenge on top of that.
challenge on top of that.
276
912720
2416
是一個更大的挑戰。
15:28
The risk is that if somebody figures out
how to crack the first challenge
how to crack the first challenge
277
916216
3487
最大的風險在於
有人想出了如何解決第一個難題
有人想出了如何解決第一個難題
15:31
without also having cracked
the additional challenge
the additional challenge
278
919703
3001
但是沒有解決第二個問題
15:34
of ensuring perfect safety.
279
922704
1901
來確保安全性萬無一失。
15:37
So I think that we should
work out a solution
work out a solution
280
925375
3331
所以我覺得我們應該先想出
15:40
to the control problem in advance,
281
928706
2822
如何"控制"的方法。
15:43
so that we have it available
by the time it is needed.
by the time it is needed.
282
931528
2660
這樣當我們需要的時候
我們可以用的到它。
我們可以用的到它。
15:46
Now it might be that we cannot solve
the entire control problem in advance
the entire control problem in advance
283
934768
3507
現在也許我們無法
完全解決「控制」的問題
完全解決「控制」的問題
15:50
because maybe some elements
can only be put in place
can only be put in place
284
938275
3024
因為有時候你要了解
你所想要控制的架構後
你所想要控制的架構後
15:53
once you know the details of the
architecture where it will be implemented.
architecture where it will be implemented.
285
941299
3997
你才能知道如何實施。
15:57
But the more of the control problem
that we solve in advance,
that we solve in advance,
286
945296
3380
但是如果我們可以
事先解決更多的難題
事先解決更多的難題
16:00
the better the odds that the transition
to the machine intelligence era
to the machine intelligence era
287
948676
4090
我們順利的進入到
機器智能時代的機率
機器智能時代的機率
16:04
will go well.
288
952766
1540
就會更高。
16:06
This to me looks like a thing
that is well worth doing
that is well worth doing
289
954306
4644
這對我來說是一個值得挑戰的事情
16:10
and I can imagine that if
things turn out okay,
things turn out okay,
290
958950
3332
而且我能想像到如果一切順利的話，
16:14
that people a million years from now
look back at this century
look back at this century
291
962282
4658
我們的後代，幾百萬年以後的人類
回顧我們這個時代的時候
回顧我們這個時代的時候
16:18
and it might well be that they say that
the one thing we did that really mattered
the one thing we did that really mattered
292
966940
4002
他們可能會說我們
所做的最重要的事就是
所做的最重要的事就是
16:22
was to get this thing right.
293
970942
1567
把這個事情弄對了。
16:24
Thank you.
294
972509
1689
謝謝
16:26
(Applause)
295
974198
2813
(觀眾掌聲)
ABOUT THE SPEAKER
Nick Bostrom - PhilosopherNick Bostrom asks big questions: What should we do, as individuals and as a species, to optimize our long-term prospects? Will humanity’s technological advancements ultimately destroy us?
Why you should listen
Philosopher Nick Bostrom envisioned a future full of human enhancement, nanotechnology and machine intelligence long before they became mainstream concerns. From his famous simulation argument -- which identified some striking implications of rejecting the Matrix-like idea that humans are living in a computer simulation -- to his work on existential risk, Bostrom approaches both the inevitable and the speculative using the tools of philosophy, probability theory, and scientific analysis.
Since 2005, Bostrom has led the Future of Humanity Institute, a research group of mathematicians, philosophers and scientists at Oxford University tasked with investigating the big picture for the human condition and its future. He has been referred to as one of the most important thinkers of our age.
Nick was honored as one of Foreign Policy's 2015 Global Thinkers .
His recent book Superintelligence advances the ominous idea that “the first ultraintelligent machine is the last invention that man need ever make.”
Nick Bostrom | Speaker | TED.com