TED2018
Max Tegmark: How to get empowered, not overpowered, by AI
麥克斯·泰格馬克: 如何讓人工智慧賦予我們力量，而非受控於它
Filmed:
Readability: 5
1,409,007 views
許多研究者預期人工智慧將會於數十年內在所有的工作上勝過我們，而唯一的限制在於物理法則，不是智慧。麻省理工學院的物理學家和人工智慧研究者麥克斯·泰格馬克區隔迷思和真正的機會，說明我們現今應該採取哪些步驟來確保人工智慧最終能為人類帶來最好（而非最糟）的結果。
Max Tegmark - Scientist, author
Max Tegmark is driven by curiosity, both about how our universe works and about how we can use the science and technology we discover to help humanity flourish rather than flounder. Full bio
Max Tegmark is driven by curiosity, both about how our universe works and about how we can use the science and technology we discover to help humanity flourish rather than flounder. Full bio
Double-click the English transcript below to play the video.
00:12
After 13.8 billion years
of cosmic history,
of cosmic history,
0
760
4416
在 138 億年的宇宙歷史之後，
00:17
our universe has woken up
1
5200
2096
我們的宇宙終於覺醒了，
00:19
and become aware of itself.
2
7320
1520
開始意識到自己。
00:21
From a small blue planet,
3
9480
1936
從一顆藍色的小星球，
00:23
tiny, conscious parts of our universe
have begun gazing out into the cosmos
have begun gazing out into the cosmos
4
11440
4136
我們宇宙中微小、有意識的部分
開始用望遠鏡窺視宇宙，
00:27
with telescopes,
5
15600
1376
00:29
discovering something humbling.
6
17000
1480
有了讓人謙卑的發現。
00:31
We've discovered that our universe
is vastly grander
is vastly grander
7
19320
2896
我們發現宇宙比我們祖先所想像的
00:34
than our ancestors imagined
8
22240
1336
要大很多很多，
00:35
and that life seems to be an almost
imperceptibly small perturbation
imperceptibly small perturbation
9
23600
4256
而且生命似乎是非常小的擾動，
小到幾乎無法感知到，
小到幾乎無法感知到，
00:39
on an otherwise dead universe.
10
27880
1720
若沒有它，宇宙就是死寂的了。
00:42
But we've also discovered
something inspiring,
something inspiring,
11
30320
3016
但我們也有振奮人心的發現：
00:45
which is that the technology
we're developing has the potential
we're developing has the potential
12
33360
2976
那就是我們所開發的技術，
有潛能可以協助生命
變得前所未有的繁盛，
變得前所未有的繁盛，
00:48
to help life flourish like never before,
13
36360
2856
00:51
not just for centuries
but for billions of years,
but for billions of years,
14
39240
3096
不是只有數世紀，
而是能持續數十億年。
而是能持續數十億年。
00:54
and not just on earth but throughout
much of this amazing cosmos.
much of this amazing cosmos.
15
42360
4120
不只是在地球上，
還是在這整個不可思議的宇宙中。
還是在這整個不可思議的宇宙中。
00:59
I think of the earliest life as "Life 1.0"
16
47680
3336
我把最早的生命
視為是「生命 1.0」，
視為是「生命 1.0」，
01:03
because it was really dumb,
17
51040
1376
因為它其實很蠢，
01:04
like bacteria, unable to learn
anything during its lifetime.
anything during its lifetime.
18
52440
4296
像細菌，它在一生中
無法學習任何東西。
無法學習任何東西。
01:08
I think of us humans as "Life 2.0"
because we can learn,
because we can learn,
19
56760
3376
我把人類視為是「生命 2.0」，
因為我們可以學習，
因為我們可以學習，
01:12
which we in nerdy, geek speak,
20
60160
1496
用很宅的方式來說，
01:13
might think of as installing
new software into our brains,
new software into our brains,
21
61680
3216
可以視為是把新軟體
安裝到我們的大腦中，
安裝到我們的大腦中，
01:16
like languages and job skills.
22
64920
2120
就像語言以及工作技能。
01:19
"Life 3.0," which can design not only
its software but also its hardware
its software but also its hardware
23
67680
4296
「生命 3.0」不只能設計
它的軟體，還能設計硬體，
它的軟體，還能設計硬體，
01:24
of course doesn't exist yet.
24
72000
1656
當然，它還不存在。
01:25
But perhaps our technology
has already made us "Life 2.1,"
has already made us "Life 2.1,"
25
73680
3776
但，也許我們的科技已經
讓我們成為「生命 2.1」了，
讓我們成為「生命 2.1」了，
01:29
with our artificial knees,
pacemakers and cochlear implants.
pacemakers and cochlear implants.
26
77480
4336
因為我們現在有人工膝蓋、
心律調節器，以及耳蝸植入。
心律調節器，以及耳蝸植入。
01:33
So let's take a closer look
at our relationship with technology, OK?
at our relationship with technology, OK?
27
81840
3880
咱們來更進一步談談
我們與科技的關係，好嗎？
我們與科技的關係，好嗎？
01:38
As an example,
28
86800
1216
舉個例子，
01:40
the Apollo 11 moon mission
was both successful and inspiring,
was both successful and inspiring,
29
88040
5296
阿波羅 11 月球任務
很成功也很鼓舞人心，
很成功也很鼓舞人心，
01:45
showing that when we humans
use technology wisely,
use technology wisely,
30
93360
3016
展示出當我們人類聰明地使用科技時，
01:48
we can accomplish things
that our ancestors could only dream of.
that our ancestors could only dream of.
31
96400
3936
我們能達成祖先只能夢想的事情。
01:52
But there's an even more inspiring journey
32
100360
2976
但還有一趟更鼓舞人心的旅程，
01:55
propelled by something
more powerful than rocket engines,
more powerful than rocket engines,
33
103360
2680
由比火箭引擎
更強大的東西所推動，
更強大的東西所推動，
01:59
where the passengers
aren't just three astronauts
aren't just three astronauts
34
107200
2336
乘客也不只是三個太空人，
02:01
but all of humanity.
35
109560
1776
而是全人類。
02:03
Let's talk about our collective
journey into the future
journey into the future
36
111360
2936
咱們來談談我們全體與人工智慧
02:06
with artificial intelligence.
37
114320
2000
一起前往未來的旅程，
02:08
My friend Jaan Tallinn likes to point out
that just as with rocketry,
that just as with rocketry,
38
116960
4536
我的朋友楊·塔林常說，
就像火箭學一樣，
就像火箭學一樣，
光是讓我們的科技
有強大的力量是不足夠的。
有強大的力量是不足夠的。
02:13
it's not enough to make
our technology powerful.
our technology powerful.
39
121520
3160
02:17
We also have to figure out,
if we're going to be really ambitious,
if we're going to be really ambitious,
40
125560
3175
若我們真的很有野心的話，
我們還得要想出
我們還得要想出
02:20
how to steer it
41
128759
1416
如何操控它，
02:22
and where we want to go with it.
42
130199
1681
及我們要和它一起去到哪裡。
02:24
So let's talk about all three
for artificial intelligence:
for artificial intelligence:
43
132880
2840
所以，咱們針對人工智慧
來談談這三點：
來談談這三點：
02:28
the power, the steering
and the destination.
and the destination.
44
136440
3056
力量、操控，以及目的地。
02:31
Let's start with the power.
45
139520
1286
咱們先從力量談起。
02:33
I define intelligence very inclusively --
46
141600
3096
我對於人工智慧的定義非常全面，
02:36
simply as our ability
to accomplish complex goals,
to accomplish complex goals,
47
144720
4336
就是我們能夠完成
複雜目標的能力，
複雜目標的能力，
02:41
because I want to include both
biological and artificial intelligence.
biological and artificial intelligence.
48
149080
3816
因為我想要把生物智慧
和人工智慧都包含進去。
和人工智慧都包含進去。
02:44
And I want to avoid
the silly carbon-chauvinism idea
the silly carbon-chauvinism idea
49
152920
4016
且我想要排除愚蠢的碳沙文主義，
它認為若你很聰明，
你就一定是肉做的。
你就一定是肉做的。
02:48
that you can only be smart
if you're made of meat.
if you're made of meat.
50
156960
2360
02:52
It's really amazing how the power
of AI has grown recently.
of AI has grown recently.
51
160880
4176
人工智慧的力量
在近期的成長十分驚人。
在近期的成長十分驚人。
02:57
Just think about it.
52
165080
1256
試想看看。
02:58
Not long ago, robots couldn't walk.
53
166360
3200
沒多久以前，機器人還無法走路。
03:03
Now, they can do backflips.
54
171040
1720
現在，它們還會後翻。
03:06
Not long ago,
55
174080
1816
沒多久以前，
03:07
we didn't have self-driving cars.
56
175920
1760
我們還沒有自動駕駛的汽車。
03:10
Now, we have self-flying rockets.
57
178920
2480
現在，我們有自動飛行的火箭。
03:15
Not long ago,
58
183960
1416
沒多久以前，
03:17
AI couldn't do face recognition.
59
185400
2616
人工智慧還無法做臉孔辨識。
03:20
Now, AI can generate fake faces
60
188040
2976
現在，人工智慧能產生出假臉孔，
03:23
and simulate your face
saying stuff that you never said.
saying stuff that you never said.
61
191040
4160
並模擬你的臉孔，
說出你從來沒有說過的話。
說出你從來沒有說過的話。
03:28
Not long ago,
62
196400
1576
沒多久以前，
03:30
AI couldn't beat us at the game of Go.
63
198000
1880
人工智慧無法在圍棋中打敗人類。
03:32
Then, Google DeepMind's AlphaZero AI
took 3,000 years of human Go games
took 3,000 years of human Go games
64
200400
5096
接著， Google DeepMind
AlphaZero 的人工智慧
AlphaZero 的人工智慧
拿來人類三千年的圍棋譜和圍棋智慧，
03:37
and Go wisdom,
65
205520
1256
全部擺在一邊，
03:38
ignored it all and became the world's best
player by just playing against itself.
player by just playing against itself.
66
206800
4976
透過和自己比賽的練習，
變成了世界上最厲害的圍棋手。
變成了世界上最厲害的圍棋手。
03:43
And the most impressive feat here
wasn't that it crushed human gamers,
wasn't that it crushed human gamers,
67
211800
3696
這裡最讓人印象深刻的功績
並不是它擊垮了人類的棋手，
並不是它擊垮了人類的棋手，
03:47
but that it crushed human AI researchers
68
215520
2576
而是它擊垮了人類的
人工智慧研究者，
人工智慧研究者，
03:50
who had spent decades
handcrafting game-playing software.
handcrafting game-playing software.
69
218120
3680
這些研究者花了數十年
手工打造下棋軟體。
手工打造下棋軟體。
03:54
And AlphaZero crushed human AI researchers
not just in Go but even at chess,
not just in Go but even at chess,
70
222200
4656
除了圍棋，AlphaZero 也在西洋棋
擊垮了人類的人工智慧研究者，
擊垮了人類的人工智慧研究者，
03:58
which we have been working on since 1950.
71
226880
2480
西洋棋從 1950 年起就被研究著。
04:02
So all this amazing recent progress in AI
really begs the question:
really begs the question:
72
230000
4240
所以，近期這些驚人的
人工智慧進步，讓大家想問：
人工智慧進步，讓大家想問：
04:07
How far will it go?
73
235280
1560
它能做到什麼程度？
04:09
I like to think about this question
74
237800
1696
我在思考這個問題時，
04:11
in terms of this abstract
landscape of tasks,
landscape of tasks,
75
239520
2976
想從工作任務的抽象地景來切入，
04:14
where the elevation represents
how hard it is for AI to do each task
how hard it is for AI to do each task
76
242520
3456
高度表示人工智慧要把每一項工作
做到人類水平的難度，
04:18
at human level,
77
246000
1216
04:19
and the sea level represents
what AI can do today.
what AI can do today.
78
247240
2760
海平面高度表示現今的
人工智慧能做什麼。
人工智慧能做什麼。
04:23
The sea level is rising
as AI improves,
as AI improves,
79
251120
2056
海平面隨著人工智慧的改進而上升，
04:25
so there's a kind of global warming
going on here in the task landscape.
going on here in the task landscape.
80
253200
3440
所以，在這工作任務地景上，
有類似全球暖化的現象發生。
有類似全球暖化的現象發生。
04:30
And the obvious takeaway
is to avoid careers at the waterfront --
is to avoid careers at the waterfront --
81
258040
3335
很顯然的結論是：
避免從事在海濱的職業——
避免從事在海濱的職業——
04:33
(Laughter)
82
261399
1257
（笑聲）
04:34
which will soon be
automated and disrupted.
automated and disrupted.
83
262680
2856
這些工作很快就會被自動化所取代。
04:37
But there's a much
bigger question as well.
bigger question as well.
84
265560
2976
但這裡還有一個更大的問題。
04:40
How high will the water end up rising?
85
268560
1810
水面最後會升到多高？
04:43
Will it eventually rise
to flood everything,
to flood everything,
86
271440
3200
它最後是否會升高到淹沒一切，
04:47
matching human intelligence at all tasks.
87
275840
2496
在所有工作任務上
都能和人類的智慧匹敵？
都能和人類的智慧匹敵？
04:50
This is the definition
of artificial general intelligence --
of artificial general intelligence --
88
278360
3736
這是「強人工智慧」的定義，
04:54
AGI,
89
282120
1296
縮寫 AGI，
04:55
which has been the holy grail
of AI research since its inception.
of AI research since its inception.
90
283440
3080
打從它的一開始，
它就是人工智慧研究的聖杯。
它就是人工智慧研究的聖杯。
04:59
By this definition, people who say,
91
287000
1776
依這個定義，當有人說：
05:00
"Ah, there will always be jobs
that humans can do better than machines,"
that humans can do better than machines,"
92
288800
3416
「啊，總是會有些工作，
人類能做得比機器好。」
人類能做得比機器好。」
05:04
are simply saying
that we'll never get AGI.
that we'll never get AGI.
93
292240
2920
他們說的只是
「我們永遠不會有 AGI。」
「我們永遠不會有 AGI。」
05:07
Sure, we might still choose
to have some human jobs
to have some human jobs
94
295680
3576
當然，我們仍可選擇
保留一些人類的工作，
保留一些人類的工作，
05:11
or to give humans income
and purpose with our jobs,
and purpose with our jobs,
95
299280
3096
或是用工作讓人類保有收入和目的，
05:14
but AGI will in any case
transform life as we know it
transform life as we know it
96
302400
3736
但不論如何，AGI 都會
轉變我們所認知的生活，
轉變我們所認知的生活，
05:18
with humans no longer being
the most intelligent.
the most intelligent.
97
306160
2736
人類將不再是最有智慧的。
05:20
Now, if the water level does reach AGI,
98
308920
3696
如果海平面真的升到 AGI 的高度，
05:24
then further AI progress will be driven
mainly not by humans but by AI,
mainly not by humans but by AI,
99
312640
5296
那麼進一步的人工智慧進步將會
由人工智慧來主導，而非人類，
由人工智慧來主導，而非人類，
05:29
which means that there's a possibility
100
317960
1856
意思是，有可能進一步的
人工智慧進展會非常快，
人工智慧進展會非常快，
05:31
that further AI progress
could be way faster
could be way faster
101
319840
2336
05:34
than the typical human research
and development timescale of years,
and development timescale of years,
102
322200
3376
超越用「年」來計算
典型人類研究和發展的時間，
典型人類研究和發展的時間，
05:37
raising the controversial possibility
of an intelligence explosion
of an intelligence explosion
103
325600
4016
提高很受爭議的智慧噴發的可能性，
05:41
where recursively self-improving AI
104
329640
2296
即，不斷遞迴的
自我改進的人工智慧
自我改進的人工智慧
很快就會遠遠超越人類的智慧，
05:43
rapidly leaves human
intelligence far behind,
intelligence far behind,
105
331960
3416
創造出所謂的超級人工智慧。
05:47
creating what's known
as superintelligence.
as superintelligence.
106
335400
2440
05:51
Alright, reality check:
107
339800
2280
好了，回歸現實：
05:55
Are we going to get AGI any time soon?
108
343120
2440
我們很快就會有 AGI 嗎？
05:58
Some famous AI researchers,
like Rodney Brooks,
like Rodney Brooks,
109
346360
2696
有些知名的人工智慧研究者，
像羅德尼·布魯克斯，
像羅德尼·布魯克斯，
認為在數百年內還不會發生。
06:01
think it won't happen
for hundreds of years.
for hundreds of years.
110
349080
2496
06:03
But others, like Google DeepMind
founder Demis Hassabis,
founder Demis Hassabis,
111
351600
3896
但其他人，像 Google DeepMind 的
創辦人傑米斯·哈薩比斯，
創辦人傑米斯·哈薩比斯，
就比較樂觀，
06:07
are more optimistic
112
355520
1256
且努力想要讓它早點發生。
06:08
and are working to try to make
it happen much sooner.
it happen much sooner.
113
356800
2576
06:11
And recent surveys have shown
that most AI researchers
that most AI researchers
114
359400
3296
近期的調查顯示，
大部分的人工智慧研究者
大部分的人工智慧研究者
其實和傑米斯一樣樂觀，
06:14
actually share Demis's optimism,
115
362720
2856
預期我們會在數十年內就有 AGI，
06:17
expecting that we will
get AGI within decades,
get AGI within decades,
116
365600
3080
06:21
so within the lifetime of many of us,
117
369640
2256
所以，許多人
在有生之年就能看到，
在有生之年就能看到，
06:23
which begs the question -- and then what?
118
371920
1960
這就讓人不禁想問：接下來呢？
06:27
What do we want the role of humans to be
119
375040
2216
我們希望人類扮演什麼角色，
06:29
if machines can do everything better
and cheaper than us?
and cheaper than us?
120
377280
2680
如果機器每件事都做得
比人類好、成本又更低的話？
比人類好、成本又更低的話？
06:35
The way I see it, we face a choice.
121
383000
2000
依我所見，我們面臨一個選擇。
選項之一是不假思索的滿足。
06:38
One option is to be complacent.
122
386000
1576
06:39
We can say, "Oh, let's just build machines
that can do everything we can do
that can do everything we can do
123
387600
3776
我們可以說：「咱們來打造機器，
讓它們做所有我們能做的事，
讓它們做所有我們能做的事，
06:43
and not worry about the consequences.
124
391400
1816
不要擔心結果。
06:45
Come on, if we build technology
that makes all humans obsolete,
that makes all humans obsolete,
125
393240
3256
拜託，如果我們能打造科技，
讓全人類變成過時，
讓全人類變成過時，
06:48
what could possibly go wrong?"
126
396520
2096
怎可能會出錯的？」
06:50
(Laughter)
127
398640
1656
（笑聲）
06:52
But I think that would be
embarrassingly lame.
embarrassingly lame.
128
400320
2760
但我覺得那樣是差勁到令人難堪。
06:56
I think we should be more ambitious --
in the spirit of TED.
in the spirit of TED.
129
404080
3496
我認為我們該更有野心——
抱持 TED 精神。
抱持 TED 精神。
06:59
Let's envision a truly inspiring
high-tech future
high-tech future
130
407600
3496
咱們來想像一下
一個真正鼓舞人心的高科技未來，
一個真正鼓舞人心的高科技未來，
07:03
and try to steer towards it.
131
411120
1400
並試著朝它邁進。
07:05
This brings us to the second part
of our rocket metaphor: the steering.
of our rocket metaphor: the steering.
132
413720
3536
這就帶我們來到了
火箭比喻的第二部分：操控。
火箭比喻的第二部分：操控。
07:09
We're making AI more powerful,
133
417280
1896
我們讓人工智慧的力量更強大，
07:11
but how can we steer towards a future
134
419200
3816
但我們要如何將人工智慧導向
協助人類繁盛而非掙扎無助的未來？
07:15
where AI helps humanity flourish
rather than flounder?
rather than flounder?
135
423040
3080
07:18
To help with this,
136
426760
1256
為了協助做到這點，
我共同創辦了「生命未來研究所」。
07:20
I cofounded the Future of Life Institute.
137
428040
1976
07:22
It's a small nonprofit promoting
beneficial technology use,
beneficial technology use,
138
430040
2776
它是個小型的非營利機構，
旨在促進有益的科技使用，
旨在促進有益的科技使用，
我們的目標很簡單：
希望生命的未來能夠存在，
希望生命的未來能夠存在，
07:24
and our goal is simply
for the future of life to exist
for the future of life to exist
139
432840
2736
且越是鼓舞人心越好。
07:27
and to be as inspiring as possible.
140
435600
2056
07:29
You know, I love technology.
141
437680
3176
你們知道的，我很愛科技。
07:32
Technology is why today
is better than the Stone Age.
is better than the Stone Age.
142
440880
2920
現今之所以比石器時代更好，
就是因為科技。
就是因為科技。
07:36
And I'm optimistic that we can create
a really inspiring high-tech future ...
a really inspiring high-tech future ...
143
444600
4080
我很樂觀地認為我們能創造出
真的很鼓舞人心的高科技未來……
真的很鼓舞人心的高科技未來……
07:41
if -- and this is a big if --
144
449680
1456
如果——這個「如果」很重要——
07:43
if we win the wisdom race --
145
451160
2456
如果我們能贏得這場智慧賽跑——
07:45
the race between the growing
power of our technology
power of our technology
146
453640
2856
這場賽跑的兩位競爭者是
不斷成長的科技力量
和不斷成長的管理科技智慧。
和不斷成長的管理科技智慧。
07:48
and the growing wisdom
with which we manage it.
with which we manage it.
147
456520
2200
07:51
But this is going to require
a change of strategy
a change of strategy
148
459240
2296
這會需要策略的改變，
07:53
because our old strategy
has been learning from mistakes.
has been learning from mistakes.
149
461560
3040
因為我們的舊策略
是從錯誤中學習。
是從錯誤中學習。
07:57
We invented fire,
150
465280
1536
我們發明了火，
07:58
screwed up a bunch of times --
151
466840
1536
搞砸了很多次——
08:00
invented the fire extinguisher.
152
468400
1816
發明了滅火器。
08:02
(Laughter)
153
470240
1336
（笑聲）
08:03
We invented the car,
screwed up a bunch of times --
screwed up a bunch of times --
154
471600
2416
我們發明了汽車，搞砸了很多次——
08:06
invented the traffic light,
the seat belt and the airbag,
the seat belt and the airbag,
155
474040
2667
發明了紅綠燈、
安全帶，和安全氣囊，
安全帶，和安全氣囊，
08:08
but with more powerful technology
like nuclear weapons and AGI,
like nuclear weapons and AGI,
156
476731
3845
但對於更強大的科技，
比如核子武器和 AGI，
比如核子武器和 AGI，
08:12
learning from mistakes
is a lousy strategy,
is a lousy strategy,
157
480600
3376
從錯誤中學習是很糟的策略，
08:16
don't you think?
158
484000
1216
對吧？（笑聲）
08:17
(Laughter)
159
485240
1016
事前的主動比事後的反應更好；
08:18
It's much better to be proactive
rather than reactive;
rather than reactive;
160
486280
2576
08:20
plan ahead and get things
right the first time
right the first time
161
488880
2296
先計畫好，第一次就把事情做對，
08:23
because that might be
the only time we'll get.
the only time we'll get.
162
491200
2496
因為我們可能只有一次機會。
08:25
But it is funny because
sometimes people tell me,
sometimes people tell me,
163
493720
2336
好笑的是，有時人們告訴我：
08:28
"Max, shhh, don't talk like that.
164
496080
2736
「麥克斯，噓，別那樣說話。
08:30
That's Luddite scaremongering."
165
498840
1720
那是危言聳聽。」
08:34
But it's not scaremongering.
166
502040
1536
但那並非危言聳聽。
08:35
It's what we at MIT
call safety engineering.
call safety engineering.
167
503600
2880
我們在麻省理工學院
稱之為安全工程。
稱之為安全工程。
08:39
Think about it:
168
507200
1216
想想看：
08:40
before NASA launched
the Apollo 11 mission,
the Apollo 11 mission,
169
508440
2216
在美國太空總署的
阿波羅 11 任務之前，
阿波羅 11 任務之前，
08:42
they systematically thought through
everything that could go wrong
everything that could go wrong
170
510680
3136
他們系統性地設想過
所有可能出錯的狀況，
所有可能出錯的狀況，
08:45
when you put people
on top of explosive fuel tanks
on top of explosive fuel tanks
171
513840
2376
畢竟是要把人類
放在易爆燃料槽上，
放在易爆燃料槽上，
08:48
and launch them somewhere
where no one could help them.
where no one could help them.
172
516240
2616
再將他們發射到
沒有人能協助他們的地方。
沒有人能協助他們的地方。
08:50
And there was a lot that could go wrong.
173
518880
1936
可能會出錯的狀況非常多。
08:52
Was that scaremongering?
174
520840
1480
那是危言聳聽嗎？
08:55
No.
175
523159
1217
不。
08:56
That's was precisely
the safety engineering
the safety engineering
176
524400
2016
那正是安全工程，
用來確保任務能夠成功。
08:58
that ensured the success of the mission,
177
526440
1936
那正是我認為處理 AGI 時
應該採用的策略。
應該採用的策略。
09:00
and that is precisely the strategy
I think we should take with AGI.
I think we should take with AGI.
178
528400
4176
09:04
Think through what can go wrong
to make sure it goes right.
to make sure it goes right.
179
532600
4056
想清楚有什麼可能會出錯，
確保它能不要出錯。
確保它能不要出錯。
09:08
So in this spirit,
we've organized conferences,
we've organized conferences,
180
536680
2536
基於這種精神，我們辦了一些會議，
集合了最領先的人工智慧
研究者和其他思想家，
研究者和其他思想家，
09:11
bringing together leading
AI researchers and other thinkers
AI researchers and other thinkers
181
539240
2816
09:14
to discuss how to grow this wisdom
we need to keep AI beneficial.
we need to keep AI beneficial.
182
542080
3736
討論要如何發展這項必要的智慧，
確保人工智慧是有益的。
確保人工智慧是有益的。
我們最近一次會議是去年
在加州的阿西洛馬會議中舉辦，
在加州的阿西洛馬會議中舉辦，
09:17
Our last conference
was in Asilomar, California last year
was in Asilomar, California last year
183
545840
3296
09:21
and produced this list of 23 principles
184
549160
3056
得出了這 23 條原則。
09:24
which have since been signed
by over 1,000 AI researchers
by over 1,000 AI researchers
185
552240
2896
從那之後，已經有
超過一千名人工智慧研究者
超過一千名人工智慧研究者
與重要產業領導者簽署。
09:27
and key industry leaders,
186
555160
1296
09:28
and I want to tell you
about three of these principles.
about three of these principles.
187
556480
3176
我想要和各位談其中三條原則。
09:31
One is that we should avoid an arms race
and lethal autonomous weapons.
and lethal autonomous weapons.
188
559680
4960
其一是我們應該要避免
軍備競賽以及自主的致命武器。
軍備競賽以及自主的致命武器。
09:37
The idea here is that any science
can be used for new ways of helping people
can be used for new ways of helping people
189
565480
3616
想法是，任何科學都能被
用作助人或傷人的新方法。
用作助人或傷人的新方法。
09:41
or new ways of harming people.
190
569120
1536
09:42
For example, biology and chemistry
are much more likely to be used
are much more likely to be used
191
570680
3936
比如，生物和化學更有可能會被用來
09:46
for new medicines or new cures
than for new ways of killing people,
than for new ways of killing people,
192
574640
4856
做新的藥物或治療方法，
而不是殺人的新方法，
而不是殺人的新方法，
09:51
because biologists
and chemists pushed hard --
and chemists pushed hard --
193
579520
2176
因為生物學家和化學家
很努力推動——
很努力推動——
09:53
and successfully --
194
581720
1256
且很成功——
09:55
for bans on biological
and chemical weapons.
and chemical weapons.
195
583000
2176
禁止生物及化學武器的禁令。
09:57
And in the same spirit,
196
585200
1256
基於同樣的精神，
09:58
most AI researchers want to stigmatize
and ban lethal autonomous weapons.
and ban lethal autonomous weapons.
197
586480
4440
大部分的人工智慧研究者
想要譴責和禁用自主的致命武器。
想要譴責和禁用自主的致命武器。
10:03
Another Asilomar AI principle
198
591600
1816
另一條阿西洛馬會議人工智慧原則
10:05
is that we should mitigate
AI-fueled income inequality.
AI-fueled income inequality.
199
593440
3696
是我們應該要減輕
由人工智慧引起的收入不平等。
由人工智慧引起的收入不平等。
10:09
I think that if we can grow
the economic pie dramatically with AI
the economic pie dramatically with AI
200
597160
4456
我認為如果能用人工智慧
讓經濟大餅大幅地成長，
讓經濟大餅大幅地成長，
10:13
and we still can't figure out
how to divide this pie
how to divide this pie
201
601640
2456
而仍無法弄清楚如何分割這塊餅
10:16
so that everyone is better off,
202
604120
1576
來讓每個人都過得更好，
10:17
then shame on us.
203
605720
1256
那我們真該感到羞恥。
10:19
(Applause)
204
607000
4096
（掌聲）
10:23
Alright, now raise your hand
if your computer has ever crashed.
if your computer has ever crashed.
205
611120
3600
好，如果你的電腦曾當過機，請舉手。
10:27
(Laughter)
206
615480
1256
（笑聲）
10:28
Wow, that's a lot of hands.
207
616760
1656
哇，好多人舉手。
10:30
Well, then you'll appreciate
this principle
this principle
208
618440
2176
那麼你們就會欣賞這條原則：
10:32
that we should invest much more
in AI safety research,
in AI safety research,
209
620640
3136
我們應該要更投入
人工智慧安全的研究，
人工智慧安全的研究，
10:35
because as we put AI in charge
of even more decisions and infrastructure,
of even more decisions and infrastructure,
210
623800
3656
因為當我們讓人工智慧來主導
更多決策和基礎設施時。
更多決策和基礎設施時。
10:39
we need to figure out how to transform
today's buggy and hackable computers
today's buggy and hackable computers
211
627480
3616
我們得要想出該如何將現今
有程式錯誤且可能被駭入的電腦，
有程式錯誤且可能被駭入的電腦，
10:43
into robust AI systems
that we can really trust,
that we can really trust,
212
631120
2416
轉變成我們能真正信任的
穩定的人工智慧系統。
穩定的人工智慧系統。
10:45
because otherwise,
213
633560
1216
要不然，所有這些了不起的新科技
都可能故障、傷害我們，
都可能故障、傷害我們，
10:46
all this awesome new technology
can malfunction and harm us,
can malfunction and harm us,
214
634800
2816
10:49
or get hacked and be turned against us.
215
637640
1976
或被駭入而轉成對抗我們。
10:51
And this AI safety work
has to include work on AI value alignment,
has to include work on AI value alignment,
216
639640
5696
這項人工智慧安全性的工作必須要
包含人工智慧價值校準的工作，
包含人工智慧價值校準的工作，
10:57
because the real threat
from AGI isn't malice,
from AGI isn't malice,
217
645360
2816
因為 AGI 帶來的
真正威脅不是惡意——
真正威脅不是惡意——
11:00
like in silly Hollywood movies,
218
648200
1656
像愚蠢的好萊塢電影裡那樣——
11:01
but competence --
219
649880
1736
而是能力——
11:03
AGI accomplishing goals
that just aren't aligned with ours.
that just aren't aligned with ours.
220
651640
3416
AGI 要完成的目標就是
與我們的目標不一致。
與我們的目標不一致。
11:07
For example, when we humans drove
the West African black rhino extinct,
the West African black rhino extinct,
221
655080
4736
比如，當我們人類
讓西非的黑犀牛濱臨絕種時，
讓西非的黑犀牛濱臨絕種時，
11:11
we didn't do it because we were a bunch
of evil rhinoceros haters, did we?
of evil rhinoceros haters, did we?
222
659840
3896
並不是因為我們邪惡、痛恨犀牛
才這麼做的，對吧？
才這麼做的，對吧？
11:15
We did it because
we were smarter than them
we were smarter than them
223
663760
2056
會這麼做，是因為我們比牠們聰明，
11:17
and our goals weren't aligned with theirs.
224
665840
2576
我們的目標和牠們的目標不一致。
11:20
But AGI is by definition smarter than us,
225
668440
2656
但 AGI 在定義上
就是比我們聰明的，
就是比我們聰明的，
11:23
so to make sure that we don't put
ourselves in the position of those rhinos
ourselves in the position of those rhinos
226
671120
3576
所以要確保我們在
創造了 AGI 之後，
創造了 AGI 之後，
11:26
if we create AGI,
227
674720
1976
不會淪落到那些犀牛的處境，
11:28
we need to figure out how
to make machines understand our goals,
to make machines understand our goals,
228
676720
4176
我們就得要想出如何
讓機器了解我們的目標，
讓機器了解我們的目標，
11:32
adopt our goals and retain our goals.
229
680920
3160
採用並且保持我們的目標。
11:37
And whose goals should these be, anyway?
230
685320
2856
不過，這些目標該是誰的目標？
11:40
Which goals should they be?
231
688200
1896
這些目標該是哪些目標？
11:42
This brings us to the third part
of our rocket metaphor: the destination.
of our rocket metaphor: the destination.
232
690120
3560
這就帶我們到了火箭比喻的
第三部分：目的地。
第三部分：目的地。
11:47
We're making AI more powerful,
233
695160
1856
我們要讓人工智慧的力量更強大，
11:49
trying to figure out how to steer it,
234
697040
1816
試圖想辦法來操控它，
11:50
but where do we want to go with it?
235
698880
1680
但我們如何帶它到哪裡？
11:53
This is the elephant in the room
that almost nobody talks about --
that almost nobody talks about --
236
701760
3656
這是幾乎沒人談論的房中大象
（顯而易見又被忽略）——
（顯而易見又被忽略）——
11:57
not even here at TED --
237
705440
1856
即使在 TED 也沒人在談——
11:59
because we're so fixated
on short-term AI challenges.
on short-term AI challenges.
238
707320
4080
因為我們都把目光放在
短期的人工智慧挑戰。
短期的人工智慧挑戰。
12:04
Look, our species is trying to build AGI,
239
712080
4656
聽著，我們人類正試著建造 AGI，
12:08
motivated by curiosity and economics,
240
716760
3496
動機是好奇心和經濟，
12:12
but what sort of future society
are we hoping for if we succeed?
are we hoping for if we succeed?
241
720280
3680
但如果成功了，我們希望
創造出什麼樣的未來社會？
創造出什麼樣的未來社會？
12:16
We did an opinion poll on this recently,
242
724680
1936
最近我們針對這點做了意見調查，
12:18
and I was struck to see
243
726640
1216
結果讓我很驚訝，
12:19
that most people actually
want us to build superintelligence:
want us to build superintelligence:
244
727880
2896
大部分的人其實希望
我們建造超級人工智慧：
我們建造超級人工智慧：
12:22
AI that's vastly smarter
than us in all ways.
than us in all ways.
245
730800
3160
全面比我們聰明的人工智慧。
12:27
What there was the greatest agreement on
was that we should be ambitious
was that we should be ambitious
246
735120
3416
大家最一致的意見，
就是我們應該要有野心，
就是我們應該要有野心，
12:30
and help life spread into the cosmos,
247
738560
2016
並協助生命在宇宙中散播，
12:32
but there was much less agreement
about who or what should be in charge.
about who or what should be in charge.
248
740600
4496
但對於該由誰或由什麼來主導，
大家的意見就不那麼一致了。
大家的意見就不那麼一致了。
12:37
And I was actually quite amused
249
745120
1736
有件事讓我覺得很有趣的
12:38
to see that there's some some people
who want it to be just machines.
who want it to be just machines.
250
746880
3456
是我看到有些人希望就由機器主導。
12:42
(Laughter)
251
750360
1696
（笑聲）
12:44
And there was total disagreement
about what the role of humans should be,
about what the role of humans should be,
252
752080
3856
至於人類該扮演什麼角色，
意見就完全不一致了，
意見就完全不一致了，
12:47
even at the most basic level,
253
755960
1976
即使在最基礎的層級也一樣。
12:49
so let's take a closer look
at possible futures
at possible futures
254
757960
2816
讓咱們來近看
我們或許會選擇的可能未來，好嗎？
12:52
that we might choose
to steer toward, alright?
to steer toward, alright?
255
760800
2736
12:55
So don't get be wrong here.
256
763560
1336
別誤會我的意思，
我並不是在談太空旅行，
12:56
I'm not talking about space travel,
257
764920
2056
只是要談人類進入未來的比喻之旅。
12:59
merely about humanity's
metaphorical journey into the future.
metaphorical journey into the future.
258
767000
3200
13:02
So one option that some
of my AI colleagues like
of my AI colleagues like
259
770920
3496
我的一些人工智慧同事
很喜歡的一個選擇是
很喜歡的一個選擇是
建造超級人工智慧，
並保持讓它被人類控制，
並保持讓它被人類控制，
13:06
is to build superintelligence
and keep it under human control,
and keep it under human control,
260
774440
3616
就像能被奴役的神一樣，
13:10
like an enslaved god,
261
778080
1736
和網路沒有連結，
13:11
disconnected from the internet
262
779840
1576
用來創造無法想像的科技和財富，
13:13
and used to create unimaginable
technology and wealth
technology and wealth
263
781440
3256
全交給控制它的人。
13:16
for whoever controls it.
264
784720
1240
13:18
But Lord Acton warned us
265
786800
1456
阿克頓男爵警告我們：
「權力會產生腐敗，
絕對權力絕對腐敗。」
絕對權力絕對腐敗。」
13:20
that power corrupts,
and absolute power corrupts absolutely,
and absolute power corrupts absolutely,
266
788280
3616
13:23
so you might worry that maybe
we humans just aren't smart enough,
we humans just aren't smart enough,
267
791920
4056
你可能會擔心
也許我們人類就是不夠聰明，
也許我們人類就是不夠聰明，
13:28
or wise enough rather,
268
796000
1536
或是沒有足夠的智慧
13:29
to handle this much power.
269
797560
1240
來操作這麼多權力。
13:31
Also, aside from any
moral qualms you might have
moral qualms you might have
270
799640
2536
除了奴役更優秀的智慧的
任何道德疑慮之外，
任何道德疑慮之外，
13:34
about enslaving superior minds,
271
802200
2296
13:36
you might worry that maybe
the superintelligence could outsmart us,
the superintelligence could outsmart us,
272
804520
3976
你也許會擔心
或許超級人工智慧會智勝我們，
13:40
break out and take over.
273
808520
2240
衝破籓籬和掌管。
13:43
But I also have colleagues
who are fine with AI taking over
who are fine with AI taking over
274
811560
3416
但我也有些同事覺得
讓人工智慧接管也沒不好，
讓人工智慧接管也沒不好，
13:47
and even causing human extinction,
275
815000
2296
甚至造成人類絕種也無妨，
13:49
as long as we feel the the AIs
are our worthy descendants,
are our worthy descendants,
276
817320
3576
只要我們覺得人工智慧
配得上做我們的後裔就好，
配得上做我們的後裔就好，
13:52
like our children.
277
820920
1736
就像我們的孩子一樣。
13:54
But how would we know that the AIs
have adopted our best values
have adopted our best values
278
822680
5616
但，我們要如何確知人工智慧
已經採用了我們最好的價值觀，
已經採用了我們最好的價值觀，
14:00
and aren't just unconscious zombies
tricking us into anthropomorphizing them?
tricking us into anthropomorphizing them?
279
828320
4376
而不只是有意識的殭屍，
騙我們將人性賦予它們？
騙我們將人性賦予它們？
14:04
Also, shouldn't those people
who don't want human extinction
who don't want human extinction
280
832720
2856
此外，不希望人類絕種的那些人
14:07
have a say in the matter, too?
281
835600
1440
也應被容許對此事表達意見吧？
14:10
Now, if you didn't like either
of those two high-tech options,
of those two high-tech options,
282
838200
3376
如果這兩個高科技選項
都不合你們的意，
都不合你們的意，
14:13
it's important to remember
that low-tech is suicide
that low-tech is suicide
283
841600
3176
很重要的是要記得
從宇宙的觀點來看低科技是自殺，
從宇宙的觀點來看低科技是自殺，
14:16
from a cosmic perspective,
284
844800
1256
14:18
because if we don't go far
beyond today's technology,
beyond today's technology,
285
846080
2496
因為如果我們不遠遠超越
現今的科技，
現今的科技，
14:20
the question isn't whether humanity
is going to go extinct,
is going to go extinct,
286
848600
2816
問題就不是人類是否會絕種，
而是讓我們絕種的會是下一次的
14:23
merely whether
we're going to get taken out
we're going to get taken out
287
851440
2016
14:25
by the next killer asteroid, supervolcano
288
853480
2136
巨型慧星撞擊、超級火山爆發，
14:27
or some other problem
that better technology could have solved.
that better technology could have solved.
289
855640
3096
或是更優的科技
本可解決的其他問題？
本可解決的其他問題？
14:30
So, how about having
our cake and eating it ...
our cake and eating it ...
290
858760
3576
所以，何不接受和吃下這蛋糕……
14:34
with AGI that's not enslaved
291
862360
1840
這個不是被奴役的 AGI，
14:37
but treats us well because its values
are aligned with ours?
are aligned with ours?
292
865120
3176
而是價值觀和我們一致，
善待我們的 AGI 呢？
善待我們的 AGI 呢？
14:40
This is the gist of what Eliezer Yudkowsky
has called "friendly AI,"
has called "friendly AI,"
293
868320
4176
那就是亞里艾瑟·尤考斯基
所謂的「友善的人工智慧」，
所謂的「友善的人工智慧」，
14:44
and if we can do this,
it could be awesome.
it could be awesome.
294
872520
2680
若我們能做到這點，可能會很棒。
14:47
It could not only eliminate negative
experiences like disease, poverty,
experiences like disease, poverty,
295
875840
4816
它可能不只會除去負面的遭遇，
比如疾病、貧困、
14:52
crime and other suffering,
296
880680
1456
犯罪，和其他苦難，
14:54
but it could also give us
the freedom to choose
the freedom to choose
297
882160
2816
它也可能會給予我們自由，
讓我們從各式各樣
新的正面經驗中做選擇——
新的正面經驗中做選擇——
14:57
from a fantastic new diversity
of positive experiences --
of positive experiences --
298
885000
4056
15:01
basically making us
the masters of our own destiny.
the masters of our own destiny.
299
889080
3160
基本上，就是讓我們
成為自己命運的主宰。
成為自己命運的主宰。
15:06
So in summary,
300
894280
1376
所以，總結一下，
15:07
our situation with technology
is complicated,
is complicated,
301
895680
3096
我們在科技方面的情況很複雜，
15:10
but the big picture is rather simple.
302
898800
2416
但整體來看是很簡單的。
15:13
Most AI researchers
expect AGI within decades,
expect AGI within decades,
303
901240
3456
多數人工智慧研究者預期
AGI 會在數十年內出現，
AGI 會在數十年內出現，
15:16
and if we just bumble
into this unprepared,
into this unprepared,
304
904720
3136
如果我們沒有先準備好面對它，
15:19
it will probably be
the biggest mistake in human history --
the biggest mistake in human history --
305
907880
3336
那可能會成為人類史上最大的錯誤。
15:23
let's face it.
306
911240
1416
讓我們正視事實吧，
15:24
It could enable brutal,
global dictatorship
global dictatorship
307
912680
2576
它會讓殘酷的
全球獨裁主義成為可能，
全球獨裁主義成為可能，
15:27
with unprecedented inequality,
surveillance and suffering,
surveillance and suffering,
308
915280
3536
造成前所未有的不平等、
監控，以及苦難，
監控，以及苦難，
15:30
and maybe even human extinction.
309
918840
1976
甚至讓人類絕種。
15:32
But if we steer carefully,
310
920840
2320
但如果我們小心地操控，
15:36
we could end up in a fantastic future
where everybody's better off:
where everybody's better off:
311
924040
3896
我們可能會有個美好的未來，
人人都過得更好：
人人都過得更好：
15:39
the poor are richer, the rich are richer,
312
927960
2376
貧窮的人有錢，有錢的人更有錢，
15:42
everybody is healthy
and free to live out their dreams.
and free to live out their dreams.
313
930360
3960
每個人都健康，能自由自在地
去實現他們的夢想。
去實現他們的夢想。
15:47
Now, hang on.
314
935000
1536
等等，別急。
15:48
Do you folks want the future
that's politically right or left?
that's politically right or left?
315
936560
4576
你們希望未來在政治上
是右派還是左派？
是右派還是左派？
15:53
Do you want the pious society
with strict moral rules,
with strict moral rules,
316
941160
2856
你們想要一個有著嚴格
道德規則的虔誠社會，
道德規則的虔誠社會，
15:56
or do you an hedonistic free-for-all,
317
944040
1816
或一個人人可參與的享樂主義社會，
15:57
more like Burning Man 24/7?
318
945880
2216
就像全年無休的燃燒人節慶？
16:00
Do you want beautiful beaches,
forests and lakes,
forests and lakes,
319
948120
2416
你們想要美麗的海灘、森林和湖泊，
16:02
or would you prefer to rearrange
some of those atoms with the computers,
some of those atoms with the computers,
320
950560
3416
或是偏好用電腦重新排列原子，
16:06
enabling virtual experiences?
321
954000
1715
產生出虛擬經驗？
16:07
With friendly AI, we could simply
build all of these societies
build all of these societies
322
955739
3157
有了友善的人工智慧，
我們就能建立出所有這些社會，
我們就能建立出所有這些社會，
16:10
and give people the freedom
to choose which one they want to live in
to choose which one they want to live in
323
958920
3216
並讓大家有自由去選擇
他們想要住在哪個社會中，
他們想要住在哪個社會中，
因為我們不會再受智慧的限制，
16:14
because we would no longer
be limited by our intelligence,
be limited by our intelligence,
324
962160
3096
唯一的限制是物理法則。
16:17
merely by the laws of physics.
325
965280
1456
16:18
So the resources and space
for this would be astronomical --
for this would be astronomical --
326
966760
4616
所以，資源和空間會非常龐大 ——
16:23
literally.
327
971400
1320
天文級的龐大。
16:25
So here's our choice.
328
973320
1200
我們的選擇如下：
16:27
We can either be complacent
about our future,
about our future,
329
975880
2320
我們可以對未來感到滿足，
16:31
taking as an article of blind faith
330
979440
2656
帶著盲目的信念，
16:34
that any new technology
is guaranteed to be beneficial,
is guaranteed to be beneficial,
331
982120
4016
相信任何新科技都必然有益，
16:38
and just repeat that to ourselves
as a mantra over and over and over again
as a mantra over and over and over again
332
986160
4136
當作真言，不斷對自己
一次又一次地重述，
一次又一次地重述，
16:42
as we drift like a rudderless ship
towards our own obsolescence.
towards our own obsolescence.
333
990320
3680
而我們像無舵的船漂向淘汰。
16:46
Or we can be ambitious --
334
994920
1880
或是我們可以有野心，
16:49
thinking hard about how
to steer our technology
to steer our technology
335
997840
2456
努力去想出如何操控我們的科技，
16:52
and where we want to go with it
336
1000320
1936
以及我們想要去的目的地，
16:54
to create the age of amazement.
337
1002280
1760
創造出驚奇的時代。
16:57
We're all here to celebrate
the age of amazement,
the age of amazement,
338
1005000
2856
我們在這裡讚頌驚奇的時代，
16:59
and I feel that its essence should lie
in becoming not overpowered
in becoming not overpowered
339
1007880
4440
我覺得精髓應該在於不受控於科技，
17:05
but empowered by our technology.
340
1013240
2616
而是讓它賦予我們力量。
17:07
Thank you.
341
1015880
1376
謝謝。
17:09
(Applause)
342
1017280
3080
（掌聲）
ABOUT THE SPEAKER
Max Tegmark - Scientist, authorMax Tegmark is driven by curiosity, both about how our universe works and about how we can use the science and technology we discover to help humanity flourish rather than flounder.
Why you should listen
Max Tegmark is an MIT professor who loves thinking about life's big questions. He's written two popular books, Our Mathematical Universe: My Quest for the Ultimate Nature of Reality and the recently published Life 3.0: Being Human in the Age of Artificial Intelligence, as well as more than 200 nerdy technical papers on topics from cosmology to AI.
He writes: "In my spare time, I'm president of the Future of Life Institute, which aims to ensure that we develop not only technology but also the wisdom required to use it beneficially."
Max Tegmark | Speaker | TED.com