TED2019
Arnav Kapur: How AI could become an extension of your mind
阿纳·卡普尔: 人工智能如何成为思维的延伸？
Filmed:
Readability: 4.9
601,116 views
尝试不开张嘴巴自言自语，只是在内心说出词语。如果你可以在网络上像这样进行搜索——然后得到结果呢？TED Fellow阿纳·卡普尔第一次公开在现场介绍演示他的新技术，AlterEgo：一个可穿戴的人工智能装置，可以让你不发出声音说话并且从计算机中获得信息，就像你大脑里的声音。了解更多关于这个装置如何工作以及这个新型人机互动产生的广泛影响的意义。
Arnav Kapur - Technologist
TED Fellow Arnav Kapur invents wearable AI devices that augment human cognition and give voice to those who have lost their ability to speak. Full bio
TED Fellow Arnav Kapur invents wearable AI devices that augment human cognition and give voice to those who have lost their ability to speak. Full bio
Double-click the English transcript below to play the video.
00:13
Ever since computers were invented,
0
1917
1767
自从电脑被发明出来，
我们一直致力于使它们更加聪明和强大。
00:15
we've been trying to make them
smarter and more powerful.
smarter and more powerful.
1
3708
3601
从算盘，到房间大小的机器，
00:19
From the abacus, to room-sized machines,
2
7333
2768
到台式机，再到可以
放进我们口袋里的电脑。
放进我们口袋里的电脑。
00:22
to desktops, to computers in our pockets.
3
10125
2851
现在我们开始设计人工智能，
来完成要求人类智能
来完成要求人类智能
00:25
And are now designing
artificial intelligence to automate tasks
artificial intelligence to automate tasks
4
13000
3434
来完成的自动化工作。
00:28
that would require human intelligence.
5
16458
1834
00:31
If you look at the history of computing,
6
19542
1934
如果你看一下计算机的历史，
我们总是将电脑看成一个外部的装置，
00:33
we've always treated computers
as external devices
as external devices
7
21500
3226
代表我们来进行计算和行动。
00:36
that compute and act on our behalf.
8
24750
2333
00:40
What I want to do is I want to weave
computing, AI and internet as part of us.
computing, AI and internet as part of us.
9
28167
5892
我想做的，则是把计算、人工智能
和网络整合在一起成为我们的一部分。
和网络整合在一起成为我们的一部分。
作为人类认知的一部分，
00:46
As part of human cognition,
10
34083
2560
它们使我们可以自由地与
周围的世界互动。
周围的世界互动。
00:48
freeing us to interact
with the world around us.
with the world around us.
11
36667
2250
00:51
Integrate human and machine intelligence
12
39833
2643
人类能力强化，可以通过
使人类和机械智能
使人类和机械智能
在我们的身体里成为一个整体来实现，
而不是让它弱化我们，或者替代我们。
而不是让它弱化我们，或者替代我们。
00:54
right inside our own bodies to augment us,
instead of diminishing us or replacing us.
instead of diminishing us or replacing us.
13
42500
4833
01:00
Could we combine what people do best,
such as creative and intuitive thinking,
such as creative and intuitive thinking,
14
48583
4185
我们可不可以把人类最擅长的，
比如创造力和直觉思维
比如创造力和直觉思维
和电脑最擅长的，
01:04
with what computers do best,
15
52792
1392
比如信息处理和
完美的记忆力结合起来呢？
完美的记忆力结合起来呢？
01:06
such as processing information
and perfectly memorizing stuff?
and perfectly memorizing stuff?
16
54208
3768
这会不会产生1加1大于2的效果呢？
01:10
Could this whole be better
than the sum of its parts?
than the sum of its parts?
17
58000
2833
01:14
We have a device
that could make that possible.
that could make that possible.
18
62167
2809
我们有一个装置
可以使这个设想成为可能。
可以使这个设想成为可能。
它叫做AlterEgo，
是一个可穿戴的装置，
是一个可穿戴的装置，
01:17
It's called AlterEgo,
and it's a wearable device
and it's a wearable device
19
65000
2643
01:19
that gives you the experience
of a conversational AI
of a conversational AI
20
67667
3017
你会感觉到它是一个
生活在你大脑里的，
生活在你大脑里的，
可以进行对话的人工智能，
01:22
that lives inside your head,
21
70708
1601
这种感觉就像是你在和
你心里的自己对话。
你心里的自己对话。
01:24
that you could talk to in likeness
to talking to yourself internally.
to talking to yourself internally.
22
72333
3584
01:29
We have a new prototype
that we're showing here,
that we're showing here,
23
77279
2239
我们这里展示的是我们的一个新的样本，
这是我们第一次在TED进行展示，
它是这么工作的。
它是这么工作的。
01:31
for the first time at TED,
and here's how it works.
and here's how it works.
24
79542
3000
01:36
Normally, when we speak,
25
84292
1267
通常来说，当我们说话的时候，
大脑会通过神经
01:37
the brain sends neurosignals
through the nerves
through the nerves
26
85583
2810
向你的内部语言系统传递神经信号
01:40
to your internal speech systems,
27
88417
2142
来激活它们，让你的声带可以开始讲话。
01:42
to activate them and your vocal cords
to produce speech.
to produce speech.
28
90583
3459
01:47
One of the most complex
cognitive and motor tasks
cognitive and motor tasks
29
95107
2286
这是我们作为人类，进行的
最复杂的与认知和肌肉运动
最复杂的与认知和肌肉运动
有关的工作之一。
01:49
that we do as human beings.
30
97417
1458
01:52
Now, imagine talking to yourself
31
100125
2059
现在，想象一下和你自己说话，
不要发出声音，
不要动嘴，
不要动嘴，
01:54
without vocalizing,
without moving your mouth,
without moving your mouth,
32
102208
2601
不要动下巴，
01:56
without moving your jaw,
33
104833
1601
只需要在心里清晰的将这些单词说出来，
01:58
but by simply articulating
those words internally.
those words internally.
34
106458
2667
02:02
Thereby very subtly engaging
your internal speech systems,
your internal speech systems,
35
110208
4768
也就是非常巧妙地运用
你的内部语言系统，
你的内部语言系统，
比如你的舌头和上颚的后半部分。
02:07
such as your tongue
and back of your palate.
and back of your palate.
36
115000
2625
02:11
When that happens,
37
119000
1768
在这个时候，
大脑会发送非常微弱的信号
给这些内部语言系统，
给这些内部语言系统，
02:12
the brain sends extremely weak signals
to these internal speech systems.
to these internal speech systems.
38
120792
4392
AlterEgo有传感器，
02:17
AlterEgo has sensors,
39
125208
1726
嵌入在一个薄薄的、
柔韧且透明的塑料装置里，
柔韧且透明的塑料装置里，
02:18
embedded in a thin plastic,
flexible and transparent device
flexible and transparent device
40
126958
4226
可以像贴纸一样贴在你的脖子上。
02:23
that sits on your neck
just like a sticker.
just like a sticker.
41
131208
2209
02:27
These sensors pick up
on these internal signals
on these internal signals
42
135542
2642
这些传感器会透过表面的皮肤，
在口腔的深处获得
02:30
sourced deep within the mouth cavity,
43
138208
1893
并感应到这些内部信号。
02:32
right from the surface of the skin.
44
140125
1708
02:35
An AI program running in the background
45
143910
1858
一个人工智能程序在后台运行，
试图分辨出用户想说的话。
02:37
then tries to figure out
what the user's trying to say.
what the user's trying to say.
46
145792
3309
之后它会通过骨骼传导
02:41
It then feeds back an answer to the user
47
149125
3393
将答案反馈给用户，
02:44
by means of bone conduction,
48
152542
1767
声音通过颅骨传到用户的内耳，
02:46
audio conducted through the skull
into the user's inner ear,
into the user's inner ear,
49
154333
3643
用户所听到的声音
02:50
that the user hears,
50
158000
1268
覆盖在自然听觉环境上面，
02:51
overlaid on top of the user's
natural hearing of the environment,
natural hearing of the environment,
51
159292
3101
并不会屏蔽它。
02:54
without blocking it.
52
162417
1250
02:58
The combination of all these parts,
the input, the output and the AI,
the input, the output and the AI,
53
166542
3809
这些部分的结合：
输入，输出和人工智能，
输入，输出和人工智能，
使你脑中的界面有了
一个非常主观的经验，
一个非常主观的经验，
03:02
gives a net subjective experience
of an interface inside your head
of an interface inside your head
54
170375
3809
就是你可以像跟自己说话一样说话。
03:06
that you could talk to
in likeness to talking to yourself.
in likeness to talking to yourself.
55
174208
3125
03:11
Just to be very clear, the device
does not record or read your thoughts.
does not record or read your thoughts.
56
179750
3976
需要阐明的一点是，这个装备
不会记录或者读取你的想法。
不会记录或者读取你的想法。
它只会记录你想交流的
03:15
It records deliberate information
that you want to communicate
that you want to communicate
57
183750
2934
并通过你内部语言系统的有意参与
而透露出来的信息。
而透露出来的信息。
03:18
through deliberate engagement
of your internal speech systems.
of your internal speech systems.
58
186708
3060
人们不想被读取想法，
他们更愿意写出来。
他们更愿意写出来。
03:21
People don't want to be read,
they want to write.
they want to write.
59
189792
2726
这就是为什么我们设计这个系统来
03:24
Which is why we designed the system
60
192542
1934
特意的从外周神经系统来记录信息。
03:26
to deliberately record
from the peripheral nervous system.
from the peripheral nervous system.
61
194500
3708
03:31
Which is why the control
in all situations resides with the user.
in all situations resides with the user.
62
199750
3500
这就是为什么所有情况下的
控制权都属于用户。
控制权都属于用户。
03:37
I want to stop here for a second
and show you a live demo.
and show you a live demo.
63
205042
2750
这里我想暂停一下，
给你们一个现场的演示。
给你们一个现场的演示。
03:40
What I'm going to do is,
I'm going to ask Eric a question.
I'm going to ask Eric a question.
64
208875
2809
我将向Eric问一个问题。
他将搜索这个问题的信息，
03:43
And he's going to search
for that information
for that information
65
211708
2101
不是通过语言，也不是通过打字，
也不会移动他的手指，
也不会移动他的手指，
03:45
without vocalizing, without typing,
without moving his fingers,
without moving his fingers,
66
213833
3893
或者是动一下嘴。
03:49
without moving his mouth.
67
217750
1268
只是在心里问这个问题。
03:51
Simply by internally asking that question.
68
219042
2892
人工智能将会找出这个答案
并过声音，通过这个装置
并过声音，通过这个装置
03:53
The AI will then figure out the answer
and feed it back to Eric,
and feed it back to Eric,
69
221958
3185
发送给Eric。
03:57
through audio, through the device.
70
225167
1625
03:59
While you see a laptop
in front of him, he's not using it.
in front of him, he's not using it.
71
227917
3101
你看到他面前有一个笔记本电脑，
他不会用这个电脑。
他不会用这个电脑。
所有的东西都在装置上面。
04:03
Everything lives on the device.
72
231042
2351
他需要的只是那个像贴纸一样的装置，
去连接人工智能和互联网。
去连接人工智能和互联网。
04:05
All he needs is that sticker device
to interface with the AI and the internet.
to interface with the AI and the internet.
73
233417
3791
04:11
So, Eric, what's the weather
in Vancouver like, right now?
in Vancouver like, right now?
74
239458
3959
那么Eric，现在温哥华的天气怎么样？
04:22
What you see on the screen
75
250917
1684
现在你在屏幕上面看到的
就是Eric现在在心里对他自己说的话。
04:24
are the words that Eric
is speaking to himself right now.
is speaking to himself right now.
76
252625
4517
这是实时发生的。
04:29
This is happening in real time.
77
257166
1500
04:31
Eric: It's 50 degrees
and rainy here in Vancouver.
and rainy here in Vancouver.
78
259875
3393
Eric：现在温哥华的气温为
10摄氏度，有雨。
10摄氏度，有雨。
所以，人工智能将答案
04:35
Arnav Kapur: What happened is
that the AI sent the answer
that the AI sent the answer
79
263292
2726
通过语音和装置发回给Eric。
04:38
through audio, through
the device, back to Eric.
the device, back to Eric.
80
266042
2250
04:41
What could the implications
of something like this be?
of something like this be?
81
269125
2542
这会给我们什么样的启示呢？
04:44
Imagine perfectly memorizing things,
82
272542
3351
想象一下完美的记忆，
在你无声表达的时候
可以完美的记录信息，
可以完美的记录信息，
04:47
where you perfectly record information
that you silently speak,
that you silently speak,
83
275917
2953
然后等到你需要的时候可以去听，
04:50
and then hear them later when you want to,
84
278894
2040
还可以在内部搜寻信息，
04:52
internally searching for information,
85
280958
2393
处理数字的速度就像电脑那样，
04:55
crunching numbers at speeds computers do,
86
283375
2601
静静的向其他人发送信息。
04:58
silently texting other people.
87
286000
1917
05:00
Suddenly becoming multilingual,
88
288917
2309
突然变得多语言化，
那么当你在心里说一种语言，
05:03
so that you internally
speak in one language,
speak in one language,
89
291250
2268
你会听到另外一种语言的翻译。
05:05
and hear the translation
in your head in another.
in your head in another.
90
293542
2666
05:09
The potential could be far-reaching.
91
297125
2125
这个装置的潜力影响深远。
05:12
There are millions of people
around the world
around the world
92
300708
2101
现在世界上有数以百万计的人
挣扎于无法使用自然语言。
05:14
who struggle with using natural speech.
93
302833
2476
比如身患肌萎缩侧索硬化症，
或者葛雷克氏症的人们，
或者葛雷克氏症的人们，
05:17
People with conditions such as ALS,
or Lou Gehrig's disease,
or Lou Gehrig's disease,
94
305333
3310
中风，以及口腔癌，
05:20
stroke and oral cancer,
95
308667
1684
还有除此以外的其它情况。
05:22
amongst many other conditions.
96
310375
1708
05:25
For them, communicating is
a painstakingly slow and tiring process.
a painstakingly slow and tiring process.
97
313917
3625
对于他们来说，交流是一个
艰苦漫长和疲惫的过程。
艰苦漫长和疲惫的过程。
05:30
This is Doug.
98
318667
1601
这是道格。
道格在12年前被诊断出患了
肌萎缩侧索硬化症，
肌萎缩侧索硬化症，
05:32
Doug was diagnosed with ALS
about 12 years ago
about 12 years ago
99
320292
2309
并且从那时候开始就丧失了说话的能力。
05:34
and has since lost the ability to speak.
100
322625
1917
05:37
Today, he uses an on-screen keyboard
101
325292
1976
今天，他使用一个屏幕键盘，
通过他脑部的活动
来打出一个个单独的字母。
来打出一个个单独的字母。
05:39
where he types in individual letters
using his head movements.
using his head movements.
102
327292
3351
通常这需要好几分钟
才能交流一个简单的句子。
才能交流一个简单的句子。
05:42
And it takes several minutes
to communicate a single sentence.
to communicate a single sentence.
103
330667
2916
05:46
So we went to Doug and asked him
104
334583
1851
所以我们找到道格并问他，
通过我们的系统，你想使用
或者说出的第一个词是什么。
或者说出的第一个词是什么。
05:48
what were the first words
he'd like to use or say, using our system.
he'd like to use or say, using our system.
105
336458
4334
05:54
Perhaps a greeting, like,
"Hello, how are you?"
"Hello, how are you?"
106
342292
2851
可能是一个招呼，比如，
“嗨，你好吗？”
“嗨，你好吗？”
或者表示他需要某些帮助。
05:57
Or indicate that he needed
help with something.
help with something.
107
345167
2208
06:01
What Doug said that he wanted
to use our system for
to use our system for
108
349125
2434
道格说他希望用我们的系统
去重新启动他的旧系统，
因为旧的系统一直在崩溃。
因为旧的系统一直在崩溃。
06:03
is to reboot the old system he had,
because that old system kept on crashing.
because that old system kept on crashing.
109
351583
3643
（笑声）
06:07
(Laughter)
110
355250
1750
06:10
We never could have predicted that.
111
358083
1709
这是我们完全没有预料到的。
06:12
I'm going to show you a short clip of Doug
using our system for the first time.
using our system for the first time.
112
360917
3750
我将向大家展示一个道格
第一次使用我们系统的短片。
第一次使用我们系统的短片。
06:22
(Voice) Reboot computer.
113
370417
1625
（声音）重启电脑。
06:26
AK: What you just saw there
114
374375
1309
你们在这里看到的，
是道格自从失去说话的能力以来，
06:27
was Doug communicating or speaking
in real time for the first time
in real time for the first time
115
375708
3601
第一次实时的进行交流或者说话。
06:31
since he lost the ability to speak.
116
379333
1834
06:34
There are millions of people
117
382125
1559
有无数的人，
他们可能可以像道格一样
和其他人，他们的朋友，
和其他人，他们的朋友，
06:35
who might be able to communicate
in real time like Doug,
in real time like Doug,
118
383708
3393
他们的家人进行实时的交流。
06:39
with other people, with their friends
and with their families.
and with their families.
119
387125
3000
06:43
My hope is to be able to help them
express their thoughts and ideas.
express their thoughts and ideas.
120
391125
3458
我的希望是可以帮助他们
表达他们的想法和观点。
表达他们的想法和观点。
06:48
I believe computing, AI and the internet
121
396875
2226
我相信计算、人工智能和互联网
可以作为思维的延伸消失并融入我们，
06:51
would disappear into us
as extensions of our cognition,
as extensions of our cognition,
122
399125
3726
而不是成为外部的存在或者对手，
06:54
instead of being external
entities or adversaries,
entities or adversaries,
123
402875
3601
（它们可以）增强人类的创造力，
06:58
amplifying human ingenuity,
124
406500
2309
赋予我们无法想象的能力，
解锁我们真正的潜力。
解锁我们真正的潜力。
07:00
giving us unimaginable abilities
and unlocking our true potential.
and unlocking our true potential.
125
408833
3875
07:05
And perhaps even freeing us
to becoming better at being human.
to becoming better at being human.
126
413833
3834
也许甚至可以让我们成为更好的人类。
07:10
Thank you so much.
127
418500
1268
谢谢大家。
（掌声）
07:11
(Applause)
128
419792
5767
Shoham Arad：请站到这边。
07:17
Shoham Arad: Come over here.
129
425583
1375
07:23
OK.
130
431083
1250
好的。
07:25
I want to ask you a couple of questions,
they're going to clear the stage.
they're going to clear the stage.
131
433917
3517
我想问你一些问题，
他们需要时间来清理一下演讲台。
他们需要时间来清理一下演讲台。
我觉得这太令人惊叹了，
很新颖的点子，
很新颖的点子，
07:29
I feel like this is amazing,
it's innovative,
it's innovative,
132
437458
3518
不过听起来很诡异，有点吓人。
07:33
it's creepy, it's terrifying.
133
441000
3476
你可不可以告诉我们，
07:36
Can you tell us what I think ...
134
444500
2059
我想现在房间里
充斥着一些不安的气氛，
充斥着一些不安的气氛，
07:38
I think there are some
uncomfortable feelings around this.
uncomfortable feelings around this.
135
446583
2726
告诉我们，这个装置
会读取你的想法吗？
会读取你的想法吗？
07:41
Tell us, is this reading your thoughts,
136
449333
2310
它在五年的发展后，
会读取你的想法吗？
会读取你的想法吗？
07:43
will it in five years,
137
451667
1267
会有一个武器化版本吗，
如果有会是什么样的？
如果有会是什么样的？
07:44
is there a weaponized version of this,
what does it look like?
what does it look like?
138
452958
2917
07:48
AK: So our first design principle,
before we started working on this,
before we started working on this,
139
456833
3268
AK：在我们开始动手之前，
第一个设计原则
第一个设计原则
就是把伦理放在第一位。
07:52
was to not render ethics
as an afterthought.
as an afterthought.
140
460125
3518
所以我们把伦理融入了设计里。
07:55
So we wanted to bake ethics
right into the design.
right into the design.
141
463667
3434
我们进行了逆向设计，
07:59
We flipped the design.
142
467125
1268
我们的设计是从语言神经系统读取，
08:00
Instead of reading
from the brain directly,
from the brain directly,
143
468417
2059
而不是直接从大脑读取，
08:02
we're reading from
the voluntary nervous system
the voluntary nervous system
144
470500
2226
这样一来你就必须有意的
与装置进行交流，
与装置进行交流，
08:04
that you deliberately have to engage
to communicate with the device,
to communicate with the device,
145
472750
3191
同时仍然可以保留一个思维装置的优点。
08:07
while still bringing the benefits
of a thinking or a thought device.
of a thinking or a thought device.
146
475965
3204
这是一个对两个世界都好的方法。
08:11
The best of both worlds in a way.
147
479183
1626
SA：好的，我想大家都会
有很多问题想问你。
有很多问题想问你。
08:12
SA: OK, I think people are going to have
a lot more questions for you.
a lot more questions for you.
148
480833
3601
你说它像个贴纸。
08:16
Also, you said that it's a sticker.
149
484458
1685
所以就贴在这儿是吗？
08:18
So right now it sits just right here?
150
486167
1976
这是最终的版本吗，
08:20
Is that the final iteration,
151
488167
1684
你希望最终版是什么样的？
08:21
what the final design you hope looks like?
152
489875
3333
08:26
AK: Our goal is for the technology
to disappear completely.
to disappear completely.
153
494167
4726
AK：我们的目标是
让这个技术可以隐形。
让这个技术可以隐形。
SA：这是什么意思？
08:30
SA: What does that mean?
154
498917
1267
AK：当你把它戴上的时候，
我不应该看到它。
我不应该看到它。
08:32
AK: If you're wearing it,
I shouldn't be able to see it.
I shouldn't be able to see it.
155
500208
2643
你并不想让这个技术在你脸上，
你是想让它在后台，
你是想让它在后台，
08:34
You don't want technology on your face,
you want it in the background,
you want it in the background,
156
502875
3351
在看不见的地方强化你的能力。
08:38
to augment you in the background.
157
506250
1643
我们有一个贴纸的版本，
与皮肤颜色相似，
与皮肤颜色相似，
08:39
So we have a sticker version
that conforms to the skin,
that conforms to the skin,
158
507917
2601
看起来就像是皮肤，
08:42
that looks like the skin,
159
510542
1267
但是我们正在尝试做一个更小的版本，
08:43
but we're trying to make
an even smaller version
an even smaller version
160
511833
2268
贴在这个地方。
08:46
that would sit right here.
161
514125
1292
08:49
SA: OK.
162
517167
1266
SA：好的。
如果任何人有任何问题想问阿纳，
08:50
I feel like if anyone has any questions
they want to ask Arnav,
they want to ask Arnav,
163
518457
2976
他一整周都会在这里。
08:53
he'll be here all week.
164
521457
1269
好的，谢谢你，阿纳。
08:54
OK, thank you so much, Arnav.
165
522750
1433
谢谢。
08:56
AK: Thanks, Shoham.
166
524207
1376
Translated by Jingdan Niu
ABOUT THE SPEAKER
Arnav Kapur - TechnologistTED Fellow Arnav Kapur invents wearable AI devices that augment human cognition and give voice to those who have lost their ability to speak.
Why you should listen
At the MIT Media Lab, Arnav Kapur is inventing a device called Alter Ego, which lets users converse with machines, AI assistants and other people by articulating words internally. Utilizing bone conduction to transmit and receive streams of information, the result is a totally discreet and completely internal method of communication.
As Kapur says: "This work is about enabling people and extending human intelligence. Can we weave artificial intelligence and computing into the human condition, as an extension of our cognition -- combining human creativity and intuition with the power of AI, computation and information? Can we build technology that enables us, not distracts us; that augments us, instead of replacing us; that disappears into the background of the human experience, raising us to new levels of curiosity and creativity?"
Arnav Kapur | Speaker | TED.com