What AI is -- and isn't | Sebastian Thrun and Chris Anderson

259,143 views ・ 2017-12-21

TED


Please double-click on the English subtitles below to play the video.

00:12
Chris Anderson: Help us understand what machine learning is,
0
12904
2886
00:15
because that seems to be the key driver
1
15814
2054
00:17
of so much of the excitement and also of the concern
2
17892
2737
00:20
around artificial intelligence.
3
20653
1494
00:22
How does machine learning work?
4
22171
1643
00:23
Sebastian Thrun: So, artificial intelligence and machine learning
5
23838
3896
00:27
is about 60 years old
6
27758
2002
00:29
and has not had a great day in its past until recently.
7
29784
4269
00:34
And the reason is that today,
8
34077
2924
00:37
we have reached a scale of computing and datasets
9
37025
3973
00:41
that was necessary to make machines smart.
10
41022
2637
00:43
So here's how it works.
11
43683
1751
00:45
If you program a computer today, say, your phone,
12
45458
3497
00:48
then you hire software engineers
13
48979
2335
00:51
that write a very, very long kitchen recipe,
14
51338
3854
00:55
like, "If the water is too hot, turn down the temperature.
15
55216
3132
00:58
If it's too cold, turn up the temperature."
16
58372
2279
01:00
The recipes are not just 10 lines long.
17
60675
2849
01:03
They are millions of lines long.
18
63548
2603
01:06
A modern cell phone has 12 million lines of code.
19
66175
4084
01:10
A browser has five million lines of code.
20
70283
2646
01:12
And each bug in this recipe can cause your computer to crash.
21
72953
4969
01:17
That's why a software engineer makes so much money.
22
77946
3075
01:21
The new thing now is that computers can find their own rules.
23
81953
3660
01:25
So instead of an expert deciphering, step by step,
24
85637
3606
01:29
a rule for every contingency,
25
89267
2148
01:31
what you do now is you give the computer examples
26
91439
3074
01:34
and have it infer its own rules.
27
94537
1581
01:36
A really good example is AlphaGo, which recently was won by Google.
28
96142
4306
01:40
Normally, in game playing, you would really write down all the rules,
29
100472
3687
01:44
but in AlphaGo's case,
30
104183
1785
01:45
the system looked over a million games
31
105992
2066
01:48
and was able to infer its own rules
32
108082
2192
01:50
and then beat the world's residing Go champion.
33
110298
2738
01:53
That is exciting, because it relieves the software engineer
34
113853
3509
01:57
of the need of being super smart,
35
117386
1819
01:59
and pushes the burden towards the data.
36
119229
2325
02:01
As I said, the inflection point where this has become really possible --
37
121578
4534
02:06
very embarrassing, my thesis was about machine learning.
38
126136
2746
02:08
It was completely insignificant, don't read it,
39
128906
2205
02:11
because it was 20 years ago
40
131135
1350
02:12
and back then, the computers were as big as a cockroach brain.
41
132509
2907
02:15
Now they are powerful enough to really emulate
42
135440
2331
02:17
kind of specialized human thinking.
43
137795
2076
02:19
And then the computers take advantage of the fact
44
139895
2313
02:22
that they can look at much more data than people can.
45
142232
2500
02:24
So I'd say AlphaGo looked at more than a million games.
46
144756
3080
02:27
No human expert can ever study a million games.
47
147860
2839
02:30
Google has looked at over a hundred billion web pages.
48
150723
3182
02:33
No person can ever study a hundred billion web pages.
49
153929
2650
02:36
So as a result, the computer can find rules
50
156603
2714
02:39
that even people can't find.
51
159341
1755
02:41
CA: So instead of looking ahead to, "If he does that, I will do that,"
52
161120
4312
02:45
it's more saying, "Here is what looks like a winning pattern,
53
165456
3072
02:48
here is what looks like a winning pattern."
54
168552
2079
02:50
ST: Yeah. I mean, think about how you raise children.
55
170655
2517
02:53
You don't spend the first 18 years giving kids a rule for every contingency
56
173196
3644
02:56
and set them free and they have this big program.
57
176864
2347
02:59
They stumble, fall, get up, they get slapped or spanked,
58
179235
2719
03:01
and they have a positive experience, a good grade in school,
59
181978
2884
03:04
and they figure it out on their own.
60
184886
1834
03:06
That's happening with computers now,
61
186744
1737
03:08
which makes computer programming so much easier all of a sudden.
62
188505
3029
03:11
Now we don't have to think anymore. We just give them lots of data.
63
191558
3175
03:14
CA: And so, this has been key to the spectacular improvement
64
194757
3422
03:18
in power of self-driving cars.
65
198203
3064
03:21
I think you gave me an example.
66
201291
1739
03:23
Can you explain what's happening here?
67
203054
2685
03:25
ST: This is a drive of a self-driving car
68
205763
3564
03:29
that we happened to have at Udacity
69
209351
1957
03:31
and recently made into a spin-off called Voyage.
70
211332
2398
03:33
We have used this thing called deep learning
71
213754
2574
03:36
to train a car to drive itself,
72
216352
1623
03:37
and this is driving from Mountain View, California,
73
217999
2387
03:40
to San Francisco
74
220410
1168
03:41
on El Camino Real on a rainy day,
75
221602
2259
03:43
with bicyclists and pedestrians and 133 traffic lights.
76
223885
3524
03:47
And the novel thing here is,
77
227433
2636
03:50
many, many moons ago, I started the Google self-driving car team.
78
230093
3120
03:53
And back in the day, I hired the world's best software engineers
79
233237
3181
03:56
to find the world's best rules.
80
236442
1607
03:58
This is just trained.
81
238073
1754
03:59
We drive this road 20 times,
82
239851
3336
04:03
we put all this data into the computer brain,
83
243211
2447
04:05
and after a few hours of processing,
84
245682
2082
04:07
it comes up with behavior that often surpasses human agility.
85
247788
3926
04:11
So it's become really easy to program it.
86
251738
2017
04:13
This is 100 percent autonomous, about 33 miles, an hour and a half.
87
253779
3803
04:17
CA: So, explain it -- on the big part of this program on the left,
88
257606
3630
04:21
you're seeing basically what the computer sees as trucks and cars
89
261260
3257
04:24
and those dots overtaking it and so forth.
90
264541
2886
04:27
ST: On the right side, you see the camera image, which is the main input here,
91
267451
3762
04:31
and it's used to find lanes, other cars, traffic lights.
92
271237
2676
04:33
The vehicle has a radar to do distance estimation.
93
273937
2489
04:36
This is very commonly used in these kind of systems.
94
276450
2621
04:39
On the left side you see a laser diagram,
95
279095
1992
04:41
where you see obstacles like trees and so on depicted by the laser.
96
281111
3200
04:44
But almost all the interesting work is centering on the camera image now.
97
284335
3436
04:47
We're really shifting over from precision sensors like radars and lasers
98
287795
3476
04:51
into very cheap, commoditized sensors.
99
291295
1842
04:53
A camera costs less than eight dollars.
100
293161
1987
04:55
CA: And that green dot on the left thing, what is that?
101
295172
2793
04:57
Is that anything meaningful?
102
297989
1371
04:59
ST: This is a look-ahead point for your adaptive cruise control,
103
299384
3668
05:03
so it helps us understand how to regulate velocity
104
303076
2477
05:05
based on how far the cars in front of you are.
105
305577
2634
05:08
CA: And so, you've also got an example, I think,
106
308235
2716
05:10
of how the actual learning part takes place.
107
310975
2381
05:13
Maybe we can see that. Talk about this.
108
313380
2458
05:15
ST: This is an example where we posed a challenge to Udacity students
109
315862
3643
05:19
to take what we call a self-driving car Nanodegree.
110
319529
3131
05:22
We gave them this dataset
111
322684
1495
05:24
and said "Hey, can you guys figure out how to steer this car?"
112
324203
3054
05:27
And if you look at the images,
113
327281
1624
05:28
it's, even for humans, quite impossible to get the steering right.
114
328929
4073
05:33
And we ran a competition and said, "It's a deep learning competition,
115
333026
3591
05:36
AI competition,"
116
336641
1173
05:37
and we gave the students 48 hours.
117
337838
1887
05:39
So if you are a software house like Google or Facebook,
118
339749
4172
05:43
something like this costs you at least six months of work.
119
343945
2717
05:46
So we figured 48 hours is great.
120
346686
2202
05:48
And within 48 hours, we got about 100 submissions from students,
121
348912
3467
05:52
and the top four got it perfectly right.
122
352403
3370
05:55
It drives better than I could drive on this imagery,
123
355797
2640
05:58
using deep learning.
124
358461
1189
05:59
And again, it's the same methodology.
125
359674
1799
06:01
It's this magical thing.
126
361497
1164
06:02
When you give enough data to a computer now,
127
362685
2085
06:04
and give enough time to comprehend the data,
128
364794
2140
06:06
it finds its own rules.
129
366958
1445
06:09
CA: And so that has led to the development of powerful applications
130
369339
4845
06:14
in all sorts of areas.
131
374208
1525
06:15
You were talking to me the other day about cancer.
132
375757
2668
06:18
Can I show this video?
133
378449
1189
06:19
ST: Yeah, absolutely, please. CA: This is cool.
134
379662
2354
06:22
ST: This is kind of an insight into what's happening
135
382040
3534
06:25
in a completely different domain.
136
385598
2429
06:28
This is augmenting, or competing --
137
388051
3752
06:31
it's in the eye of the beholder --
138
391827
1749
06:33
with people who are being paid 400,000 dollars a year,
139
393600
3454
06:37
dermatologists,
140
397078
1237
06:38
highly trained specialists.
141
398339
1983
06:40
It takes more than a decade of training to be a good dermatologist.
142
400346
3561
06:43
What you see here is the machine learning version of it.
143
403931
3196
06:47
It's called a neural network.
144
407151
1841
06:49
"Neural networks" is the technical term for these machine learning algorithms.
145
409016
3742
06:52
They've been around since the 1980s.
146
412782
1789
06:54
This one was invented in 1988 by a Facebook Fellow called Yann LeCun,
147
414595
4640
06:59
and it propagates data stages
148
419259
3558
07:02
through what you could think of as the human brain.
149
422841
2578
07:05
It's not quite the same thing, but it emulates the same thing.
150
425443
2966
07:08
It goes stage after stage.
151
428433
1302
07:09
In the very first stage, it takes the visual input and extracts edges
152
429759
3637
07:13
and rods and dots.
153
433420
2612
07:16
And the next one becomes more complicated edges
154
436056
3037
07:19
and shapes like little half-moons.
155
439117
3191
07:22
And eventually, it's able to build really complicated concepts.
156
442332
4443
07:26
Andrew Ng has been able to show
157
446799
2048
07:28
that it's able to find cat faces and dog faces
158
448871
3480
07:32
in vast amounts of images.
159
452375
1661
07:34
What my student team at Stanford has shown is that
160
454060
2724
07:36
if you train it on 129,000 images of skin conditions,
161
456808
6073
07:42
including melanoma and carcinomas,
162
462905
2565
07:45
you can do as good a job
163
465494
3301
07:48
as the best human dermatologists.
164
468819
2197
07:51
And to convince ourselves that this is the case,
165
471040
2549
07:53
we captured an independent dataset that we presented to our network
166
473613
3990
07:57
and to 25 board-certified Stanford-level dermatologists,
167
477627
4342
08:01
and compared those.
168
481993
1672
08:03
And in most cases,
169
483689
1504
08:05
they were either on par or above the performance classification accuracy
170
485217
3875
08:09
of human dermatologists.
171
489116
1467
08:10
CA: You were telling me an anecdote.
172
490607
1746
08:12
I think about this image right here.
173
492377
1957
08:14
What happened here?
174
494358
1484
08:15
ST: This was last Thursday. That's a moving piece.
175
495866
4008
08:19
What we've shown before and we published in "Nature" earlier this year
176
499898
3600
08:23
was this idea that we show dermatologists images
177
503522
2484
08:26
and our computer program images,
178
506030
1539
08:27
and count how often they're right.
179
507593
1627
08:29
But all these images are past images.
180
509244
1778
08:31
They've all been biopsied to make sure we had the correct classification.
181
511046
3460
08:34
This one wasn't.
182
514530
1172
08:35
This one was actually done at Stanford by one of our collaborators.
183
515726
3179
08:38
The story goes that our collaborator,
184
518929
2314
08:41
who is a world-famous dermatologist, one of the three best, apparently,
185
521267
3391
08:44
looked at this mole and said, "This is not skin cancer."
186
524682
2935
08:47
And then he had a second moment, where he said,
187
527641
2476
08:50
"Well, let me just check with the app."
188
530141
1866
08:52
So he took out his iPhone and ran our piece of software,
189
532031
2699
08:54
our "pocket dermatologist," so to speak,
190
534754
2121
08:56
and the iPhone said: cancer.
191
536899
2994
08:59
It said melanoma.
192
539917
1306
09:01
And then he was confused.
193
541849
1233
09:03
And he decided, "OK, maybe I trust the iPhone a little bit more than myself,"
194
543106
4551
09:07
and he sent it out to the lab to get it biopsied.
195
547681
2735
09:10
And it came up as an aggressive melanoma.
196
550440
2469
09:13
So I think this might be the first time that we actually found,
197
553545
3067
09:16
in the practice of using deep learning,
198
556636
2487
09:19
an actual person whose melanoma would have gone unclassified,
199
559147
3372
09:22
had it not been for deep learning.
200
562543
2115
09:24
CA: I mean, that's incredible.
201
564682
1560
09:26
(Applause)
202
566266
1769
09:28
It feels like there'd be an instant demand for an app like this right now,
203
568059
3600
09:31
that you might freak out a lot of people.
204
571683
1966
09:33
Are you thinking of doing this, making an app that allows self-checking?
205
573673
3527
09:37
ST: So my in-box is flooded about cancer apps,
206
577224
4973
09:42
with heartbreaking stories of people.
207
582221
2303
09:44
I mean, some people have had 10, 15, 20 melanomas removed,
208
584548
3204
09:47
and are scared that one might be overlooked, like this one,
209
587776
3952
09:51
and also, about, I don't know,
210
591752
1741
09:53
flying cars and speaker inquiries these days, I guess.
211
593517
2732
09:56
My take is, we need more testing.
212
596273
2738
09:59
I want to be very careful.
213
599449
1778
10:01
It's very easy to give a flashy result and impress a TED audience.
214
601251
3666
10:04
It's much harder to put something out that's ethical.
215
604941
2627
10:07
And if people were to use the app
216
607592
2394
10:10
and choose not to consult the assistance of a doctor
217
610010
2797
10:12
because we get it wrong,
218
612831
1583
10:14
I would feel really bad about it.
219
614438
1653
10:16
So we're currently doing clinical tests,
220
616115
1925
10:18
and if these clinical tests commence and our data holds up,
221
618064
2798
10:20
we might be able at some point to take this kind of technology
222
620886
2990
10:23
and take it out of the Stanford clinic
223
623900
1892
10:25
and bring it to the entire world,
224
625816
1658
10:27
places where Stanford doctors never, ever set foot.
225
627498
2456
10:30
CA: And do I hear this right,
226
630617
2580
10:33
that it seemed like what you were saying,
227
633221
1966
10:35
because you are working with this army of Udacity students,
228
635211
4254
10:39
that in a way, you're applying a different form of machine learning
229
639489
3221
10:42
than might take place in a company,
230
642734
1735
10:44
which is you're combining machine learning with a form of crowd wisdom.
231
644493
3484
10:48
Are you saying that sometimes you think that could actually outperform
232
648001
3384
10:51
what a company can do, even a vast company?
233
651409
2050
10:53
ST: I believe there's now instances that blow my mind,
234
653483
2940
10:56
and I'm still trying to understand.
235
656447
1758
10:58
What Chris is referring to is these competitions that we run.
236
658229
3937
11:02
We turn them around in 48 hours,
237
662190
2268
11:04
and we've been able to build a self-driving car
238
664482
2252
11:06
that can drive from Mountain View to San Francisco on surface streets.
239
666758
3387
11:10
It's not quite on par with Google after seven years of Google work,
240
670169
3584
11:13
but it's getting there.
241
673777
2528
11:16
And it took us only two engineers and three months to do this.
242
676329
3084
11:19
And the reason is, we have an army of students
243
679437
2856
11:22
who participate in competitions.
244
682317
1850
11:24
We're not the only ones who use crowdsourcing.
245
684191
2220
11:26
Uber and Didi use crowdsource for driving.
246
686435
2223
11:28
Airbnb uses crowdsourcing for hotels.
247
688682
2759
11:31
There's now many examples where people do bug-finding crowdsourcing
248
691465
4007
11:35
or protein folding, of all things, in crowdsourcing.
249
695496
2804
11:38
But we've been able to build this car in three months,
250
698324
2915
11:41
so I am actually rethinking
251
701263
3655
11:44
how we organize corporations.
252
704942
2238
11:47
We have a staff of 9,000 people who are never hired,
253
707204
4696
11:51
that I never fire.
254
711924
1308
11:53
They show up to work and I don't even know.
255
713256
2362
11:55
Then they submit to me maybe 9,000 answers.
256
715642
3058
11:58
I'm not obliged to use any of those.
257
718724
2176
12:00
I end up -- I pay only the winners,
258
720924
1991
12:02
so I'm actually very cheapskate here, which is maybe not the best thing to do.
259
722939
3718
12:06
But they consider it part of their education, too, which is nice.
260
726681
3185
12:09
But these students have been able to produce amazing deep learning results.
261
729890
4201
12:14
So yeah, the synthesis of great people and great machine learning is amazing.
262
734115
3861
12:18
CA: I mean, Gary Kasparov said on the first day [of TED2017]
263
738000
2814
12:20
that the winners of chess, surprisingly, turned out to be two amateur chess players
264
740848
5412
12:26
with three mediocre-ish, mediocre-to-good, computer programs,
265
746284
5371
12:31
that could outperform one grand master with one great chess player,
266
751679
3163
12:34
like it was all part of the process.
267
754866
1743
12:36
And it almost seems like you're talking about a much richer version
268
756633
3335
12:39
of that same idea.
269
759992
1200
12:41
ST: Yeah, I mean, as you followed the fantastic panels yesterday morning,
270
761216
3857
12:45
two sessions about AI,
271
765097
1994
12:47
robotic overlords and the human response,
272
767115
2167
12:49
many, many great things were said.
273
769306
1982
12:51
But one of the concerns is that we sometimes confuse
274
771312
2687
12:54
what's actually been done with AI with this kind of overlord threat,
275
774023
4062
12:58
where your AI develops consciousness, right?
276
778109
3424
13:01
The last thing I want is for my AI to have consciousness.
277
781557
2971
13:04
I don't want to come into my kitchen
278
784552
1716
13:06
and have the refrigerator fall in love with the dishwasher
279
786292
4193
13:10
and tell me, because I wasn't nice enough,
280
790509
2124
13:12
my food is now warm.
281
792657
1837
13:14
I wouldn't buy these products, and I don't want them.
282
794518
2891
13:17
But the truth is, for me,
283
797825
1802
13:19
AI has always been an augmentation of people.
284
799651
2720
13:22
It's been an augmentation of us,
285
802893
1676
13:24
to make us stronger.
286
804593
1457
13:26
And I think Kasparov was exactly correct.
287
806074
2831
13:28
It's been the combination of human smarts and machine smarts
288
808929
3849
13:32
that make us stronger.
289
812802
1464
13:34
The theme of machines making us stronger is as old as machines are.
290
814290
4587
13:39
The agricultural revolution took place because it made steam engines
291
819567
3758
13:43
and farming equipment that couldn't farm by itself,
292
823349
2666
13:46
that never replaced us; it made us stronger.
293
826039
2122
13:48
And I believe this new wave of AI will make us much, much stronger
294
828185
3738
13:51
as a human race.
295
831947
1183
13:53
CA: We'll come on to that a bit more,
296
833765
1813
13:55
but just to continue with the scary part of this for some people,
297
835602
3671
13:59
like, what feels like it gets scary for people is when you have
298
839297
3558
14:02
a computer that can, one, rewrite its own code,
299
842879
4618
14:07
so, it can create multiple copies of itself,
300
847521
3584
14:11
try a bunch of different code versions,
301
851129
1897
14:13
possibly even at random,
302
853050
1775
14:14
and then check them out and see if a goal is achieved and improved.
303
854849
3632
14:18
So, say the goal is to do better on an intelligence test.
304
858505
3641
14:22
You know, a computer that's moderately good at that,
305
862170
3894
14:26
you could try a million versions of that.
306
866088
2509
14:28
You might find one that was better,
307
868621
2090
14:30
and then, you know, repeat.
308
870735
2004
14:32
And so the concern is that you get some sort of runaway effect
309
872763
3040
14:35
where everything is fine on Thursday evening,
310
875827
3008
14:38
and you come back into the lab on Friday morning,
311
878859
2336
14:41
and because of the speed of computers and so forth,
312
881219
2449
14:43
things have gone crazy, and suddenly --
313
883692
1903
14:45
ST: I would say this is a possibility,
314
885619
2020
14:47
but it's a very remote possibility.
315
887663
1916
14:49
So let me just translate what I heard you say.
316
889603
3337
14:52
In the AlphaGo case, we had exactly this thing:
317
892964
2704
14:55
the computer would play the game against itself
318
895692
2315
14:58
and then learn new rules.
319
898031
1250
14:59
And what machine learning is is a rewriting of the rules.
320
899305
3235
15:02
It's the rewriting of code.
321
902564
1769
15:04
But I think there was absolutely no concern
322
904357
2845
15:07
that AlphaGo would take over the world.
323
907226
2426
15:09
It can't even play chess.
324
909676
1464
15:11
CA: No, no, no, but now, these are all very single-domain things.
325
911164
5147
15:16
But it's possible to imagine.
326
916335
2879
15:19
I mean, we just saw a computer that seemed nearly capable
327
919238
3089
15:22
of passing a university entrance test,
328
922351
2655
15:25
that can kind of -- it can't read and understand in the sense that we can,
329
925030
3688
15:28
but it can certainly absorb all the text
330
928742
1987
15:30
and maybe see increased patterns of meaning.
331
930753
2899
15:33
Isn't there a chance that, as this broadens out,
332
933676
3694
15:37
there could be a different kind of runaway effect?
333
937394
2466
15:39
ST: That's where I draw the line, honestly.
334
939884
2078
15:41
And the chance exists -- I don't want to downplay it --
335
941986
2643
15:44
but I think it's remote, and it's not the thing that's on my mind these days,
336
944653
3672
15:48
because I think the big revolution is something else.
337
948349
2512
15:50
Everything successful in AI to the present date
338
950885
2922
15:53
has been extremely specialized,
339
953831
2214
15:56
and it's been thriving on a single idea,
340
956069
2489
15:58
which is massive amounts of data.
341
958582
2739
16:01
The reason AlphaGo works so well is because of massive numbers of Go plays,
342
961345
4147
16:05
and AlphaGo can't drive a car or fly a plane.
343
965516
3255
16:08
The Google self-driving car or the Udacity self-driving car
344
968795
3031
16:11
thrives on massive amounts of data, and it can't do anything else.
345
971850
3240
16:15
It can't even control a motorcycle.
346
975114
1727
16:16
It's a very specific, domain-specific function,
347
976865
2762
16:19
and the same is true for our cancer app.
348
979651
1907
16:21
There has been almost no progress on this thing called "general AI,"
349
981582
3236
16:24
where you go to an AI and say, "Hey, invent for me special relativity
350
984842
4000
16:28
or string theory."
351
988866
1666
16:30
It's totally in the infancy.
352
990556
1931
16:32
The reason I want to emphasize this,
353
992511
2127
16:34
I see the concerns, and I want to acknowledge them.
354
994662
3838
16:38
But if I were to think about one thing,
355
998524
2886
16:41
I would ask myself the question, "What if we can take anything repetitive
356
1001434
5563
16:47
and make ourselves 100 times as efficient?"
357
1007021
3473
16:51
It so turns out, 300 years ago, we all worked in agriculture
358
1011170
4249
16:55
and did farming and did repetitive things.
359
1015443
2051
16:57
Today, 75 percent of us work in offices
360
1017518
2556
17:00
and do repetitive things.
361
1020098
2124
17:02
We've become spreadsheet monkeys.
362
1022246
2183
17:04
And not just low-end labor.
363
1024453
2054
17:06
We've become dermatologists doing repetitive things,
364
1026531
2754
17:09
lawyers doing repetitive things.
365
1029309
1749
17:11
I think we are at the brink of being able to take an AI,
366
1031082
3823
17:14
look over our shoulders,
367
1034929
1718
17:16
and they make us maybe 10 or 50 times as effective in these repetitive things.
368
1036671
4058
17:20
That's what is on my mind.
369
1040753
1275
17:22
CA: That sounds super exciting.
370
1042052
2450
17:24
The process of getting there seems a little terrifying to some people,
371
1044526
3530
17:28
because once a computer can do this repetitive thing
372
1048080
3180
17:31
much better than the dermatologist
373
1051284
3434
17:34
or than the driver, especially, is the thing that's talked about
374
1054742
3230
17:37
so much now,
375
1057996
1290
17:39
suddenly millions of jobs go,
376
1059310
1958
17:41
and, you know, the country's in revolution
377
1061292
2695
17:44
before we ever get to the more glorious aspects of what's possible.
378
1064011
4329
17:48
ST: Yeah, and that's an issue, and it's a big issue,
379
1068364
2517
17:50
and it was pointed out yesterday morning by several guest speakers.
380
1070905
4196
17:55
Now, prior to me showing up onstage,
381
1075125
2754
17:57
I confessed I'm a positive, optimistic person,
382
1077903
3739
18:01
so let me give you an optimistic pitch,
383
1081666
2389
18:04
which is, think of yourself back 300 years ago.
384
1084079
4795
18:08
Europe just survived 140 years of continuous war,
385
1088898
3996
18:12
none of you could read or write,
386
1092918
1711
18:14
there were no jobs that you hold today,
387
1094653
2945
18:17
like investment banker or software engineer or TV anchor.
388
1097622
4096
18:21
We would all be in the fields and farming.
389
1101742
2414
18:24
Now here comes little Sebastian with a little steam engine in his pocket,
390
1104180
3573
18:27
saying, "Hey guys, look at this.
391
1107777
1548
18:29
It's going to make you 100 times as strong, so you can do something else."
392
1109349
3595
18:32
And then back in the day, there was no real stage,
393
1112968
2470
18:35
but Chris and I hang out with the cows in the stable,
394
1115462
2526
18:38
and he says, "I'm really concerned about it,
395
1118012
2100
18:40
because I milk my cow every day, and what if the machine does this for me?"
396
1120136
3652
18:43
The reason why I mention this is,
397
1123812
1702
18:46
we're always good in acknowledging past progress and the benefit of it,
398
1126360
3603
18:49
like our iPhones or our planes or electricity or medical supply.
399
1129987
3354
18:53
We all love to live to 80, which was impossible 300 years ago.
400
1133365
4245
18:57
But we kind of don't apply the same rules to the future.
401
1137634
4156
19:02
So if I look at my own job as a CEO,
402
1142621
3207
19:05
I would say 90 percent of my work is repetitive,
403
1145852
3140
19:09
I don't enjoy it,
404
1149016
1351
19:10
I spend about four hours per day on stupid, repetitive email.
405
1150391
3978
19:14
And I'm burning to have something that helps me get rid of this.
406
1154393
3641
19:18
Why?
407
1158058
1158
19:19
Because I believe all of us are insanely creative;
408
1159240
3003
19:22
I think the TED community more than anybody else.
409
1162731
3194
19:25
But even blue-collar workers; I think you can go to your hotel maid
410
1165949
3559
19:29
and have a drink with him or her,
411
1169532
2402
19:31
and an hour later, you find a creative idea.
412
1171958
2717
19:34
What this will empower is to turn this creativity into action.
413
1174699
4140
19:39
Like, what if you could build Google in a day?
414
1179265
3442
19:43
What if you could sit over beer and invent the next Snapchat,
415
1183221
3316
19:46
whatever it is,
416
1186561
1165
19:47
and tomorrow morning it's up and running?
417
1187750
2187
19:49
And that is not science fiction.
418
1189961
1773
19:51
What's going to happen is,
419
1191758
1254
19:53
we are already in history.
420
1193036
1867
19:54
We've unleashed this amazing creativity
421
1194927
3228
19:58
by de-slaving us from farming
422
1198179
1611
19:59
and later, of course, from factory work
423
1199814
3363
20:03
and have invented so many things.
424
1203201
3162
20:06
It's going to be even better, in my opinion.
425
1206387
2178
20:08
And there's going to be great side effects.
426
1208589
2072
20:10
One of the side effects will be
427
1210685
1489
20:12
that things like food and medical supply and education and shelter
428
1212198
4795
20:17
and transportation
429
1217017
1177
20:18
will all become much more affordable to all of us,
430
1218218
2441
20:20
not just the rich people.
431
1220683
1322
20:22
CA: Hmm.
432
1222029
1182
20:23
So when Martin Ford argued, you know, that this time it's different
433
1223235
4341
20:27
because the intelligence that we've used in the past
434
1227600
3453
20:31
to find new ways to be
435
1231077
2483
20:33
will be matched at the same pace
436
1233584
2279
20:35
by computers taking over those things,
437
1235887
2291
20:38
what I hear you saying is that, not completely,
438
1238202
3078
20:41
because of human creativity.
439
1241304
2951
20:44
Do you think that that's fundamentally different from the kind of creativity
440
1244279
3785
20:48
that computers can do?
441
1248088
2696
20:50
ST: So, that's my firm belief as an AI person --
442
1250808
4434
20:55
that I haven't seen any real progress on creativity
443
1255266
3803
20:59
and out-of-the-box thinking.
444
1259949
1407
21:01
What I see right now -- and this is really important for people to realize,
445
1261380
3623
21:05
because the word "artificial intelligence" is so threatening,
446
1265027
2903
21:07
and then we have Steve Spielberg tossing a movie in,
447
1267954
2523
21:10
where all of a sudden the computer is our overlord,
448
1270501
2413
21:12
but it's really a technology.
449
1272938
1452
21:14
It's a technology that helps us do repetitive things.
450
1274414
2982
21:17
And the progress has been entirely on the repetitive end.
451
1277420
2913
21:20
It's been in legal document discovery.
452
1280357
2228
21:22
It's been contract drafting.
453
1282609
1680
21:24
It's been screening X-rays of your chest.
454
1284313
4223
21:28
And these things are so specialized,
455
1288560
1773
21:30
I don't see the big threat of humanity.
456
1290357
2391
21:32
In fact, we as people --
457
1292772
1794
21:34
I mean, let's face it: we've become superhuman.
458
1294590
2385
21:36
We've made us superhuman.
459
1296999
1764
21:38
We can swim across the Atlantic in 11 hours.
460
1298787
2632
21:41
We can take a device out of our pocket
461
1301443
2074
21:43
and shout all the way to Australia,
462
1303541
2147
21:45
and in real time, have that person shouting back to us.
463
1305712
2600
21:48
That's physically not possible. We're breaking the rules of physics.
464
1308336
3624
21:51
When this is said and done, we're going to remember everything
465
1311984
2943
21:54
we've ever said and seen,
466
1314951
1213
21:56
you'll remember every person,
467
1316188
1496
21:57
which is good for me in my early stages of Alzheimer's.
468
1317708
2626
22:00
Sorry, what was I saying? I forgot.
469
1320358
1677
22:02
CA: (Laughs)
470
1322059
1578
22:03
ST: We will probably have an IQ of 1,000 or more.
471
1323661
3077
22:06
There will be no more spelling classes for our kids,
472
1326762
3425
22:10
because there's no spelling issue anymore.
473
1330211
2086
22:12
There's no math issue anymore.
474
1332321
1832
22:14
And I think what really will happen is that we can be super creative.
475
1334177
3510
22:17
And we are. We are creative.
476
1337711
1857
22:19
That's our secret weapon.
477
1339592
1552
22:21
CA: So the jobs that are getting lost,
478
1341168
2153
22:23
in a way, even though it's going to be painful,
479
1343345
2494
22:25
humans are capable of more than those jobs.
480
1345863
2047
22:27
This is the dream.
481
1347934
1218
22:29
The dream is that humans can rise to just a new level of empowerment
482
1349176
4247
22:33
and discovery.
483
1353447
1657
22:35
That's the dream.
484
1355128
1452
22:36
ST: And think about this:
485
1356604
1643
22:38
if you look at the history of humanity,
486
1358271
2021
22:40
that might be whatever -- 60-100,000 years old, give or take --
487
1360316
3328
22:43
almost everything that you cherish in terms of invention,
488
1363668
3726
22:47
of technology, of things we've built,
489
1367418
2151
22:49
has been invented in the last 150 years.
490
1369593
3099
22:53
If you toss in the book and the wheel, it's a little bit older.
491
1373756
3048
22:56
Or the axe.
492
1376828
1169
22:58
But your phone, your sneakers,
493
1378021
2790
23:00
these chairs, modern manufacturing, penicillin --
494
1380835
3551
23:04
the things we cherish.
495
1384410
1714
23:06
Now, that to me means
496
1386148
3658
23:09
the next 150 years will find more things.
497
1389830
3041
23:12
In fact, the pace of invention has gone up, not gone down, in my opinion.
498
1392895
4154
23:17
I believe only one percent of interesting things have been invented yet. Right?
499
1397073
4905
23:22
We haven't cured cancer.
500
1402002
1988
23:24
We don't have flying cars -- yet. Hopefully, I'll change this.
501
1404014
3718
23:27
That used to be an example people laughed about. (Laughs)
502
1407756
3257
23:31
It's funny, isn't it? Working secretly on flying cars.
503
1411037
2992
23:34
We don't live twice as long yet. OK?
504
1414053
2683
23:36
We don't have this magic implant in our brain
505
1416760
2785
23:39
that gives us the information we want.
506
1419569
1832
23:41
And you might be appalled by it,
507
1421425
1526
23:42
but I promise you, once you have it, you'll love it.
508
1422975
2444
23:45
I hope you will.
509
1425443
1166
23:46
It's a bit scary, I know.
510
1426633
1909
23:48
There are so many things we haven't invented yet
511
1428566
2254
23:50
that I think we'll invent.
512
1430844
1268
23:52
We have no gravity shields.
513
1432136
1306
23:53
We can't beam ourselves from one location to another.
514
1433466
2553
23:56
That sounds ridiculous,
515
1436043
1151
23:57
but about 200 years ago,
516
1437218
1288
23:58
experts were of the opinion that flight wouldn't exist,
517
1438530
2667
24:01
even 120 years ago,
518
1441221
1324
24:02
and if you moved faster than you could run,
519
1442569
2582
24:05
you would instantly die.
520
1445175
1520
24:06
So who says we are correct today that you can't beam a person
521
1446719
3569
24:10
from here to Mars?
522
1450312
2249
24:12
CA: Sebastian, thank you so much
523
1452585
1569
24:14
for your incredibly inspiring vision and your brilliance.
524
1454178
2682
24:16
Thank you, Sebastian Thrun.
525
1456884
1323
24:18
That was fantastic. (Applause)
526
1458231
1895
About this website

This site will introduce you to YouTube videos that are useful for learning English. You will see English lessons taught by top-notch teachers from around the world. Double-click on the English subtitles displayed on each video page to play the video from there. The subtitles scroll in sync with the video playback. If you have any comments or requests, please contact us using this contact form.

https://forms.gle/WvT1wiN1qDtmnspy7