TEDxBerkeley
Ken Goldberg: 4 lessons from robots about being human
Filmed:
Readability: 4.6
387,467 views
The more that robots ingrain themselves into our everyday lives, the more we're forced to examine ourselves as people. At TEDxBerkeley, Ken Goldberg shares four very human lessons that he's learned from working with robots.
Ken Goldberg - Roboticist
Ken Goldberg works reflect the intersection of robotics, social media, and art. Full bio
Ken Goldberg works reflect the intersection of robotics, social media, and art. Full bio
Double-click the English transcript below to play the video.
00:16
I know this is going to sound strange,
0
577
2475
00:18
but I think robots can inspire us
1
3052
3294
00:22
to be better humans.
2
6346
2388
00:24
See, I grew up in Bethlehem, Pennsylvania,
3
8734
3629
00:28
the home of Bethlehem Steel.
4
12363
2321
00:30
My father was an engineer,
5
14684
2079
00:32
and when I was growing up, he would teach me
6
16763
2851
00:35
how things worked.
7
19614
1159
00:36
We would build projects together,
8
20773
2815
00:39
like model rockets and slot cars.
9
23588
2792
00:42
Here's the go-kart that we built together.
10
26380
3284
00:45
That's me behind the wheel,
11
29664
2219
00:47
with my sister and my best friend at the time,
12
31883
3819
00:51
and one day,
13
35702
2123
00:53
he came home, when I was about 10 years old,
14
37825
3107
00:56
and at the dinner table, he announced
15
40932
2304
00:59
that for our next project, we were going to build a robot.
16
43236
6438
01:05
A robot.
17
49674
1109
01:06
Now, I was thrilled about this,
18
50783
2252
01:08
because at school,
19
53035
1590
01:10
there was a bully named Kevin,
20
54625
2075
01:12
and he was picking on me
21
56700
2246
01:14
because I was the only Jewish kid in class.
22
58946
2414
01:17
So I couldn't wait to get started to work on this
23
61360
2813
01:20
so I could introduce Kevin to my robot. (Laughter)
24
64173
3822
01:23
(Robot noises)
25
67995
11063
01:34
But that wasn't the kind of robot my dad had in mind.
26
79058
4875
01:39
See, he owned a chromium plating company,
27
83933
3728
01:43
and they had to move
28
87661
2176
01:45
heavy steel parts between tanks of chemicals,
29
89837
3096
01:48
and so he needed an industrial robot like this
30
92933
3790
01:52
that could basically do the heavy lifting.
31
96723
3099
01:55
But my dad didn't get the kind of robot he wanted, either.
32
99822
3905
01:59
He and I worked on it for several years,
33
103727
2597
02:02
but it was the 1970s,
34
106324
1978
02:04
and the technology that was available to amateurs
35
108302
2526
02:06
just wasn't there yet.
36
110828
2657
02:09
So Dad continued to do this kind of work by hand,
37
113485
3745
02:13
and a few years later,
38
117230
2275
02:15
he was diagnosed with cancer.
39
119505
4139
02:19
You see, what the robot we were trying to build
40
123644
3305
02:22
was telling him was not about doing the heavy lifting.
41
126949
3267
02:26
It was a warning about his exposure to the toxic chemicals.
42
130216
4308
02:30
He didn't recognize that at the time,
43
134524
2977
02:33
and he contracted leukemia,
44
137501
2139
02:35
and he died at the age of 45.
45
139640
3222
02:38
I was devastated by this,
46
142862
2841
02:41
and I never forgot the robot that he and I tried to build.
47
145703
4575
02:46
When I was in college, I decided to study engineering, like him.
48
150278
4311
02:50
And I went to Carnegie Mellon, and I earned my PhD in robotics.
49
154589
4946
02:55
I've been studying robots ever since.
50
159535
3094
02:58
So what I'd like to tell you about
51
162629
1638
03:00
are four robot projects
52
164267
2714
03:02
and how they've inspired me to be a better human.
53
166981
7044
03:09
By 1993, I was a young professor at USC,
54
174025
5749
03:15
and I was just building up my own robotics lab,
55
179774
3037
03:18
and this was the year that the World Wide Web came out.
56
182811
3585
03:22
And I remember my students were the ones
57
186396
1176
03:23
who told me about it,
58
187572
1704
03:25
and we would -- we were just amazed.
59
189276
2619
03:27
We started playing with this, and that afternoon,
60
191895
3434
03:31
we realized that we could use this new, universal interface
61
195329
3853
03:35
to allow anyone in the world
62
199182
2543
03:37
to operate the robot in our lab.
63
201725
3048
03:40
So, rather than have it fight or do industrial work,
64
204773
5708
03:46
we decided to build a planter,
65
210481
2486
03:48
put the robot into the center of it,
66
212967
1976
03:50
and we called it the Telegarden.
67
214943
2353
03:53
And we had put a camera in the gripper of the hand
68
217296
3603
03:56
of the robot, and we wrote some special scripts
69
220899
2709
03:59
and software so that anyone in the world could come in
70
223608
3161
04:02
and by clicking on the screen
71
226769
2329
04:04
they could move the robot around
72
229098
2173
04:07
and visit the garden.
73
231271
2316
04:09
But we also allowed, set up some other software
74
233587
3620
04:13
that lets you participate and help us water the garden
75
237207
3419
04:16
remotely, and if you water it a few times,
76
240626
3045
04:19
we'd give you your own seed to plant.
77
243671
3585
04:23
Now, this was a project, an engineering project,
78
247256
3271
04:26
and we published some papers on the design,
79
250527
2776
04:29
the system design of it, but we also thought of it
80
253303
2249
04:31
as an art installation.
81
255552
3086
04:34
It was invited, after the first year,
82
258638
2173
04:36
by the Ars Electronica Museum in Austria
83
260811
3044
04:39
to have it installed in their lobby,
84
263855
3000
04:42
and I'm happy to say it remained online there,
85
266855
2418
04:45
24 hours a day, for almost nine years.
86
269273
4983
04:50
That robot was operated by more people
87
274256
3799
04:53
than any other robot in history.
88
278055
3086
04:57
Now, one day,
89
281141
1554
04:58
I got a call out of the blue
90
282695
2317
05:00
from a student,
91
285012
2015
05:02
who asked a very simple but profound question.
92
287027
4670
05:07
He said, "Is the robot real?"
93
291697
4538
05:12
Now, everyone else had assumed it was,
94
296235
2765
05:14
and we knew it was because we were working with it.
95
299000
2331
05:17
But I knew what he meant,
96
301331
1540
05:18
because it would be possible to take a bunch of pictures
97
302871
2593
05:21
of flowers in a garden and then, basically, index them
98
305464
4361
05:25
in a computer system such that it would appear
99
309825
2229
05:27
that there was a real robot when there wasn't.
100
312054
3074
05:31
And the more I thought about it, I couldn't think
101
315128
1241
05:32
of a good answer for how he could tell the difference.
102
316369
3551
05:35
This was right about the time that I was offered a position
103
319920
2887
05:38
here at Berkeley,
104
322807
2073
05:40
and when I got here, I looked up Hubert Dreyfus,
105
324880
3552
05:44
who's a world-renowned professor of philosophy,
106
328432
3665
05:47
and I talked with him about this, and he said,
107
332097
2374
05:50
"This is one of the oldest and most central problems
108
334471
3560
05:53
in philosophy. It goes back to the Skeptics,
109
338031
3602
05:57
and up through Descartes.
110
341633
1960
05:59
It's the issue of epistemology,
111
343593
3170
06:02
the study of how do we know that something is true."
112
346763
3750
06:06
So he and I started working together,
113
350513
2223
06:08
and we coined a new term: telepistemology,
114
352736
3041
06:11
the study of knowledge at a distance.
115
355777
3526
06:15
We invited leading artists, engineers,
116
359303
2752
06:17
and philosophers to write essays about this,
117
362055
3064
06:21
and the results, the results are collected in this book
118
365119
2340
06:23
from MIT Press.
119
367459
2621
06:25
So thanks to this student who questioned
120
370080
2332
06:28
what everyone else had assumed to be true,
121
372412
2803
06:31
this project taught me an important lesson about life,
122
375215
4208
06:35
which is to always question assumptions.
123
379423
4073
06:39
Now, the second project I'll tell you about
124
383496
2648
06:42
grew out of the Telegarden.
125
386144
1879
06:43
As it was operating, my students and I were very interested
126
388023
2576
06:46
in how people were interacting with each other
127
390599
2762
06:49
and what they were doing with the garden.
128
393361
1621
06:50
So we started thinking, what if the robot could leave
129
394982
2262
06:53
the garden and go out into some other
130
397244
2083
06:55
interesting environment?
131
399327
1992
06:57
Like, for example, what if it could go to a dinner party
132
401319
2179
06:59
at the White House? (Laughter)
133
403498
5198
07:04
So, because we were interested more in the system design
134
408696
2841
07:07
and the user interface than in the hardware,
135
411537
3286
07:10
we decided that, rather than have
136
414823
2034
07:12
a robot replace the human to go to the party,
137
416857
3678
07:16
we'd have a human replace the robot.
138
420535
2843
07:19
We called it the Tele-Actor.
139
423378
2600
07:21
We got a human,
140
425978
2032
07:23
someone who's very outgoing and gregarious,
141
428010
3007
07:26
and she was outfitted with a helmet
142
431017
3142
07:30
with various equipment, cameras and microphones,
143
434159
2570
07:32
and then a backpack with wireless Internet connection,
144
436729
3740
07:36
and the idea was that she could go into a remote and
145
440469
3359
07:39
interesting environment, and then over the Internet,
146
443828
3815
07:43
people could experience what she was experiencing,
147
447643
3151
07:46
so they could see what she was seeing,
148
450794
2937
07:49
but then, more importantly, they could participate
149
453731
3307
07:52
by interacting with each other
150
457038
2948
07:55
and coming up with ideas about what she should do next
151
459986
3705
07:59
and where she should go,
152
463691
2216
08:01
and then conveying those to the Tele-Actor.
153
465907
3162
08:04
So we got a chance to take the Tele-Actor
154
469069
2445
08:07
to the Webby Awards in San Francisco,
155
471514
3615
08:11
and that year, Sam Donaldson was the host.
156
475129
4111
08:15
Just before the curtain went up, I had about 30 seconds
157
479240
3642
08:18
to explain to Mr. Donaldson what we were gonna do,
158
482882
4458
08:23
and I said, "The Tele-Actor
159
487340
1864
08:25
is going to be joining you on stage,
160
489204
2456
08:27
and this is a new experimental project,
161
491660
2198
08:29
and people are watching her on their screens,
162
493858
2625
08:32
and she's got -- there's cameras involved and there's
163
496483
2305
08:34
microphones and she's got an earbud in her ear,
164
498788
2911
08:37
and people over the network are giving her advice
165
501699
1447
08:39
about what to do next."
166
503146
1368
08:40
And he said, "Wait a second,
167
504514
3209
08:43
that's what I do." (Laughter)
168
507723
6375
08:49
So he loved the concept,
169
514098
1931
08:51
and when the Tele-Actor walked onstage,
170
516029
2340
08:54
she walked right up to him, and she gave him a big kiss
171
518369
2821
08:57
right on the lips. (Laughter)
172
521190
3106
09:00
We were totally surprised.
173
524296
1131
09:01
We had no idea that would happen.
174
525427
2017
09:03
And he was great. He just gave her a big hug in return,
175
527444
2692
09:06
and it worked out great.
176
530136
1769
09:07
But that night, as we were packing up,
177
531905
2064
09:09
I asked the Tele-Actor, how did the Tele-Directors
178
533969
3608
09:13
decide that they would give a kiss to Sam Donaldson?
179
537577
5558
09:19
And she said they hadn't.
180
543135
2212
09:21
She said, when she was just about to walk on stage,
181
545347
2434
09:23
the Tele-Directors were still trying to agree on what to do,
182
547781
2312
09:25
and so she just walked on stage and did
183
550093
2407
09:28
what felt most natural. (Laughter)
184
552500
5522
09:33
So, the success of the Tele-Actor that night
185
558022
3670
09:37
was due to the fact that she was a wonderful actor.
186
561692
4373
09:41
She knew when to trust her instincts,
187
566065
2357
09:44
and so that project taught me another lesson about life,
188
568422
3864
09:48
which is that, when in doubt, improvise. (Laughter)
189
572286
6379
09:54
Now, the third project grew out of
190
578665
3080
09:57
my experience when my father was in the hospital.
191
581745
4893
10:02
He was undergoing a treatment,
192
586638
2240
10:04
chemotherapy treatments, and there's a related treatment
193
588878
3633
10:08
called brachytherapy, where tiny, radioactive seeds
194
592511
5048
10:13
are placed into the body to treat cancerous tumors.
195
597559
4211
10:17
And the way it's done, as you can see here,
196
601770
2097
10:19
is that surgeons insert needles into the body
197
603867
4391
10:24
to deliver the seeds, and all this,
198
608258
2592
10:26
all these needles are inserted in parallel,
199
610850
3433
10:30
so it's very common that some of the needles
200
614283
2975
10:33
penetrate sensitive organs, and as a result,
201
617258
4817
10:37
the needles damage these organs, cause damage
202
622075
5063
10:43
which leads to trauma and side effects.
203
627138
3487
10:46
So my students and I wondered, what if we could
204
630625
2500
10:49
modify the system
205
633125
3485
10:52
so that the needles could come in at different angles?
206
636610
3785
10:56
So we simulated this, and we developed some
207
640395
2777
10:59
optimization algorithms and we simulated this,
208
643172
2589
11:01
and we were able to show that we are able to avoid
209
645761
2277
11:03
the delicate organs and yet still achieve the coverage
210
648038
3767
11:07
of the tumors with the radiation.
211
651805
3508
11:11
So now, we're working with doctors at UCSF
212
655313
3498
11:14
and engineers at Johns Hopkins
213
658811
2696
11:17
and we're building a robot that has a number of,
214
661507
3582
11:20
it's a specialized design with different joints that can allow
215
665089
3257
11:24
the needles to come in at an infinite variety of angles,
216
668346
4258
11:28
and as you can see here, they can avoid delicate organs
217
672604
3166
11:31
and still reach the targets they're aiming for.
218
675770
4138
11:35
So, by questioning this assumption that all the needles
219
679908
3317
11:39
have to be parallel, this project also taught me
220
683225
2945
11:42
an important lesson: When in doubt --
221
686170
3300
11:45
When your path is blocked, pivot.
222
689470
4367
11:49
And the last project also has to do with medical robotics.
223
693837
4401
11:54
And this is something that's grown out of a system called
224
698238
4040
11:58
the da Vinci surgical robot,
225
702278
3588
12:01
and this is a commercially available device.
226
705866
2468
12:04
It's being used in over 2,000 hospitals around the world,
227
708334
3332
12:07
and the idea is it allows the surgeon
228
711666
2528
12:10
to operate comfortably in his own coordinate frame,
229
714194
4249
12:14
but many of the subtasks in surgery
230
718443
4991
12:19
are very routine and tedious, like suturing,
231
723434
3047
12:22
and currently, all of these are performed
232
726481
2365
12:24
under the specific and immediate control of the surgeon,
233
728846
4420
12:29
so the surgeon becomes fatigued over time.
234
733266
2658
12:31
And we've been wondering,
235
735924
1295
12:33
what if we could program the robot
236
737219
2265
12:35
to perform some of these subtasks,
237
739484
2487
12:37
and thereby free the surgeons to focus
238
741971
1720
12:39
on the more complicated parts of the surgery,
239
743691
2656
12:42
and also cut down on the time that the surgery would take
240
746347
3248
12:45
if we could get the robot to do them a little bit faster?
241
749595
3023
12:48
Now, it's hard to program a robot to do delicate things
242
752618
2434
12:50
like this, but it turns out my colleague, Pieter Abbeel,
243
755052
4079
12:55
who's here at Berkeley, has develeloped
244
759131
2416
12:57
a new set of techniques for teaching robots from example.
245
761547
5623
13:03
So he's gotten robots to fly helicopters,
246
767170
2767
13:05
do incredibly interesting, beautiful acrobatics,
247
769937
3104
13:08
by watching human experts fly them.
248
773041
2719
13:11
So we got one of these robots.
249
775760
2588
13:14
We started working with Pieter and his students,
250
778348
2182
13:16
and we asked a surgeon to perform
251
780530
2663
13:19
a task, and what we do is we, with the robot,
252
783193
4451
13:23
so what we're doing is asking the robot,
253
787644
2063
13:25
the surgeon to perform the task,
254
789707
1278
13:26
and we record the motions of the robot.
255
790985
2272
13:29
So here's an example. I'll use a figure eight,
256
793257
2128
13:31
tracing out a figure eight as an example.
257
795385
2240
13:33
So here's what it looks like when the robot,
258
797625
3630
13:37
this is what the robot's path looks like,
259
801255
2317
13:39
those three examples.
260
803572
1174
13:40
Now, those are much better than what a novice
261
804746
2462
13:43
like I could do, but they're still jerky and imprecise.
262
807208
4657
13:47
So we record all these examples, the data,
263
811865
2072
13:49
and then we go through a sequence of steps.
264
813937
3712
13:53
First, we used a technique called dynamic time warping
265
817649
3632
13:57
from speech recognition, and this allows us to
266
821281
2182
13:59
temporally align all of the examples,
267
823463
2840
14:02
and then we apply Kalman filtering,
268
826303
2929
14:05
a technique from control theory, that allows us
269
829232
2983
14:08
to statistically analyze all the noise
270
832215
2672
14:10
and extract the desired trajectory that underlies them.
271
834887
6183
14:16
Now, so what we're doing is that we take those
272
841070
1994
14:18
human demonstrations, they're all noisy and imperfect,
273
843064
2023
14:20
and we extract from them an inferred task trajectory
274
845087
3091
14:24
and control sequence for the robot.
275
848178
3003
14:27
We then execute that on the robot,
276
851181
2184
14:29
we observe what happens,
277
853365
2172
14:31
then we adjust the controls using a sequence of techniques
278
855537
2662
14:34
called iterative learning.
279
858199
2930
14:37
Then what we do is, we increase the velocity a little bit.
280
861129
3977
14:41
We observe the results, adjust the controls again,
281
865106
3563
14:44
and observe what happens.
282
868669
2522
14:47
And we go through this several rounds.
283
871191
2136
14:49
And here's the result.
284
873327
1356
14:50
That's the inferred task trajectory,
285
874683
1845
14:52
and here's the robot moving at the speed of the human.
286
876528
3463
14:55
Here's four times the speed of the human.
287
879991
2442
14:58
Here's seven times.
288
882433
2571
15:00
And here's the robot operating at 10 times
289
885004
3637
15:04
the speed of the human.
290
888641
2200
15:06
So we're able to get a robot to perform a delicate task,
291
890841
2950
15:09
like a surgical subtask,
292
893791
3224
15:12
at 10 times the speed of a human.
293
897015
3247
15:16
So this project also, because of its involved practicing
294
900262
4223
15:20
and learning, doing something over and over again,
295
904485
2173
15:22
this project also has a lesson, which is,
296
906658
2757
15:25
if you want to do something well,
297
909415
3126
15:28
there's no substitute for practice, practice, practice.
298
912541
7948
15:36
So these are four of the lessons that I've learned
299
920505
3120
15:39
from robots over the years,
300
923625
3127
15:42
and robotics, the field of robotics has gotten much better
301
926752
5369
15:48
over time.
302
932121
2168
15:50
Nowadays, high school students can build robots
303
934289
2167
15:52
like the industrial robot my dad and I tried to build.
304
936456
4025
15:56
And now, I have a daughter,
305
940481
6982
16:03
named Odessa.
306
947463
2386
16:05
She's eight years old,
307
949849
1833
16:07
and she likes robots, too.
308
951682
2424
16:10
Maybe it runs in the family. (Laughter)
309
954106
2414
16:12
I wish she could meet my dad.
310
956520
3647
16:16
And now I get to teach her how things work,
311
960167
2900
16:18
and we get to build projects together, and I wonder
312
963067
2772
16:21
what kind of lessons that she'll learn from them.
313
965839
4308
16:26
Robots are the most human
314
970147
2607
16:28
of our machines.
315
972754
2229
16:30
They can't solve all of the world's problems,
316
974983
3085
16:33
but I think they have something important to teach us.
317
978068
4358
16:38
I invite all of you to think about the innovations
318
982426
3343
16:41
that you're interested in,
319
985769
2675
16:44
the machines that you wish for,
320
988444
3216
16:47
and think about what they might be telling you,
321
991660
3169
16:50
because I have a hunch
322
994829
2063
16:52
that many of our technological innovations,
323
996892
2087
16:54
the devices we dream about,
324
998979
2658
16:57
can inspire us to be better humans.
325
1001637
3964
17:01
Thank you. (Applause)
326
1005601
3585
ABOUT THE SPEAKER
Ken Goldberg - RoboticistKen Goldberg works reflect the intersection of robotics, social media, and art.
Why you should listen
Ken Goldberg is a Professor of Industrial Engineering and Operations Research in Robotics, Automation, and New Media at UC Berkeley and holds a position at UC San Francisco Medical School where he researches medical applications for robotics. Born in Nigeria and raised in Bethlehem, Pennsylvania, Ken hold degrees in Electrical Engineering and Economics from the University of Pennsylvania and received his Ph.D. in Computer Science from Carnegie Mellon University. He is widely recognized as an engineer, a teacher, and an artist – receiving the Joseph F. Engelberger Robotics Award in 2000, the IEEE Major Educational Innovation Award in 2001, and Isadora Duncan Award in 2006 for his Ballet Mori project, performed by the San Francisco Ballet. His works have been exhibited at the Whitney Biennial in New York City, the Pompidou Centre in Paris, and the Ars Electronica in Linz. His book, The Robot in the Garden, was published in March of 2000 by the MIT Press.
Ken Goldberg | Speaker | TED.com