TEDGlobal 2012
Shyam Sankar: The rise of human-computer cooperation
Filmed:
Readability: 6.3
947,418 views
Brute computing force alone can't solve the world's problems. Data mining innovator Shyam Sankar explains why solving big problems (like catching terrorists or identifying huge hidden trends) is not a question of finding the right algorithm, but rather the right symbiotic relationship between computation and human creativity.
Shyam Sankar - Data Intelligence Agent
An advocate of human-computer symbiosis, Shyam Sankar looks for clues in big and disparate data sets. Full bio
An advocate of human-computer symbiosis, Shyam Sankar looks for clues in big and disparate data sets. Full bio
Double-click the English transcript below to play the video.
00:16
I'd like to tell you about two games of chess.
0
512
2556
00:18
The first happened in 1997, in which Garry Kasparov,
1
3068
3864
00:22
a human, lost to Deep Blue, a machine.
2
6932
3716
00:26
To many, this was the dawn of a new era,
3
10648
2240
00:28
one where man would be dominated by machine.
4
12888
2779
00:31
But here we are, 20 years on, and the greatest change
5
15667
3334
00:34
in how we relate to computers is the iPad,
6
19001
2690
00:37
not HAL.
7
21691
2045
00:39
The second game was a freestyle chess tournament
8
23736
2648
00:42
in 2005, in which man and machine could enter together
9
26384
2969
00:45
as partners, rather than adversaries, if they so chose.
10
29353
4666
00:49
At first, the results were predictable.
11
34019
1851
00:51
Even a supercomputer was beaten by a grandmaster
12
35870
2497
00:54
with a relatively weak laptop.
13
38367
2312
00:56
The surprise came at the end. Who won?
14
40679
2985
00:59
Not a grandmaster with a supercomputer,
15
43664
2776
01:02
but actually two American amateurs
16
46440
1493
01:03
using three relatively weak laptops.
17
47933
3822
01:07
Their ability to coach and manipulate their computers
18
51755
2596
01:10
to deeply explore specific positions
19
54351
2435
01:12
effectively counteracted the superior chess knowledge
20
56786
2390
01:15
of the grandmasters and the superior computational power
21
59176
2609
01:17
of other adversaries.
22
61785
1909
01:19
This is an astonishing result: average men,
23
63694
2905
01:22
average machines beating the best man, the best machine.
24
66599
4081
01:26
And anyways, isn't it supposed to be man versus machine?
25
70680
3199
01:29
Instead, it's about cooperation, and the right type of cooperation.
26
73879
4152
01:33
We've been paying a lot of attention to Marvin Minsky's
27
78031
2857
01:36
vision for artificial intelligence over the last 50 years.
28
80888
3242
01:40
It's a sexy vision, for sure. Many have embraced it.
29
84130
2262
01:42
It's become the dominant school of thought in computer science.
30
86392
2753
01:45
But as we enter the era of big data, of network systems,
31
89145
3072
01:48
of open platforms, and embedded technology,
32
92217
2698
01:50
I'd like to suggest it's time to reevaluate an alternative vision
33
94915
3392
01:54
that was actually developed around the same time.
34
98307
3070
01:57
I'm talking about J.C.R. Licklider's human-computer symbiosis,
35
101377
3332
02:00
perhaps better termed "intelligence augmentation," I.A.
36
104709
3808
02:04
Licklider was a computer science titan who had a profound
37
108517
2640
02:07
effect on the development of technology and the Internet.
38
111157
3006
02:10
His vision was to enable man and machine to cooperate
39
114163
2868
02:12
in making decisions, controlling complex situations
40
117031
3590
02:16
without the inflexible dependence
41
120621
1770
02:18
on predetermined programs.
42
122391
2533
02:20
Note that word "cooperate."
43
124924
2498
02:23
Licklider encourages us not to take a toaster
44
127422
2747
02:26
and make it Data from "Star Trek,"
45
130169
2284
02:28
but to take a human and make her more capable.
46
132453
3535
02:31
Humans are so amazing -- how we think,
47
135988
1911
02:33
our non-linear approaches, our creativity,
48
137899
2618
02:36
iterative hypotheses, all very difficult if possible at all
49
140517
2131
02:38
for computers to do.
50
142648
1345
02:39
Licklider intuitively realized this, contemplating humans
51
143993
2452
02:42
setting the goals, formulating the hypotheses,
52
146445
2327
02:44
determining the criteria, and performing the evaluation.
53
148772
3376
02:48
Of course, in other ways, humans are so limited.
54
152148
1775
02:49
We're terrible at scale, computation and volume.
55
153923
3235
02:53
We require high-end talent management
56
157158
1836
02:54
to keep the rock band together and playing.
57
158994
2064
02:56
Licklider foresaw computers doing all the routinizable work
58
161058
2204
02:59
that was required to prepare the way for insights and decision making.
59
163262
3276
03:02
Silently, without much fanfare,
60
166538
2224
03:04
this approach has been compiling victories beyond chess.
61
168762
3354
03:08
Protein folding, a topic that shares the incredible expansiveness of chess —
62
172116
3356
03:11
there are more ways of folding a protein than there are atoms in the universe.
63
175472
3042
03:14
This is a world-changing problem with huge implications
64
178514
2353
03:16
for our ability to understand and treat disease.
65
180867
2308
03:19
And for this task, supercomputer field brute force simply isn't enough.
66
183175
4248
03:23
Foldit, a game created by computer scientists,
67
187423
2384
03:25
illustrates the value of the approach.
68
189807
2502
03:28
Non-technical, non-biologist amateurs play a video game
69
192309
3041
03:31
in which they visually rearrange the structure of the protein,
70
195350
3073
03:34
allowing the computer to manage the atomic forces
71
198423
1499
03:35
and interactions and identify structural issues.
72
199922
2957
03:38
This approach beat supercomputers 50 percent of the time
73
202879
3023
03:41
and tied 30 percent of the time.
74
205902
2584
03:44
Foldit recently made a notable and major scientific discovery
75
208486
3137
03:47
by deciphering the structure of the Mason-Pfizer monkey virus.
76
211623
3160
03:50
A protease that had eluded determination for over 10 years
77
214783
3015
03:53
was solved was by three players in a matter of days,
78
217798
2626
03:56
perhaps the first major scientific advance
79
220424
2025
03:58
to come from playing a video game.
80
222449
2323
04:00
Last year, on the site of the Twin Towers,
81
224772
2181
04:02
the 9/11 memorial opened.
82
226953
1473
04:04
It displays the names of the thousands of victims
83
228426
2721
04:07
using a beautiful concept called "meaningful adjacency."
84
231147
3063
04:10
It places the names next to each other based on their
85
234210
2166
04:12
relationships to one another: friends, families, coworkers.
86
236376
2213
04:14
When you put it all together, it's quite a computational
87
238589
3028
04:17
challenge: 3,500 victims, 1,800 adjacency requests,
88
241617
4223
04:21
the importance of the overall physical specifications
89
245840
3092
04:24
and the final aesthetics.
90
248932
2137
04:26
When first reported by the media, full credit for such a feat
91
251069
2615
04:29
was given to an algorithm from the New York City
92
253684
1892
04:31
design firm Local Projects. The truth is a bit more nuanced.
93
255576
4001
04:35
While an algorithm was used to develop the underlying framework,
94
259577
2871
04:38
humans used that framework to design the final result.
95
262448
3008
04:41
So in this case, a computer had evaluated millions
96
265456
2225
04:43
of possible layouts, managed a complex relational system,
97
267681
3335
04:46
and kept track of a very large set of measurements
98
271016
2414
04:49
and variables, allowing the humans to focus
99
273430
2410
04:51
on design and compositional choices.
100
275840
2802
04:54
So the more you look around you,
101
278642
1036
04:55
the more you see Licklider's vision everywhere.
102
279678
1962
04:57
Whether it's augmented reality in your iPhone or GPS in your car,
103
281640
3304
05:00
human-computer symbiosis is making us more capable.
104
284944
2970
05:03
So if you want to improve human-computer symbiosis,
105
287914
1655
05:05
what can you do?
106
289569
1429
05:06
You can start by designing the human into the process.
107
290998
2452
05:09
Instead of thinking about what a computer will do to solve the problem,
108
293450
2204
05:11
design the solution around what the human will do as well.
109
295654
3869
05:15
When you do this, you'll quickly realize that you spent
110
299523
1937
05:17
all of your time on the interface between man and machine,
111
301460
2879
05:20
specifically on designing away the friction in the interaction.
112
304339
3099
05:23
In fact, this friction is more important than the power
113
307438
2766
05:26
of the man or the power of the machine
114
310204
2052
05:28
in determining overall capability.
115
312256
1931
05:30
That's why two amateurs with a few laptops
116
314187
1977
05:32
handily beat a supercomputer and a grandmaster.
117
316164
2456
05:34
What Kasparov calls process is a byproduct of friction.
118
318620
3005
05:37
The better the process, the less the friction.
119
321625
2401
05:39
And minimizing friction turns out to be the decisive variable.
120
324026
4256
05:44
Or take another example: big data.
121
328282
2243
05:46
Every interaction we have in the world is recorded
122
330525
1906
05:48
by an ever growing array of sensors: your phone,
123
332431
3059
05:51
your credit card, your computer. The result is big data,
124
335490
2373
05:53
and it actually presents us with an opportunity
125
337863
1742
05:55
to more deeply understand the human condition.
126
339605
2662
05:58
The major emphasis of most approaches to big data
127
342267
2305
06:00
focus on, "How do I store this data? How do I search
128
344572
2215
06:02
this data? How do I process this data?"
129
346787
2276
06:04
These are necessary but insufficient questions.
130
349063
2204
06:07
The imperative is not to figure out how to compute,
131
351267
2471
06:09
but what to compute. How do you impose human intuition
132
353738
2184
06:11
on data at this scale?
133
355922
1791
06:13
Again, we start by designing the human into the process.
134
357713
3499
06:17
When PayPal was first starting as a business, their biggest
135
361212
2812
06:19
challenge was not, "How do I send money back and forth online?"
136
364024
2804
06:22
It was, "How do I do that without being defrauded by organized crime?"
137
366828
3872
06:26
Why so challenging? Because while computers can learn
138
370700
2088
06:28
to detect and identify fraud based on patterns,
139
372788
3144
06:31
they can't learn to do that based on patterns
140
375932
1479
06:33
they've never seen before, and organized crime
141
377411
2116
06:35
has a lot in common with this audience: brilliant people,
142
379527
2709
06:38
relentlessly resourceful, entrepreneurial spirit — (Laughter) —
143
382236
3640
06:41
and one huge and important difference: purpose.
144
385876
2712
06:44
And so while computers alone can catch all but the cleverest
145
388588
2832
06:47
fraudsters, catching the cleverest is the difference
146
391420
2253
06:49
between success and failure.
147
393673
2545
06:52
There's a whole class of problems like this, ones with
148
396218
2221
06:54
adaptive adversaries. They rarely if ever present with a
149
398439
2575
06:56
repeatable pattern that's discernable to computers.
150
401014
2736
06:59
Instead, there's some inherent component of innovation or disruption,
151
403750
3993
07:03
and increasingly these problems are buried in big data.
152
407743
2735
07:06
For example, terrorism. Terrorists are always adapting
153
410478
2500
07:08
in minor and major ways to new circumstances, and despite
154
412978
2052
07:10
what you might see on TV, these adaptations,
155
415030
3094
07:14
and the detection of them, are fundamentally human.
156
418124
2293
07:16
Computers don't detect novel patterns and new behaviors,
157
420417
3117
07:19
but humans do. Humans, using technology, testing hypotheses,
158
423534
3235
07:22
searching for insight by asking machines to do things for them.
159
426769
4620
07:27
Osama bin Laden was not caught by artificial intelligence.
160
431389
2320
07:29
He was caught by dedicated, resourceful, brilliant people
161
433709
2553
07:32
in partnerships with various technologies.
162
436262
4269
07:36
As appealing as it might sound, you cannot algorithmically
163
440531
2818
07:39
data mine your way to the answer.
164
443349
1601
07:40
There is no "Find Terrorist" button, and the more data
165
444950
2855
07:43
we integrate from a vast variety of sources
166
447805
2302
07:46
across a wide variety of data formats from very
167
450107
2133
07:48
disparate systems, the less effective data mining can be.
168
452240
3309
07:51
Instead, people will have to look at data
169
455549
2024
07:53
and search for insight, and as Licklider foresaw long ago,
170
457573
3456
07:56
the key to great results here is the right type of cooperation,
171
461029
2685
07:59
and as Kasparov realized,
172
463714
1524
08:01
that means minimizing friction at the interface.
173
465238
3031
08:04
Now this approach makes possible things like combing
174
468269
2758
08:06
through all available data from very different sources,
175
471027
3386
08:10
identifying key relationships and putting them in one place,
176
474413
2792
08:13
something that's been nearly impossible to do before.
177
477205
2928
08:16
To some, this has terrifying privacy and civil liberties
178
480133
1942
08:17
implications. To others it foretells of an era of greater
179
482075
3410
08:21
privacy and civil liberties protections,
180
485485
1909
08:23
but privacy and civil liberties are of fundamental importance.
181
487394
2936
08:26
That must be acknowledged, and they can't be swept aside,
182
490330
2193
08:28
even with the best of intents.
183
492523
2530
08:30
So let's explore, through a couple of examples, the impact
184
495053
2518
08:33
that technologies built to drive human-computer symbiosis
185
497571
2406
08:35
have had in recent time.
186
499977
2919
08:38
In October, 2007, U.S. and coalition forces raided
187
502896
3416
08:42
an al Qaeda safe house in the city of Sinjar
188
506312
2416
08:44
on the Syrian border of Iraq.
189
508728
1934
08:46
They found a treasure trove of documents:
190
510662
2376
08:48
700 biographical sketches of foreign fighters.
191
513038
2335
08:51
These foreign fighters had left their families in the Gulf,
192
515373
2584
08:53
the Levant and North Africa to join al Qaeda in Iraq.
193
517957
3146
08:57
These records were human resource forms.
194
521103
1616
08:58
The foreign fighters filled them out as they joined the organization.
195
522719
2855
09:01
It turns out that al Qaeda, too,
196
525574
1211
09:02
is not without its bureaucracy. (Laughter)
197
526785
2597
09:05
They answered questions like, "Who recruited you?"
198
529382
2098
09:07
"What's your hometown?" "What occupation do you seek?"
199
531480
2854
09:10
In that last question, a surprising insight was revealed.
200
534334
3169
09:13
The vast majority of foreign fighters
201
537503
2400
09:15
were seeking to become suicide bombers for martyrdom --
202
539903
2400
09:18
hugely important, since between 2003 and 2007, Iraq
203
542303
4338
09:22
had 1,382 suicide bombings, a major source of instability.
204
546641
4244
09:26
Analyzing this data was hard. The originals were sheets
205
550885
2058
09:28
of paper in Arabic that had to be scanned and translated.
206
552943
2742
09:31
The friction in the process did not allow for meaningful
207
555685
2192
09:33
results in an operational time frame using humans, PDFs
208
557877
3350
09:37
and tenacity alone.
209
561227
2218
09:39
The researchers had to lever up their human minds
210
563445
1953
09:41
with technology to dive deeper, to explore non-obvious
211
565398
2345
09:43
hypotheses, and in fact, insights emerged.
212
567743
3218
09:46
Twenty percent of the foreign fighters were from Libya,
213
570961
2644
09:49
50 percent of those from a single town in Libya,
214
573605
2968
09:52
hugely important since prior statistics put that figure at
215
576573
2450
09:54
three percent. It also helped to hone in on a figure
216
579023
2383
09:57
of rising importance in al Qaeda, Abu Yahya al-Libi,
217
581406
2977
10:00
a senior cleric in the Libyan Islamic fighting group.
218
584383
2631
10:02
In March of 2007, he gave a speech, after which there was
219
587014
2664
10:05
a surge in participation amongst Libyan foreign fighters.
220
589678
3466
10:09
Perhaps most clever of all, though, and least obvious,
221
593144
3106
10:12
by flipping the data on its head, the researchers were
222
596250
2073
10:14
able to deeply explore the coordination networks in Syria
223
598323
2900
10:17
that were ultimately responsible for receiving and
224
601223
2517
10:19
transporting the foreign fighters to the border.
225
603740
2464
10:22
These were networks of mercenaries, not ideologues,
226
606204
2633
10:24
who were in the coordination business for profit.
227
608837
2398
10:27
For example, they charged Saudi foreign fighters
228
611235
1904
10:29
substantially more than Libyans, money that would have
229
613139
2199
10:31
otherwise gone to al Qaeda.
230
615338
2320
10:33
Perhaps the adversary would disrupt their own network
231
617658
2045
10:35
if they knew they cheating would-be jihadists.
232
619703
3035
10:38
In January, 2010, a devastating 7.0 earthquake struck Haiti,
233
622738
3745
10:42
third deadliest earthquake of all time, left one million people,
234
626483
2916
10:45
10 percent of the population, homeless.
235
629399
2584
10:47
One seemingly small aspect of the overall relief effort
236
631983
3137
10:51
became increasingly important as the delivery of food
237
635120
2176
10:53
and water started rolling.
238
637296
2160
10:55
January and February are the dry months in Haiti,
239
639456
1458
10:56
yet many of the camps had developed standing water.
240
640914
2942
10:59
The only institution with detailed knowledge of Haiti's
241
643856
2122
11:01
floodplains had been leveled
242
645978
1297
11:03
in the earthquake, leadership inside.
243
647275
3008
11:06
So the question is, which camps are at risk,
244
650283
2575
11:08
how many people are in these camps, what's the
245
652858
1921
11:10
timeline for flooding, and given very limited resources
246
654779
2311
11:12
and infrastructure, how do we prioritize the relocation?
247
657090
3384
11:16
The data was incredibly disparate. The U.S. Army had
248
660474
2344
11:18
detailed knowledge for only a small section of the country.
249
662818
2929
11:21
There was data online from a 2006 environmental risk
250
665747
2511
11:24
conference, other geospatial data, none of it integrated.
251
668258
2664
11:26
The human goal here was to identify camps for relocation
252
670922
2958
11:29
based on priority need.
253
673880
2395
11:32
The computer had to integrate a vast amount of geospacial
254
676275
2440
11:34
information, social media data and relief organization
255
678715
2584
11:37
information to answer this question.
256
681299
3480
11:40
By implementing a superior process, what was otherwise
257
684779
2415
11:43
a task for 40 people over three months became
258
687194
2608
11:45
a simple job for three people in 40 hours,
259
689802
3176
11:48
all victories for human-computer symbiosis.
260
692978
2628
11:51
We're more than 50 years into Licklider's vision
261
695606
2054
11:53
for the future, and the data suggests that we should be
262
697660
2242
11:55
quite excited about tackling this century's hardest problems,
263
699902
3030
11:58
man and machine in cooperation together.
264
702932
2947
12:01
Thank you. (Applause)
265
705879
2197
12:03
(Applause)
266
708076
2505
ABOUT THE SPEAKER
Shyam Sankar - Data Intelligence AgentAn advocate of human-computer symbiosis, Shyam Sankar looks for clues in big and disparate data sets.
Why you should listen
Shyam Sankar is a Director at Palantir Technologies, a secretive Silicon Valley company where he oversees deployments of the company's core technology, which helps law enforcement teams and corporations analyze giant, unrelated databases for clues to potential ... anything. Palantir technologies has been used to find missing children, to detect banking fraud, and to uncover the Shadow Network, a cyber-spy ring that stooped so low as to hack the Dalai Lama's email.
As part of his work, Sankar thinks deeply about the place where human and machine intelligence meet. While artificial intelligence (AI) is the dominant paradigm, he is an advocate of JCR Licklider's "intelligence augmentation" (IA) approach, where algorithms and brains work together to solve problems.
Shyam Sankar | Speaker | TED.com