1
00:00:01,102 --> 00:00:06,180
We have historical records that allow us
to know how the ancient Greeks dressed,

2
00:00:06,180 --> 00:00:07,458
how they lived,

3
00:00:07,458 --> 00:00:09,159
how they fought ...

4
00:00:09,159 --> 00:00:11,211
but how did they think?

5
00:00:11,535 --> 00:00:16,094
One natural idea is that the deepest
aspects of human thought --

6
00:00:16,094 --> 00:00:17,976
our ability to imagine,

7
00:00:17,976 --> 00:00:19,306
to be conscious,

8
00:00:19,306 --> 00:00:20,681
to dream --

9
00:00:20,681 --> 00:00:22,701
have always been the same.

10
00:00:23,092 --> 00:00:24,601
Another possibility

11
00:00:24,601 --> 00:00:28,351
is that the social transformations
that have shaped our culture

12
00:00:28,351 --> 00:00:32,428
may have also changed the structural
columns of human thought.

13
00:00:33,096 --> 00:00:35,627
We may all have different
opinions about this.

14
00:00:35,627 --> 00:00:38,344
Actually, it's a long-standing
philosophical debate.

15
00:00:38,724 --> 00:00:42,335
But is this question
even amenable to science?

16
00:00:43,116 --> 00:00:45,555
Here I'd like to propose

17
00:00:45,555 --> 00:00:50,160
that in the same way we can reconstruct
how the ancient Greek cities looked like

18
00:00:50,160 --> 00:00:52,777
just based on a few bricks,

19
00:00:52,777 --> 00:00:56,835
that the writings of a culture
are the archeological records --

20
00:00:56,835 --> 00:00:59,467
the fossils of human thought.

21
00:01:00,199 --> 00:01:01,390
And in fact,

22
00:01:01,390 --> 00:01:03,570
doing some form of psychological analysis

23
00:01:03,570 --> 00:01:07,184
of some of the most ancient
books of human culture,

24
00:01:07,184 --> 00:01:13,028
Julian Jaynes came in the '70s with
a very wild and radical hypothesis ...

25
00:01:13,028 --> 00:01:15,444
that only 3,000 years ago,

26
00:01:15,444 --> 00:01:20,857
humans were what today
we would call schizophrenics.

27
00:01:21,883 --> 00:01:23,575
And he made this claim

28
00:01:23,575 --> 00:01:26,752
based on the fact that the first
humans [inscribing] these books

29
00:01:26,752 --> 00:01:28,792
behaved consistently,


30
00:01:28,792 --> 00:01:31,708
in different traditions
and in different places of the world,

31
00:01:31,708 --> 00:01:35,444
as if they were hearing and obeying voices

32
00:01:35,444 --> 00:01:38,394
that they perceived
as coming from the Gods,

33
00:01:38,394 --> 00:01:40,413
or from the muses --

34
00:01:40,413 --> 00:01:43,893
what today we would call hallucinations.

35
00:01:44,102 --> 00:01:45,476
And only then,

36
00:01:45,476 --> 00:01:46,780
as time went on,

37
00:01:46,780 --> 00:01:50,361
they began to recognize
that they were the creators --

38
00:01:50,361 --> 00:01:53,078
the owners of these inner voices.

39
00:01:53,510 --> 00:01:56,273
And with this they gained introspection:

40
00:01:56,273 --> 00:01:59,221
the ability to think
about their own thoughts.

41
00:01:59,965 --> 00:02:03,378
So Jaynes' theory is that consciousness --

42
00:02:03,378 --> 00:02:06,538
at least in the way we perceive it today,

43
00:02:06,538 --> 00:02:10,225
where we feel that we are the pilots
of our own existence --

44
00:02:10,225 --> 00:02:13,164
is a quite recent cultural development.

45
00:02:13,545 --> 00:02:15,337
And this theory is quite spectacular,

46
00:02:15,337 --> 00:02:16,769
but it has an obvious problem

47
00:02:16,769 --> 00:02:20,761
which is that it's built on just a few
and very specific examples.

48
00:02:21,085 --> 00:02:22,872
So the question is whether the theory

49
00:02:22,872 --> 00:02:27,913
that introspection built up in human
history only about 3,000 years ago

50
00:02:27,913 --> 00:02:31,283
can be examined in a quantitative
and objective manner.

51
00:02:31,779 --> 00:02:35,338
And the problem on how
to go about this is quite obvious.

52
00:02:35,338 --> 00:02:37,724
It's not like Plato woke up one day

53
00:02:37,724 --> 00:02:38,910
and then he wrote,

54
00:02:38,910 --> 00:02:40,596
"Hello, I'm Plato

55
00:02:40,596 --> 00:02:43,487
and as of today I have a fully
introspective consciousness."

56
00:02:43,487 --> 00:02:45,307
(Laughter)

57
00:02:45,637 --> 00:02:49,097
And this still is actually
what is the essence of the problem.

58
00:02:49,624 --> 00:02:54,071
We need to find the emergence
of a concept that's never said.

59
00:02:54,680 --> 00:02:58,942
The word introspection
does not appear a single time

60
00:02:58,942 --> 00:03:01,559
in the books we want to analyze.

61
00:03:01,971 --> 00:03:06,398
So our way to solve this
is to build the space of words.

62
00:03:06,794 --> 00:03:09,998
This is a huge space
that contains all words

63
00:03:09,998 --> 00:03:12,959
in such a way that they distance
between any two of them

64
00:03:12,959 --> 00:03:15,842
is indicative of how
closely related they are.

65
00:03:16,460 --> 00:03:17,456
So for instance,

66
00:03:17,456 --> 00:03:20,857
you want the words dog and cat
to be very close together,

67
00:03:20,857 --> 00:03:24,688
but the words grapefruit and logarithm
to be very far away.

68
00:03:25,008 --> 00:03:29,298
And this has to be true for any
two words within the space.

69
00:03:29,748 --> 00:03:33,109
And there are different ways that we
can construct the space of words.

70
00:03:33,109 --> 00:03:34,802
One is just asking the experts,

71
00:03:34,802 --> 00:03:37,199
a bit like we do with dictionaries.

72
00:03:37,199 --> 00:03:38,623
Another possibility

73
00:03:38,623 --> 00:03:40,688
is following the simple assumption

74
00:03:40,688 --> 00:03:44,640
that when two words are related
they tend to appear in the same sentences,

75
00:03:44,640 --> 00:03:46,214
in the same paragraphs,

76
00:03:46,214 --> 00:03:48,049
in the same documents,

77
00:03:48,049 --> 00:03:51,509
more often than would be expected
just by pure chance.

78
00:03:52,448 --> 00:03:54,305
And this simple hypothesis,

79
00:03:54,305 --> 00:03:55,792
this simple method,

80
00:03:55,792 --> 00:03:57,266
with some computational tricks

81
00:03:57,266 --> 00:03:58,679
that have to do with the fact

82
00:03:58,679 --> 00:04:01,995
that this is a very complex
and highly dimensional space,

83
00:04:01,995 --> 00:04:04,458
turns out to be quite effective.

84
00:04:04,458 --> 00:04:07,127
And just to give you a flavor
of how well this works,

85
00:04:07,127 --> 00:04:11,321
this is the result we get when
we analyze this for some familiar words.

86
00:04:11,607 --> 00:04:12,816
And you can see first

87
00:04:12,816 --> 00:04:16,278
that words automatically organize
into semantic neighborhoods.

88
00:04:16,278 --> 00:04:17,362
So you get the fruits,

89
00:04:17,362 --> 00:04:18,359
the body parts,

90
00:04:18,359 --> 00:04:19,360
the computer parts,

91
00:04:19,360 --> 00:04:20,359
the scientific terms

92
00:04:20,359 --> 00:04:21,357
and so on.

93
00:04:21,357 --> 00:04:25,699
The algorithm also identifies
the reorganized concepts in a hierarchy.

94
00:04:26,027 --> 00:04:27,027
So for instance,

95
00:04:27,027 --> 00:04:30,648
you can see that the scientific terms
break down into two subcategories

96
00:04:30,648 --> 00:04:33,608
of the astronomic and the physic terms.

97
00:04:33,608 --> 00:04:35,881
And then there are very fine things.

98
00:04:35,881 --> 00:04:36,878
For instance,

99
00:04:36,878 --> 00:04:38,054
the word astronomy,

100
00:04:38,056 --> 00:04:39,870
which seems a bit bizarre where it is,

101
00:04:39,870 --> 00:04:41,768
is actually exactly where it should be,

102
00:04:41,768 --> 00:04:43,219
between what it is --

103
00:04:43,219 --> 00:04:44,630
an actual science --

104
00:04:44,630 --> 00:04:46,165
and between what it describes --

105
00:04:46,165 --> 00:04:47,912
the astronomical terms.

106
00:04:48,366 --> 00:04:50,097
And we could go on and on with this.

107
00:04:50,097 --> 00:04:52,181
Actually if you stare at this for a while

108
00:04:52,181 --> 00:04:54,034
and you just build random trajectories,

109
00:04:54,034 --> 00:04:55,725
you will see that is feels well --

110
00:04:55,725 --> 00:04:58,286
actually it feels a bit like doing poetry.

111
00:04:58,286 --> 00:04:59,286
And this is because,

112
00:04:59,286 --> 00:05:00,288
in a way,

113
00:05:00,288 --> 00:05:03,482
walking in this space
is like walking in the mind.

114
00:05:03,901 --> 00:05:05,778
And the last thing

115
00:05:05,778 --> 00:05:09,938
is that this algorithm identifies
what are our intuitions

116
00:05:09,938 --> 00:05:14,002
of which words should lead
in the neighborhood of introspection.

117
00:05:14,002 --> 00:05:15,072
So for instance,

118
00:05:15,072 --> 00:05:18,982
words such as self, guilt,
reason, emotion,

119
00:05:18,982 --> 00:05:21,102
are very close to introspection,

120
00:05:21,102 --> 00:05:22,102
but other words,

121
00:05:22,102 --> 00:05:24,432
such as red, football, candle, banana,

122
00:05:24,432 --> 00:05:26,072
are just very far away.

123
00:05:26,262 --> 00:05:28,882
And so once we've built the space,

124
00:05:28,882 --> 00:05:31,945
the question of the history
of introspection,

125
00:05:31,945 --> 00:05:34,277
or of the history of any concept

126
00:05:34,277 --> 00:05:39,055
which before could seem abstract
and somehow vague,

127
00:05:39,055 --> 00:05:40,754
becomes concrete --

128
00:05:40,754 --> 00:05:43,656
becomes amenable to quantitative science.

129
00:05:44,481 --> 00:05:47,137
All that we have to do is take the books,

130
00:05:47,137 --> 00:05:48,612
we digitize them

131
00:05:48,612 --> 00:05:51,420
and we take this stream
of words as a trajectory

132
00:05:51,420 --> 00:05:53,452
and project them into the space,

133
00:05:53,452 --> 00:05:57,087
and then we ask whether this trajectory
spends significant time

134
00:05:57,087 --> 00:06:00,361
circling closely to the concept
of introspection.

135
00:06:00,911 --> 00:06:02,152
And with this,

136
00:06:02,152 --> 00:06:04,263
we could analyze
the history of introspection

137
00:06:04,263 --> 00:06:06,183
in the ancient Greek tradition,

138
00:06:06,183 --> 00:06:09,304
for which we have the best
available written record.

139
00:06:09,711 --> 00:06:12,131
So what we did is we took all the books --

140
00:06:12,131 --> 00:06:14,517
we just ordered them by time --

141
00:06:14,517 --> 00:06:15,994
for each book we take the words

142
00:06:15,994 --> 00:06:18,194
and we project them to the space,

143
00:06:18,194 --> 00:06:21,165
and then we ask for each word
how close it is to introspection,

144
00:06:21,165 --> 00:06:22,728
and we just average that.

145
00:06:22,728 --> 00:06:25,986
And then we understand
that as time goes on and on,

146
00:06:25,986 --> 00:06:29,088
these books get closer,
and closer and closer

147
00:06:29,088 --> 00:06:31,062
to the concept of introspection.

148
00:06:31,062 --> 00:06:35,387
And this is exactly what happens
in the ancient Greek tradition.

149
00:06:35,968 --> 00:06:39,146
So you can see that for the oldest books
in the Homeric tradition,

150
00:06:39,146 --> 00:06:42,540
there is a small increase with books
getting closer to introspection,

151
00:06:42,540 --> 00:06:44,816
but about four centuries before Christ,

152
00:06:44,816 --> 00:06:49,376
this starts ramping-up very rapidly
to an almost five-fold increase

153
00:06:49,376 --> 00:06:51,938
of books getting closer,
and closer and closer

154
00:06:51,938 --> 00:06:54,319
to the concept of introspection.

155
00:06:54,319 --> 00:06:56,679
And one of the nice things about this

156
00:06:56,679 --> 00:06:57,829
is that now we can ask

157
00:06:57,829 --> 00:07:02,498
whether this is also true
in a different independent tradition.

158
00:07:03,026 --> 00:07:06,248
So we just ran this same analysis
on the Judeo-Christian tradition,

159
00:07:06,248 --> 00:07:09,178
and we got virtually the same pattern.

160
00:07:09,741 --> 00:07:14,445
Again you see a small increase
for the oldest books in the Old Testament,

161
00:07:14,445 --> 00:07:16,326
and then it increases much more rapidly

162
00:07:16,326 --> 00:07:18,169
in the new books of the New Testament,

163
00:07:18,169 --> 00:07:20,518
and then we get the peak of introspection

164
00:07:20,518 --> 00:07:22,521
in the work Confessions
of Saint Augustine,

165
00:07:22,521 --> 00:07:24,721
about four centuries after Christ.

166
00:07:25,051 --> 00:07:26,985
And this was very important,

167
00:07:26,985 --> 00:07:30,351
because Saint Augustine
had been recognized by scholars --

168
00:07:30,351 --> 00:07:31,481
philologists,

169
00:07:31,481 --> 00:07:32,632
historians --

170
00:07:32,632 --> 00:07:35,298
as one of the founders of introspection.

171
00:07:35,298 --> 00:07:38,595
Actually, some believe him to be
the father of modern psychology.

172
00:07:39,155 --> 00:07:41,157
So our algorithm,

173
00:07:41,157 --> 00:07:43,804
which has the virtue
of being quantitative,

174
00:07:43,804 --> 00:07:45,066
of being objective,

175
00:07:45,066 --> 00:07:47,332
and of course of being extremely fast --

176
00:07:47,332 --> 00:07:49,636
it just runs in a fraction of a second --

177
00:07:49,636 --> 00:07:53,138
can capture some of the most
important conclusions

178
00:07:53,138 --> 00:07:55,940
of this long tradition of investigation.

179
00:07:56,547 --> 00:08:00,142
And this is in a way
one of the beauties of science,

180
00:08:00,142 --> 00:08:03,617
which is that now this idea
can be translated

181
00:08:03,617 --> 00:08:06,496
and generalized to a whole
lot of different domains.

182
00:08:06,920 --> 00:08:11,807
So in the same way that we asked
about the past of human conciousness,

183
00:08:11,807 --> 00:08:15,132
maybe the most challenging question
we can pose to ourselves,

184
00:08:15,132 --> 00:08:19,269
is whether this can tell us something
about the future of our own consciousness.

185
00:08:19,717 --> 00:08:21,186
To put it more precisely,

186
00:08:21,186 --> 00:08:23,601
whether the words we say today

187
00:08:23,601 --> 00:08:28,797
can tell us something of where
our minds will be in a few days,

188
00:08:28,797 --> 00:08:29,962
in a few months,

189
00:08:29,962 --> 00:08:31,907
or a few years from now.

190
00:08:31,907 --> 00:08:34,813
And in the same way many of us
are now wearing censors

191
00:08:34,813 --> 00:08:36,524
that detect our heart rate,

192
00:08:36,524 --> 00:08:37,936
our respiration,

193
00:08:37,936 --> 00:08:39,712
our genes,

194
00:08:39,712 --> 00:08:43,268
on the hopes that this may
help us prevent diseases,

195
00:08:43,268 --> 00:08:46,788
we can ask whether monitoring
and analyzing the words we speak --

196
00:08:46,788 --> 00:08:47,786
we tweet,

197
00:08:47,786 --> 00:08:48,789
we email,

198
00:08:48,789 --> 00:08:49,793
we write --

199
00:08:49,793 --> 00:08:54,759
can tell us ahead of time whether
something may go wrong with our minds.

200
00:08:55,301 --> 00:08:56,834
And with Guillermo Cecci,

201
00:08:56,834 --> 00:08:59,962
who has been my brother
in this adventure,

202
00:08:59,962 --> 00:09:01,517
we took on this task.

203
00:09:02,489 --> 00:09:08,054
And we did so by analyzing the recorded
speech of 34 young people

204
00:09:08,054 --> 00:09:11,168
who were at a high risk
of developing schizophrenia.

205
00:09:11,568 --> 00:09:14,502
And so what we did is we
measured speech at day one

206
00:09:14,502 --> 00:09:17,743
and then we asked whether the properties
of the speech could predict,

207
00:09:17,743 --> 00:09:20,238
within a window of almost three years,

208
00:09:20,238 --> 00:09:22,919
the future development of psychosis.

209
00:09:23,570 --> 00:09:26,039
But despite our hopes,

210
00:09:26,039 --> 00:09:29,568
we got failure after failure.

211
00:09:29,928 --> 00:09:33,801
There was just not enough
information in semantics

212
00:09:33,801 --> 00:09:36,844
to predict the future
organization of the mind.

213
00:09:36,844 --> 00:09:38,556
It was good enough

214
00:09:38,556 --> 00:09:42,883
to distinguish between a group
of schizophrenics and a control group,

215
00:09:42,883 --> 00:09:45,555
a bit like we had done
for the ancient texts,

216
00:09:45,555 --> 00:09:49,108
but not to predict the future
onto the psychosis.

217
00:09:49,320 --> 00:09:51,109
But then we realized

218
00:09:51,109 --> 00:09:55,135
that maybe the most important thing
was not so much what they were saying

219
00:09:55,135 --> 00:09:57,322
but how they were saying it.

220
00:09:57,778 --> 00:09:59,138
More specifically,

221
00:09:59,138 --> 00:10:02,012
it was not in which semantic
neighborhoods the words were,

222
00:10:02,012 --> 00:10:04,611
but how far and fast they jumped

223
00:10:04,611 --> 00:10:07,076
from one semantic neighborhood
to the other one.

224
00:10:07,336 --> 00:10:09,065
And so we came up with this measure,

225
00:10:09,065 --> 00:10:11,624
which we termed semantic coherence,

226
00:10:11,624 --> 00:10:16,427
which essentially measures the persistence
of speech within one semantic topic,

227
00:10:16,427 --> 00:10:18,770
within one semantic category.

228
00:10:19,445 --> 00:10:23,502
And it turned out to be
that for this group of 34 people,

229
00:10:23,502 --> 00:10:27,465
the algorithm based on semantic
coherence could predict,

230
00:10:27,465 --> 00:10:29,630
with 100 percent accuracy,

231
00:10:29,630 --> 00:10:32,700
who developed psychosis and who will not.

232
00:10:33,178 --> 00:10:36,150
And this was something
that could not be achieved --

233
00:10:36,150 --> 00:10:37,733
not even close --

234
00:10:37,733 --> 00:10:41,123
with all the other
existing clinical measures.

235
00:10:42,779 --> 00:10:46,401
And I remember vividly
while I was working on this,

236
00:10:46,401 --> 00:10:48,750
I was sitting on my computer

237
00:10:48,750 --> 00:10:51,219
and I saw a bunch of tweets by Polo --

238
00:10:51,219 --> 00:10:54,385
Polo had been my first student
back in Buenos Aires

239
00:10:54,385 --> 00:10:56,413
and at the time he was living in New York.

240
00:10:56,413 --> 00:10:58,613
And there was something in this tweets --

241
00:10:58,613 --> 00:11:02,273
I could not tell exactly what
because nothing was said explicitly --

242
00:11:02,273 --> 00:11:04,293
but I got this strong hunch,

243
00:11:04,293 --> 00:11:07,873
this strong intuition
that something was going wrong.

244
00:11:08,510 --> 00:11:11,094
So I picked up the phone
and I called Polo,

245
00:11:11,094 --> 00:11:13,300
and in fact he was not feeling well.

246
00:11:13,582 --> 00:11:15,544
And this simple fact

247
00:11:15,544 --> 00:11:18,004
that reading in between the lines


248
00:11:18,004 --> 00:11:22,224
I could sense through words his feelings,

249
00:11:22,224 --> 00:11:25,196
was a simple but very
effective way to help.

250
00:11:26,154 --> 00:11:27,905
What I tell you today

251
00:11:27,905 --> 00:11:30,587
is that we're getting
close to understanding

252
00:11:30,587 --> 00:11:34,607
how we can convert this intuition
that we all have,

253
00:11:34,607 --> 00:11:36,125
that we all share,

254
00:11:36,125 --> 00:11:37,801
into an algorithm.

255
00:11:38,264 --> 00:11:39,840
And in doing so,

256
00:11:39,840 --> 00:11:44,489
we may be seeing in the future
a very different form of mental health,

257
00:11:44,489 --> 00:11:50,073
based on objective, quantitative
and automated analysis

258
00:11:50,073 --> 00:11:51,934
of the words we write,

259
00:11:51,934 --> 00:11:53,470
of the words we say.

260
00:11:53,470 --> 00:11:54,738
Gracias.

261
00:11:54,738 --> 00:11:56,736
(Applause)