1
00:00:04,970 --> 00:00:02,810
you really want to be educated to be

2
00:00:07,610 --> 00:00:04,980
somebody who cannot be replaced by a

3
00:00:09,710 --> 00:00:07,620
computer and I guarantee you that he

4
00:00:11,750 --> 00:00:09,720
will never be able to replace the most

5
00:00:15,490 --> 00:00:11,760
important part of us which is the

6
00:00:15,500 --> 00:00:19,910
are you sure about that

7
00:00:24,109 --> 00:00:21,529
today we talk about the boynage

8
00:00:26,210 --> 00:00:24,119
manuscript the Zodiac Cipher specialized

9
00:00:28,609 --> 00:00:26,220
non-human languages the Dora Bella

10
00:00:30,710 --> 00:00:28,619
Cipher ciphers in general and how to go

11
00:00:32,330 --> 00:00:30,720
about decrypting them great contract is

12
00:00:33,709 --> 00:00:32,340
a professor of artificial intelligence

13
00:00:35,630 --> 00:00:33,719
investigating natural language

14
00:00:37,370 --> 00:00:35,640
processing in a way that it relates to

15
00:00:38,990 --> 00:00:37,380
language reconstruction he's employed

16
00:00:40,910 --> 00:00:39,000
these machine learning techniques to

17
00:00:43,069 --> 00:00:40,920
attempt what can be considered the most

18
00:00:45,170 --> 00:00:43,079
objective decipherment of the voynage

19
00:00:47,270 --> 00:00:45,180
manuscript which is a considerably rare

20
00:00:48,889 --> 00:00:47,280
Illustrated codex handwritten in an

21
00:00:50,810 --> 00:00:48,899
otherwise unknown writing system it's

22
00:00:52,430 --> 00:00:50,820
evaded any attempt to decode it since

23
00:00:54,049 --> 00:00:52,440
the Italian Renaissance the voynage

24
00:00:56,270 --> 00:00:54,059
manuscript is written on an expensive

25
00:00:57,889 --> 00:00:56,280
Vellum and it's just one of the several

26
00:00:59,750 --> 00:00:57,899
puzzles that Professor contract has

27
00:01:01,610 --> 00:00:59,760
tackled another example being the Dora

28
00:01:03,590 --> 00:01:01,620
Bella Cipher Greg kondrak is also known

29
00:01:05,570 --> 00:01:03,600
for proving chomsky's statement wrong

30
00:01:07,429 --> 00:01:05,580
the statement that English orthography

31
00:01:08,690 --> 00:01:07,439
is close to Optimum my name is Kurt J

32
00:01:10,609 --> 00:01:08,700
mungle I have a background in

33
00:01:11,990 --> 00:01:10,619
mathematical physics this podcast is

34
00:01:14,330 --> 00:01:12,000
called theories of everything is

35
00:01:16,429 --> 00:01:14,340
dedicated to the exploration of theories

36
00:01:18,289 --> 00:01:16,439
of everything from a theoretical physics

37
00:01:19,550 --> 00:01:18,299
perspective but as well as exploring the

38
00:01:21,770 --> 00:01:19,560
world Consciousness has to the

39
00:01:23,690 --> 00:01:21,780
fundamental laws of nature each sponsor

40
00:01:25,609 --> 00:01:23,700
as well as the patrons improves the

41
00:01:27,350 --> 00:01:25,619
quality of the videos drastically it

42
00:01:29,210 --> 00:01:27,360
improves the depth it improves the

43
00:01:30,710 --> 00:01:29,220
frequency and it goes toward paying the

44
00:01:32,149 --> 00:01:30,720
staff for instance someone who's editing

45
00:01:33,950 --> 00:01:32,159
this full time right now and then we

46
00:01:35,149 --> 00:01:33,960
have an operations manager in that vein

47
00:01:37,010 --> 00:01:35,159
I want to thank today's sponsor

48
00:01:38,690 --> 00:01:37,020
brilliant if you're familiar with tow

49
00:01:40,490 --> 00:01:38,700
you're familiar with Brilliance but for

50
00:01:42,530 --> 00:01:40,500
those who don't know Brilliance is a

51
00:01:44,690 --> 00:01:42,540
place where you go to learn math science

52
00:01:46,670 --> 00:01:44,700
and engineering through these bite-sized

53
00:01:48,230 --> 00:01:46,680
Interactive Learning experiences for

54
00:01:49,850 --> 00:01:48,240
example and I keep saying this I would

55
00:01:52,969 --> 00:01:49,860
like to do a podcast on information

56
00:01:55,249 --> 00:01:52,979
Theory particularly Chiara marleto which

57
00:01:56,450 --> 00:01:55,259
is David Deutsch's student has a theory

58
00:01:58,550 --> 00:01:56,460
of everything that she puts forward

59
00:02:00,710 --> 00:01:58,560
called Constructor Theory which is

60
00:02:02,450 --> 00:02:00,720
heavily contingent on information Theory

61
00:02:04,490 --> 00:02:02,460
so I took their course on random

62
00:02:06,590 --> 00:02:04,500
variable distributions and knowledge and

63
00:02:08,570 --> 00:02:06,600
uncertainty in order to learn a bit more

64
00:02:10,370 --> 00:02:08,580
about entropy now there's this formula

65
00:02:12,350 --> 00:02:10,380
for entropy essentially hammered into

66
00:02:14,270 --> 00:02:12,360
you as an undergraduate which seems to

67
00:02:16,010 --> 00:02:14,280
have fallen from the sky however when

68
00:02:17,510 --> 00:02:16,020
you take Brilliance course it was the

69
00:02:19,790 --> 00:02:17,520
first time that I could see that it's an

70
00:02:22,610 --> 00:02:19,800
extremely clear and intuitive form

71
00:02:24,350 --> 00:02:22,620
formula that is to say that it would be

72
00:02:27,350 --> 00:02:24,360
unnatural to Define it in any other

73
00:02:29,510 --> 00:02:27,360
manner visit brilliant.org toe that is

74
00:02:31,309 --> 00:02:29,520
t-o-e to get 20 off the annual

75
00:02:33,890 --> 00:02:31,319
subscription and I recommend that you

76
00:02:35,630 --> 00:02:33,900
don't stop before four lessons I think

77
00:02:37,250 --> 00:02:35,640
you'll be greatly surprised that the

78
00:02:39,350 --> 00:02:37,260
ease at which you can now comprehend

79
00:02:41,330 --> 00:02:39,360
subjects you previously had a difficult

80
00:02:42,589 --> 00:02:41,340
time grocking at some point I'll also go

81
00:02:44,449 --> 00:02:42,599
through the courses and give a

82
00:02:45,530 --> 00:02:44,459
recommendation in order if you'd like to

83
00:02:47,750 --> 00:02:45,540
support the theories of everything

84
00:02:49,670 --> 00:02:47,760
podcast and help assist that is the

85
00:02:51,589 --> 00:02:49,680
improved video quality the depth the

86
00:02:53,390 --> 00:02:51,599
frequency of the podcast and paying the

87
00:02:55,150 --> 00:02:53,400
staff then do consider going to

88
00:02:57,470 --> 00:02:55,160
patreon.com

89
00:02:59,089 --> 00:02:57,480
kurtjimungle the link is on screen as

90
00:03:00,650 --> 00:02:59,099
well as in the description there's a

91
00:03:02,690 --> 00:03:00,660
custom amount as well as already

92
00:03:04,309 --> 00:03:02,700
delineated tiers either way your

93
00:03:06,949 --> 00:03:04,319
viewership is thank you enough enjoy

94
00:03:09,050 --> 00:03:06,959
this conversation with Greg kondrak

95
00:03:11,449 --> 00:03:09,060
Professor what is the voynish manuscript

96
00:03:14,270 --> 00:03:11,459
and why is it important

97
00:03:18,530 --> 00:03:14,280
about a transcript is a medieval

98
00:03:21,290 --> 00:03:18,540
manuscript written in some code that is

99
00:03:25,190 --> 00:03:21,300
was actually confirmed to be

100
00:03:26,830 --> 00:03:25,200
actually genuine manuscript from the

101
00:03:31,250 --> 00:03:26,840
15th century

102
00:03:35,149 --> 00:03:31,260
it is it has illustrations it has text

103
00:03:40,130 --> 00:03:35,159
the script or the alphabet is unique for

104
00:03:40,670 --> 00:03:40,140
the manuscript has not been deciphered

105
00:03:42,830 --> 00:03:40,680
um

106
00:03:46,490 --> 00:03:42,840
yet

107
00:03:50,089 --> 00:03:46,500
why hasn't it been deciphered

108
00:03:52,009 --> 00:03:50,099
uh that's uh not we don't know why it

109
00:03:55,130 --> 00:03:52,019
hasn't been deciphered some people say

110
00:03:57,470 --> 00:03:55,140
it's because there's nothing to decipher

111
00:04:01,610 --> 00:03:57,480
and it's it's just some kind of a joke

112
00:04:04,970 --> 00:04:01,620
or other people say that the encoding

113
00:04:07,009 --> 00:04:04,980
system is very complicated

114
00:04:09,410 --> 00:04:07,019
I mean other people have other theories

115
00:04:11,630 --> 00:04:09,420
about it why do people say it's a joke

116
00:04:12,670 --> 00:04:11,640
and what do you make of that

117
00:04:17,330 --> 00:04:12,680
yeah

118
00:04:19,569 --> 00:04:17,340
well I I personally I don't think uh as

119
00:04:22,069 --> 00:04:19,579
I said that this is a testable

120
00:04:24,530 --> 00:04:22,079
scientific hypothesis

121
00:04:26,270 --> 00:04:24,540
you can you can guess that it's a joke

122
00:04:27,610 --> 00:04:26,280
but it's very hard to prove something

123
00:04:30,590 --> 00:04:27,620
like that

124
00:04:33,110 --> 00:04:30,600
obviously it cost a lot of money to

125
00:04:34,610 --> 00:04:33,120
produce this kind of manuscript in the

126
00:04:37,370 --> 00:04:34,620
Middle Ages so

127
00:04:39,710 --> 00:04:37,380
that's uh one reason that I don't think

128
00:04:46,129 --> 00:04:39,720
it was a joke

129
00:04:47,570 --> 00:04:46,139
um well but also it it's um there's some

130
00:04:50,810 --> 00:04:47,580
words showing that there are some

131
00:04:53,150 --> 00:04:50,820
statistical properties that indicates

132
00:04:54,650 --> 00:04:53,160
there is actually a language that is

133
00:04:56,689 --> 00:04:54,660
being in code

134
00:04:58,730 --> 00:04:56,699
I've seen several documentaries on the

135
00:05:00,830 --> 00:04:58,740
Voynich manuscript several so not just

136
00:05:02,870 --> 00:05:00,840
one or two there's a variety of them and

137
00:05:05,030 --> 00:05:02,880
then there's also a whole subreddit so a

138
00:05:07,430 --> 00:05:05,040
whole Reddit group dedicated to solving

139
00:05:09,770 --> 00:05:07,440
this why is this so difficult compared

140
00:05:12,350 --> 00:05:09,780
to other ciphers in the past like what

141
00:05:17,090 --> 00:05:12,360
is it about this

142
00:05:18,590 --> 00:05:17,100
well the difficulty here is that the

143
00:05:21,310 --> 00:05:18,600
first difficult is that we don't know

144
00:05:23,990 --> 00:05:21,320
what language it is

145
00:05:26,330 --> 00:05:24,000
very often we have ciphers we have

146
00:05:29,629 --> 00:05:26,340
messages but we know what the message is

147
00:05:32,689 --> 00:05:29,639
for example the German Enigma machine

148
00:05:37,010 --> 00:05:32,699
was a very hard Cipher but we knew that

149
00:05:39,409 --> 00:05:37,020
it was German being encoded which made

150
00:05:41,210 --> 00:05:39,419
it quite easier

151
00:05:43,730 --> 00:05:41,220
and the second thing is that we don't

152
00:05:45,650 --> 00:05:43,740
know the script or the alphabet

153
00:05:48,230 --> 00:05:45,660
if that alphabet was used for something

154
00:05:50,629 --> 00:05:48,240
else we would know how to speak it how

155
00:05:54,469 --> 00:05:50,639
to pronounce these words

156
00:05:55,909 --> 00:05:54,479
and third problem is that this is just a

157
00:05:57,070 --> 00:05:55,919
unique document there's no other

158
00:06:00,710 --> 00:05:57,080
document

159
00:06:04,730 --> 00:06:00,720
that is it is written in this way

160
00:06:07,070 --> 00:06:04,740
so it's all self-contained and we don't

161
00:06:08,870 --> 00:06:07,080
even know we're not exactly sure where

162
00:06:10,430 --> 00:06:08,880
it was even produced

163
00:06:12,590 --> 00:06:10,440
it's strange that there's no other

164
00:06:13,969 --> 00:06:12,600
document that's like that firstly that

165
00:06:15,170 --> 00:06:13,979
the script is different like you

166
00:06:16,969 --> 00:06:15,180
mentioned that we don't know how to

167
00:06:19,070 --> 00:06:16,979
pronounce the words though even in the

168
00:06:20,629 --> 00:06:19,080
Enigma code it's not as if the words

169
00:06:21,950 --> 00:06:20,639
were meant to be pronounced and then

170
00:06:23,930 --> 00:06:21,960
understood like that you could have

171
00:06:25,070 --> 00:06:23,940
translated it to zeros and ones and it

172
00:06:27,110 --> 00:06:25,080
still would have been difficult it still

173
00:06:28,490 --> 00:06:27,120
would have been the same problem so why

174
00:06:30,770 --> 00:06:28,500
does the fact that we don't understand

175
00:06:32,090 --> 00:06:30,780
how to pronounce the alphabet make a

176
00:06:33,710 --> 00:06:32,100
difference why can't we just say look

177
00:06:36,170 --> 00:06:33,720
this letter appears let's call that

178
00:06:38,629 --> 00:06:36,180
letter 28 let's call that other letter

179
00:06:40,850 --> 00:06:38,639
50. well yeah well so we have letters

180
00:06:42,710 --> 00:06:40,860
but we have words words are made of

181
00:06:45,409 --> 00:06:42,720
letters that's the truth in every

182
00:06:47,510 --> 00:06:45,419
language human language that you have

183
00:06:51,830 --> 00:06:47,520
words that are made of

184
00:06:53,090 --> 00:06:51,840
phonings they're usually every phone has

185
00:06:57,430 --> 00:06:53,100
its own letter

186
00:07:01,129 --> 00:06:57,440
and then there are certain

187
00:07:03,770 --> 00:07:01,139
patterns regularities for example there

188
00:07:05,809 --> 00:07:03,780
are certain words that can be pronounced

189
00:07:08,029 --> 00:07:05,819
and other words that cannot be

190
00:07:09,290 --> 00:07:08,039
pronounced and that varies from language

191
00:07:13,490 --> 00:07:09,300
to language

192
00:07:16,430 --> 00:07:13,500
so so the moment you you know what

193
00:07:18,430 --> 00:07:16,440
language we're speaking we know you

194
00:07:21,409 --> 00:07:18,440
really know the Lexicon of that language

195
00:07:23,930 --> 00:07:21,419
typically a few thousand words

196
00:07:26,150 --> 00:07:23,940
and they have different frequencies

197
00:07:28,430 --> 00:07:26,160
so that makes it a lot easier to

198
00:07:30,909 --> 00:07:28,440
decipher anything even if you just

199
00:07:32,809 --> 00:07:30,919
replace every word with a number

200
00:07:34,490 --> 00:07:32,819
that will still give you some

201
00:07:37,010 --> 00:07:34,500
information about the frequency of the

202
00:07:38,450 --> 00:07:37,020
words as long as you do something to

203
00:07:39,890 --> 00:07:38,460
compare it to

204
00:07:41,930 --> 00:07:39,900
I know there are different kinds of

205
00:07:43,490 --> 00:07:41,940
ciphers and I believe the simplest is

206
00:07:44,809 --> 00:07:43,500
called the substitution Cipher that's

207
00:07:48,110 --> 00:07:44,819
like when you're a kid and you just

208
00:07:51,469 --> 00:07:48,120
replace the letter A with c and so on

209
00:07:53,749 --> 00:07:51,479
what other kinds of ciphers are there

210
00:07:56,150 --> 00:07:53,759
yeah so the principally there are

211
00:07:59,330 --> 00:07:56,160
substitution ciphers which is replacing

212
00:08:02,570 --> 00:07:59,340
a symbols and transposition ciphers

213
00:08:04,309 --> 00:08:02,580
which is mixing them up changing their

214
00:08:08,390 --> 00:08:04,319
sequence

215
00:08:10,010 --> 00:08:08,400
and every Cipher is a combination of

216
00:08:13,309 --> 00:08:10,020
those two methods

217
00:08:14,450 --> 00:08:13,319
now we don't know what is varnish we

218
00:08:18,309 --> 00:08:14,460
don't know if it's just a simple

219
00:08:20,930 --> 00:08:18,319
substitution or is it a substitution

220
00:08:23,990 --> 00:08:20,940
combined with transposition

221
00:08:26,390 --> 00:08:24,000
the paper that we wrote assumed that the

222
00:08:28,969 --> 00:08:26,400
transposition was involved

223
00:08:33,170 --> 00:08:28,979
and we try to come up with a general

224
00:08:34,969 --> 00:08:33,180
method of breaking this kind of ciphers

225
00:08:36,350 --> 00:08:34,979
that combine substitutions in just

226
00:08:38,570 --> 00:08:36,360
positions

227
00:08:40,909 --> 00:08:38,580
I completely glossed over the history

228
00:08:42,829 --> 00:08:40,919
actually I not that I glossed over I had

229
00:08:45,170 --> 00:08:42,839
forgotten to even ask you to give the

230
00:08:47,329 --> 00:08:45,180
audience an indication as to where was

231
00:08:49,850 --> 00:08:47,339
this document found why does it matter

232
00:08:51,949 --> 00:08:49,860
why is it that Scholars even care about

233
00:08:54,230 --> 00:08:51,959
it I'm sure there's plenty about the

234
00:08:56,090 --> 00:08:54,240
past that we don't know about so why is

235
00:08:57,949 --> 00:08:56,100
it that many scholars and not only

236
00:09:00,230 --> 00:08:57,959
Scholars groups of people teams of

237
00:09:01,730 --> 00:09:00,240
people on Reddit are pouring over trying

238
00:09:03,530 --> 00:09:01,740
to figure out what the heck is this

239
00:09:06,350 --> 00:09:03,540
saying where was it from can you give a

240
00:09:10,810 --> 00:09:06,360
bit of the history of it please

241
00:09:13,970 --> 00:09:10,820
sure so sorry was found basically by the

242
00:09:16,910 --> 00:09:13,980
person whose name was Vonage that's what

243
00:09:18,230 --> 00:09:16,920
it's called the Voynich it was a kind of

244
00:09:22,430 --> 00:09:18,240
a collector

245
00:09:26,090 --> 00:09:22,440
uh in the beginning of the 20th century

246
00:09:29,630 --> 00:09:26,100
and since then the the manuscript has

247
00:09:32,630 --> 00:09:29,640
been tracked back to 17th century to the

248
00:09:34,430 --> 00:09:32,640
Court of uh Roman Emperor and that's

249
00:09:36,170 --> 00:09:34,440
where it kind of ends that's where the

250
00:09:41,050 --> 00:09:36,180
Trail Ends

251
00:09:43,670 --> 00:09:41,060
uh but as uh not long ago there was a

252
00:09:46,009 --> 00:09:43,680
chemical analysis done on the manuscript

253
00:09:49,070 --> 00:09:46,019
so we're certain that it was actually

254
00:09:52,250 --> 00:09:49,080
written in the 15th century

255
00:09:55,130 --> 00:09:52,260
so there is no doubt about that anymore

256
00:09:57,290 --> 00:09:55,140
now you also asked the other question

257
00:09:58,370 --> 00:09:57,300
which is why are people so fascinated

258
00:10:01,370 --> 00:09:58,380
with it

259
00:10:03,410 --> 00:10:01,380
I think the main reason is that we are

260
00:10:06,530 --> 00:10:03,420
fascinated by puzzles if we see

261
00:10:08,930 --> 00:10:06,540
something that seems to be a message we

262
00:10:10,790 --> 00:10:08,940
want to know what that message is we

263
00:10:14,329 --> 00:10:10,800
want to decipher it

264
00:10:15,050 --> 00:10:14,339
and Voyage is like a Mount Everest of

265
00:10:18,250 --> 00:10:15,060
all

266
00:10:21,350 --> 00:10:18,260
uh overloading

267
00:10:23,150 --> 00:10:21,360
cryptographic puzzles

268
00:10:26,810 --> 00:10:23,160
it was actually

269
00:10:30,110 --> 00:10:26,820
studied by many years by people that

270
00:10:32,690 --> 00:10:30,120
were working for the U.S government

271
00:10:35,570 --> 00:10:32,700
and that they were professional Breakers

272
00:10:37,850 --> 00:10:35,580
code Breakers that that broke cause

273
00:10:40,550 --> 00:10:37,860
during the second World War

274
00:10:41,990 --> 00:10:40,560
in the case of the Enigma code we at

275
00:10:43,730 --> 00:10:42,000
least we could presume that it has

276
00:10:46,250 --> 00:10:43,740
something to do with government secrets

277
00:10:47,810 --> 00:10:46,260
and War whereas with this do we have any

278
00:10:49,730 --> 00:10:47,820
indication as to what the subject matter

279
00:10:51,769 --> 00:10:49,740
is

280
00:10:54,829 --> 00:10:51,779
well we have the illustrations that's

281
00:11:00,230 --> 00:10:54,839
what makes it really interesting is and

282
00:11:06,769 --> 00:11:04,069
all kinds of strange illustrations that

283
00:11:10,009 --> 00:11:06,779
seem to be related to the text

284
00:11:11,930 --> 00:11:10,019
so so this is not just a text like some

285
00:11:14,329 --> 00:11:11,940
kind of ancient inscription but it is

286
00:11:16,550 --> 00:11:14,339
actually a codex which is like a

287
00:11:19,190 --> 00:11:16,560
compedium of some kind of knowledge

288
00:11:21,710 --> 00:11:19,200
which was quite common in Middle Ages

289
00:11:23,090 --> 00:11:21,720
and is it quite common among a

290
00:11:24,590 --> 00:11:23,100
particular group that speaks a

291
00:11:26,210 --> 00:11:24,600
particular language and thus we could

292
00:11:28,970 --> 00:11:26,220
figure out okay with this amount of

293
00:11:30,970 --> 00:11:28,980
probability it's from this culture or

294
00:11:35,690 --> 00:11:30,980
this group of people

295
00:11:38,210 --> 00:11:35,700
yeah so many of those codex codices are

296
00:11:41,449 --> 00:11:38,220
in Latin which was the language

297
00:11:44,810 --> 00:11:41,459
of literature and Science in Middle Ages

298
00:11:48,350 --> 00:11:44,820
and you can actually find such such

299
00:11:50,569 --> 00:11:48,360
books very similar similar looking for

300
00:11:53,870 --> 00:11:50,579
Middle Ages that are written in Latin

301
00:11:56,930 --> 00:11:53,880
the best guess is about the provenance

302
00:11:58,310 --> 00:11:56,940
of the manuscripts uh point at Northern

303
00:12:01,850 --> 00:11:58,320
Italy

304
00:12:04,069 --> 00:12:01,860
were of course Latin was the language of

305
00:12:06,170 --> 00:12:04,079
literature as well at that time

306
00:12:08,449 --> 00:12:06,180
is it controversial that it's in the

307
00:12:10,970 --> 00:12:08,459
15th century and in Northern Italy

308
00:12:14,030 --> 00:12:10,980
almost everything that you can say about

309
00:12:16,610 --> 00:12:14,040
foreign so

310
00:12:18,110 --> 00:12:16,620
this is what I consider a reasonable

311
00:12:19,850 --> 00:12:18,120
guest but

312
00:12:22,430 --> 00:12:19,860
and pretty much everybody has a

313
00:12:24,889 --> 00:12:22,440
different opinion where this uh this

314
00:12:26,509 --> 00:12:24,899
manuscript comes from and what language

315
00:12:29,389 --> 00:12:26,519
it represents

316
00:12:32,990 --> 00:12:29,399
I I don't have any it is not my research

317
00:12:36,230 --> 00:12:33,000
that points to the Northern Italy

318
00:12:37,850 --> 00:12:36,240
I only looked at this manuscript from

319
00:12:39,350 --> 00:12:37,860
the point of view of computation

320
00:12:41,269 --> 00:12:39,360
decipherment

321
00:12:43,310 --> 00:12:41,279
speaking of your research now would be a

322
00:12:44,870 --> 00:12:43,320
great time to tell the audience what is

323
00:12:46,069 --> 00:12:44,880
it that you study and then how the heck

324
00:12:48,170 --> 00:12:46,079
did you become interested in the

325
00:12:51,230 --> 00:12:48,180
voendage manuscript other than a general

326
00:12:54,410 --> 00:12:51,240
curiosity for solving puzzles

327
00:12:57,050 --> 00:12:54,420
yeah so I I'm a computational linguist

328
00:12:59,509 --> 00:12:57,060
I'd say so I I work at the computer

329
00:13:01,730 --> 00:12:59,519
science department at the University of

330
00:13:04,610 --> 00:13:01,740
Alberta in Canada

331
00:13:07,310 --> 00:13:04,620
and I work on language in general making

332
00:13:10,069 --> 00:13:07,320
computers understand human language

333
00:13:12,350 --> 00:13:10,079
making a rewriting programs that can

334
00:13:15,110 --> 00:13:12,360
process human language and do the work

335
00:13:17,269 --> 00:13:15,120
for us because there's so much text

336
00:13:19,910 --> 00:13:17,279
that is available that nobody can

337
00:13:23,090 --> 00:13:19,920
actually read that all of that all of it

338
00:13:25,670 --> 00:13:23,100
about the the decipherment

339
00:13:27,650 --> 00:13:25,680
the person that made this really

340
00:13:29,930 --> 00:13:27,660
interesting for me was Professor Kevin

341
00:13:31,370 --> 00:13:29,940
knight from University of Southern

342
00:13:35,449 --> 00:13:31,380
California

343
00:13:37,970 --> 00:13:35,459
and he worked on various interesting

344
00:13:39,650 --> 00:13:37,980
projects and I saw his presentation of

345
00:13:42,710 --> 00:13:39,660
wineish manuscript

346
00:13:45,949 --> 00:13:42,720
about 10 years ago I I say

347
00:13:46,730 --> 00:13:45,959
and it it was related to what I was

348
00:13:50,269 --> 00:13:46,740
doing

349
00:13:52,490 --> 00:13:50,279
what Kevin Dr Knight said basically is

350
00:13:54,350 --> 00:13:52,500
that everything we do with language is a

351
00:13:56,329 --> 00:13:54,360
kind of decipherment

352
00:13:59,329 --> 00:13:56,339
because language

353
00:14:01,370 --> 00:13:59,339
we is typically written that's what we

354
00:14:03,889 --> 00:14:01,380
work with a written language even if

355
00:14:06,949 --> 00:14:03,899
it's spoken we work we work with some

356
00:14:08,930 --> 00:14:06,959
form of it that is a transcription

357
00:14:11,690 --> 00:14:08,940
and whenever you have a sequence of

358
00:14:15,290 --> 00:14:11,700
symbols it becomes a decipherment

359
00:14:17,269 --> 00:14:15,300
problem basically deciphering first what

360
00:14:19,430 --> 00:14:17,279
are the phonemes behind those symbols

361
00:14:20,870 --> 00:14:19,440
and secondly is the meaning behind the

362
00:14:22,670 --> 00:14:20,880
symbols

363
00:14:25,069 --> 00:14:22,680
with respect to the images in the

364
00:14:26,990 --> 00:14:25,079
Voynich which will overlay on screen is

365
00:14:29,090 --> 00:14:27,000
it strange to depict what they depict

366
00:14:30,410 --> 00:14:29,100
the pictures yeah yeah like what are the

367
00:14:32,470 --> 00:14:30,420
pictures of and is there something

368
00:14:36,230 --> 00:14:32,480
unique about them

369
00:14:38,810 --> 00:14:36,240
yes so so the generally if I tell you

370
00:14:41,030 --> 00:14:38,820
that it depicts plans for example that's

371
00:14:43,850 --> 00:14:41,040
not strange because that's what the

372
00:14:46,790 --> 00:14:43,860
medieval codices do but if you look

373
00:14:49,370 --> 00:14:46,800
closely at those plants if you're an

374
00:14:51,650 --> 00:14:49,380
expert in Plants I am not but

375
00:14:53,269 --> 00:14:51,660
experts on Plants look at them and say

376
00:14:55,550 --> 00:14:53,279
well these don't really look like real

377
00:14:56,750 --> 00:14:55,560
plants these are look like made up

378
00:14:59,449 --> 00:14:56,760
plants

379
00:15:04,490 --> 00:14:59,459
and then there are pictures of people

380
00:15:06,230 --> 00:15:04,500
that some kind of many naked bodies

381
00:15:09,170 --> 00:15:06,240
taking baths

382
00:15:11,990 --> 00:15:09,180
in some kind of green water

383
00:15:14,329 --> 00:15:12,000
you know in general pictures of people

384
00:15:16,310 --> 00:15:14,339
are not strange but those particular

385
00:15:19,430 --> 00:15:16,320
pictures are really strange

386
00:15:21,290 --> 00:15:19,440
and they are unlike anything else that

387
00:15:23,150 --> 00:15:21,300
we know from middle age

388
00:15:25,370 --> 00:15:23,160
they're strange because they're naked or

389
00:15:27,350 --> 00:15:25,380
they're strange because they're depicted

390
00:15:29,990 --> 00:15:27,360
alongside plants like what is it

391
00:15:32,090 --> 00:15:30,000
specifically that's unique

392
00:15:34,490 --> 00:15:32,100
The Exchange because it's not clear what

393
00:15:37,370 --> 00:15:34,500
they are depicting are they depicting

394
00:15:39,650 --> 00:15:37,380
people taking bath that I don't think

395
00:15:43,610 --> 00:15:39,660
was very common in Middle Ages

396
00:15:46,550 --> 00:15:43,620
or why are why are those figures for

397
00:15:49,910 --> 00:15:46,560
example all women and why they are naked

398
00:15:53,930 --> 00:15:49,920
right in 15th century that that was not

399
00:15:55,850 --> 00:15:53,940
a normal thing to put in a book

400
00:15:59,509 --> 00:15:55,860
um there are also other things like

401
00:16:01,069 --> 00:15:59,519
zodiac signs or or pictures of plan of

402
00:16:04,389 --> 00:16:01,079
of planets

403
00:16:07,250 --> 00:16:04,399
that you would expect to be

404
00:16:10,189 --> 00:16:07,260
to be quite normal because we even know

405
00:16:14,090 --> 00:16:10,199
those the what those Zodiacs are

406
00:16:16,009 --> 00:16:14,100
but it's it's difficult to connect the

407
00:16:18,590 --> 00:16:16,019
words that describes those pictures to

408
00:16:20,629 --> 00:16:18,600
the actual pictures

409
00:16:22,790 --> 00:16:20,639
in one of the documentaries that I was

410
00:16:25,389 --> 00:16:22,800
watching about this they said that a

411
00:16:29,090 --> 00:16:25,399
remarkable element is that there are

412
00:16:31,310 --> 00:16:29,100
extremely few errors in a writing of

413
00:16:32,509 --> 00:16:31,320
this size they would expect that there

414
00:16:34,250 --> 00:16:32,519
are some errors and then maybe you

415
00:16:35,870 --> 00:16:34,260
smudge it out or however they correct

416
00:16:37,490 --> 00:16:35,880
errors and there's some way of detecting

417
00:16:39,769 --> 00:16:37,500
the frequency of the errors in a

418
00:16:42,230 --> 00:16:39,779
document and most documents have let's

419
00:16:44,629 --> 00:16:42,240
say error percentage two two percents

420
00:16:48,170 --> 00:16:44,639
they make an error every two out of 100

421
00:16:50,210 --> 00:16:48,180
words whereas this it's much lower like

422
00:16:51,470 --> 00:16:50,220
half of that or even half of half of

423
00:16:53,689 --> 00:16:51,480
that first thing I want to know is that

424
00:16:55,490 --> 00:16:53,699
even true and then secondly like why why

425
00:16:57,470 --> 00:16:55,500
do you think that's the case what could

426
00:16:59,749 --> 00:16:57,480
that mean

427
00:17:01,850 --> 00:16:59,759
yeah so first of all I did not work with

428
00:17:04,250 --> 00:17:01,860
the actual text I worked with the

429
00:17:07,429 --> 00:17:04,260
transcription that somebody made

430
00:17:09,829 --> 00:17:07,439
but it is from what I know it is true

431
00:17:12,289 --> 00:17:09,839
that there is very few Corrections

432
00:17:15,409 --> 00:17:12,299
my personal opinion that it may indicate

433
00:17:17,449 --> 00:17:15,419
that whoever was writing this copying it

434
00:17:19,970 --> 00:17:17,459
did not understand what they were

435
00:17:21,710 --> 00:17:19,980
writing you usually make Corrections if

436
00:17:24,110 --> 00:17:21,720
you write something and you see oh it's

437
00:17:25,429 --> 00:17:24,120
that's not what it should be right but

438
00:17:28,429 --> 00:17:25,439
if you write something in a language

439
00:17:30,110 --> 00:17:28,439
that you're totally unfamiliar with you

440
00:17:32,090 --> 00:17:30,120
won't be able to notice that that's

441
00:17:34,370 --> 00:17:32,100
interesting in that case that would

442
00:17:38,510 --> 00:17:34,380
imply that there's another copy no that

443
00:17:42,049 --> 00:17:38,520
means that I I there were people that

444
00:17:46,490 --> 00:17:42,059
maybe were copying the text

445
00:17:48,470 --> 00:17:46,500
into the manuscript because that was a

446
00:17:50,690 --> 00:17:48,480
you would expect there was some draft

447
00:17:53,450 --> 00:17:50,700
that we're copying from because the

448
00:17:56,029 --> 00:17:53,460
manuscript itself was very expensive uh

449
00:17:58,070 --> 00:17:56,039
to write down and we also know that

450
00:17:59,990 --> 00:17:58,080
there are different kind of hands so it

451
00:18:01,909 --> 00:18:00,000
looks like they were more than a one

452
00:18:04,370 --> 00:18:01,919
person doing the copying that's

453
00:18:06,350 --> 00:18:04,380
interesting then that means that it's

454
00:18:08,630 --> 00:18:06,360
extra difficult to decipher it because

455
00:18:10,310 --> 00:18:08,640
there are errors we don't know this but

456
00:18:11,870 --> 00:18:10,320
that would mean that there's more than

457
00:18:13,909 --> 00:18:11,880
the average amount of Errors what's the

458
00:18:15,350 --> 00:18:13,919
difference between inciphering so

459
00:18:17,810 --> 00:18:15,360
creating a cipher out of something and

460
00:18:19,669 --> 00:18:17,820
then encryption

461
00:18:21,830 --> 00:18:19,679
yeah I'm not sure if there is a

462
00:18:23,330 --> 00:18:21,840
difference you know if you really try to

463
00:18:24,830 --> 00:18:23,340
find the difference you could say that

464
00:18:27,590 --> 00:18:24,840
encryption is

465
00:18:29,450 --> 00:18:27,600
like you do encryption of a text but you

466
00:18:32,990 --> 00:18:29,460
don't really know how it is done you

467
00:18:35,690 --> 00:18:33,000
know you apply some encryption program

468
00:18:37,789 --> 00:18:35,700
whereas in ciphering implies that you go

469
00:18:40,130 --> 00:18:37,799
like letter by letter look up the key

470
00:18:41,570 --> 00:18:40,140
and then Cipher each each letter

471
00:18:44,210 --> 00:18:41,580
separately

472
00:18:45,950 --> 00:18:44,220
but in general I think it's it's pretty

473
00:18:49,130 --> 00:18:45,960
much the same thing

474
00:18:51,830 --> 00:18:49,140
what were the different methods used to

475
00:18:53,750 --> 00:18:51,840
decipher the Voynich manuscript not just

476
00:18:55,970 --> 00:18:53,760
by you but by others what techniques do

477
00:18:58,970 --> 00:18:55,980
they employ

478
00:19:02,150 --> 00:18:58,980
well that's a kind of difficult to say

479
00:19:05,210 --> 00:19:02,160
because none of those methods actually

480
00:19:08,750 --> 00:19:05,220
worked and then deciphering has not been

481
00:19:10,990 --> 00:19:08,760
achieved in spite of many claims

482
00:19:14,330 --> 00:19:11,000
to the country so

483
00:19:16,490 --> 00:19:14,340
the there isn't really any algorithm

484
00:19:20,110 --> 00:19:16,500
involved it's it's mostly based on

485
00:19:24,529 --> 00:19:20,120
people's intuition and and theories

486
00:19:27,650 --> 00:19:24,539
but but even then uh if even if there

487
00:19:31,730 --> 00:19:27,660
was some method to it it has not

488
00:19:33,590 --> 00:19:31,740
produced a readable transcription

489
00:19:34,789 --> 00:19:33,600
what are some of your attempts to

490
00:19:37,730 --> 00:19:34,799
decipher can you go through the

491
00:19:41,330 --> 00:19:37,740
successes and failures

492
00:19:43,070 --> 00:19:41,340
yes so so uh in our project our

493
00:19:46,310 --> 00:19:43,080
assumption was that the first thing to

494
00:19:49,430 --> 00:19:46,320
basically to start with is to find out

495
00:19:50,750 --> 00:19:49,440
what language this is written in if we

496
00:19:52,310 --> 00:19:50,760
don't know what the language it is

497
00:19:53,690 --> 00:19:52,320
written and then there's no way to

498
00:19:59,289 --> 00:19:53,700
decipher it

499
00:20:02,150 --> 00:19:59,299
so we devised some methods of detecting

500
00:20:04,970 --> 00:20:02,160
identifying the language of the cipher

501
00:20:08,510 --> 00:20:04,980
even without deciphering it

502
00:20:10,909 --> 00:20:08,520
and we use the sample a large sample of

503
00:20:13,909 --> 00:20:10,919
about 400 languages

504
00:20:16,070 --> 00:20:13,919
and out of those far-handed languages we

505
00:20:17,210 --> 00:20:16,080
assign like a number as score to each

506
00:20:20,150 --> 00:20:17,220
language

507
00:20:21,409 --> 00:20:20,160
in terms of the probability that this

508
00:20:23,210 --> 00:20:21,419
language is the language of the

509
00:20:26,870 --> 00:20:23,220
manuscript

510
00:20:29,150 --> 00:20:26,880
and what was interesting that we found

511
00:20:31,490 --> 00:20:29,160
was that the language that was the the

512
00:20:33,830 --> 00:20:31,500
highest scoring language or out of those

513
00:20:35,990 --> 00:20:33,840
400 was Hebrew

514
00:20:38,150 --> 00:20:36,000
how much higher was it than the second

515
00:20:41,450 --> 00:20:38,160
and third place

516
00:20:43,130 --> 00:20:41,460
was a clear difference I would say a

517
00:20:44,870 --> 00:20:43,140
significant difference between the

518
00:20:47,930 --> 00:20:44,880
second one in the list

519
00:20:50,690 --> 00:20:47,940
so that was quite striking

520
00:20:52,850 --> 00:20:50,700
and were you able to get any other

521
00:20:55,850 --> 00:20:52,860
historical documents that are written in

522
00:20:57,830 --> 00:20:55,860
Hebrew that have a similar art style and

523
00:20:59,450 --> 00:20:57,840
are of similar length just to see well

524
00:21:01,430 --> 00:20:59,460
is this common is this a common practice

525
00:21:05,210 --> 00:21:01,440
to the people who write in Hebrew or is

526
00:21:08,630 --> 00:21:05,220
this aberrant is this extremely unique

527
00:21:10,789 --> 00:21:08,640
so the Hebrew manuscripts exist and they

528
00:21:11,750 --> 00:21:10,799
were written throughout Middle Ages in

529
00:21:15,049 --> 00:21:11,760
Hebrew

530
00:21:18,710 --> 00:21:15,059
by by the Jewish Scholars

531
00:21:21,470 --> 00:21:18,720
and I'm not the only person that that uh

532
00:21:24,789 --> 00:21:21,480
that hypothesized that this this was

533
00:21:27,169 --> 00:21:24,799
actually coming from from the Jewish

534
00:21:30,289 --> 00:21:27,179
scholar community

535
00:21:32,049 --> 00:21:30,299
uh now nobody used this kind of

536
00:21:34,990 --> 00:21:32,059
particular script

537
00:21:39,590 --> 00:21:35,000
but this script

538
00:21:42,710 --> 00:21:39,600
does have some some similarities to

539
00:21:47,570 --> 00:21:42,720
Hebrew script for example

540
00:21:49,909 --> 00:21:47,580
I actually don't speak hero uh or or I

541
00:21:51,590 --> 00:21:49,919
don't know much about it but I know that

542
00:21:54,289 --> 00:21:51,600
the Hebrew script does not include

543
00:21:55,970 --> 00:21:54,299
letters sorry include vowels which makes

544
00:21:58,610 --> 00:21:55,980
the words shorter

545
00:22:01,730 --> 00:21:58,620
and this is what we observe in vanished

546
00:22:04,970 --> 00:22:01,740
is that the words are quite short

547
00:22:08,149 --> 00:22:04,980
and then the number uh the number of uh

548
00:22:10,970 --> 00:22:08,159
different symbols suggest that it is

549
00:22:13,130 --> 00:22:10,980
something like a substitution Cipher

550
00:22:14,930 --> 00:22:13,140
because the number of symbol is similar

551
00:22:16,010 --> 00:22:14,940
to the number of phonemes in a typical

552
00:22:18,169 --> 00:22:16,020
language

553
00:22:19,490 --> 00:22:18,179
okay so now that you have at least

554
00:22:22,310 --> 00:22:19,500
potentially identified the language

555
00:22:25,789 --> 00:22:22,320
what's the next step

556
00:22:28,250 --> 00:22:25,799
yeah a very good point so the next point

557
00:22:30,649 --> 00:22:28,260
the next step is obviously try to match

558
00:22:32,690 --> 00:22:30,659
every symbol to a different letter of

559
00:22:34,490 --> 00:22:32,700
the say Hebrew

560
00:22:38,330 --> 00:22:34,500
alphabet

561
00:22:41,390 --> 00:22:38,340
and that usually is easy the breaking

562
00:22:43,789 --> 00:22:41,400
simple substitution Cipher's disease but

563
00:22:47,350 --> 00:22:43,799
it doesn't work in this case it does not

564
00:22:50,570 --> 00:22:47,360
produce any sensible decipherment

565
00:22:53,270 --> 00:22:50,580
so we came up with this hypothesis that

566
00:22:56,570 --> 00:22:53,280
the letters within words are actually

567
00:22:58,730 --> 00:22:56,580
transposed to make it more difficult

568
00:23:02,450 --> 00:22:58,740
to decipher

569
00:23:04,669 --> 00:23:02,460
and when you kind of move the letters

570
00:23:06,770 --> 00:23:04,679
around it becomes very difficult to

571
00:23:08,510 --> 00:23:06,780
decipher it so we came up with a method

572
00:23:10,070 --> 00:23:08,520
that could handle this kind of

573
00:23:12,950 --> 00:23:10,080
transposition

574
00:23:16,570 --> 00:23:12,960
within words and we tested it on other

575
00:23:20,029 --> 00:23:16,580
languages and it worked very well

576
00:23:22,430 --> 00:23:20,039
however when we apply this to uh the

577
00:23:25,190 --> 00:23:22,440
varnish manuscript it still does not

578
00:23:26,750 --> 00:23:25,200
produce any kind of a readable

579
00:23:28,130 --> 00:23:26,760
decipherment

580
00:23:30,549 --> 00:23:28,140
when you're testing it with other

581
00:23:33,169 --> 00:23:30,559
languages are you testing it with

582
00:23:35,330 --> 00:23:33,179
ensipherments that you contrive or are

583
00:23:38,630 --> 00:23:35,340
you testing it with ciphers that already

584
00:23:41,870 --> 00:23:38,640
exist from those other languages

585
00:23:44,330 --> 00:23:41,880
no we tested it on a mass scale with uh

586
00:23:46,130 --> 00:23:44,340
with synthetic Cipher so computer

587
00:23:48,049 --> 00:23:46,140
generated ciphers

588
00:23:50,149 --> 00:23:48,059
but these are generated from the actual

589
00:23:52,250 --> 00:23:50,159
text in those languages

590
00:23:53,990 --> 00:23:52,260
you mentioned that there's two kinds of

591
00:23:55,730 --> 00:23:54,000
ciphers at least so far so there's

592
00:23:58,250 --> 00:23:55,740
substitution and then transposed or

593
00:24:00,169 --> 00:23:58,260
transposition what else is there so Pig

594
00:24:03,830 --> 00:24:00,179
Latin where you just you add some words

595
00:24:09,950 --> 00:24:07,270
no I think Pig Landing is like a game

596
00:24:11,450 --> 00:24:09,960
what I'm saying is that it seems like

597
00:24:12,950 --> 00:24:11,460
there are other methods that exist even

598
00:24:15,409 --> 00:24:12,960
if they're silly so there's

599
00:24:17,750 --> 00:24:15,419
transposition there's substitution is

600
00:24:19,730 --> 00:24:17,760
that primarily it or is there a seldom

601
00:24:22,549 --> 00:24:19,740
third

602
00:24:24,770 --> 00:24:22,559
well I would say even big Latin you can

603
00:24:28,310 --> 00:24:24,780
express it probably as some kind of

604
00:24:31,730 --> 00:24:28,320
substitution uh or and transposition

605
00:24:35,510 --> 00:24:31,740
right so so these serious ciphers like

606
00:24:37,730 --> 00:24:35,520
in Enigma is is basically again a

607
00:24:41,090 --> 00:24:37,740
combination of substitution and intense

608
00:24:42,649 --> 00:24:41,100
position uh actually Enigma is just pure

609
00:24:45,049 --> 00:24:42,659
substitution really there's no

610
00:24:46,909 --> 00:24:45,059
transposition there except that of

611
00:24:48,470 --> 00:24:46,919
course the spaces between words are

612
00:24:50,690 --> 00:24:48,480
removed

613
00:24:52,669 --> 00:24:50,700
well that's disappointing you're like

614
00:24:54,289 --> 00:24:52,679
okay great I've made a Headway I found

615
00:24:56,270 --> 00:24:54,299
out that it's Hebrew at least you're

616
00:24:58,909 --> 00:24:56,280
somewhat confident it's Hebrew then you

617
00:25:01,130 --> 00:24:58,919
say well okay let me devise some way of

618
00:25:03,350 --> 00:25:01,140
deciphering any substitution plus

619
00:25:05,510 --> 00:25:03,360
transposition combination it works on

620
00:25:07,430 --> 00:25:05,520
other ciphers great let me apply it here

621
00:25:09,350 --> 00:25:07,440
doesn't work so now what are you

622
00:25:11,750 --> 00:25:09,360
thinking and what's next

623
00:25:14,090 --> 00:25:11,760
well first of all I'm not confident that

624
00:25:17,149 --> 00:25:14,100
it is actually Hebrew all I can say is

625
00:25:20,750 --> 00:25:17,159
that out of those 400 languages that we

626
00:25:22,310 --> 00:25:20,760
had samples of this is the one that got

627
00:25:24,950 --> 00:25:22,320
the highest score

628
00:25:28,010 --> 00:25:24,960
so if I had to pick one of those 400

629
00:25:30,769 --> 00:25:28,020
then I would pick here but the language

630
00:25:33,710 --> 00:25:30,779
may actually not not be in that 400

631
00:25:35,450 --> 00:25:33,720
sample it may be a there's thousands of

632
00:25:37,850 --> 00:25:35,460
languages in the world

633
00:25:40,149 --> 00:25:37,860
and in addition it may not actually be

634
00:25:42,409 --> 00:25:40,159
actually any human language some people

635
00:25:45,590 --> 00:25:42,419
hypothesize that it's a made-up language

636
00:25:50,390 --> 00:25:45,600
like Esperanto can this thing

637
00:25:53,149 --> 00:25:50,400
so of course we were excited to see some

638
00:25:54,970 --> 00:25:53,159
kind of clear preference for one of the

639
00:25:56,769 --> 00:25:54,980
languages

640
00:26:01,010 --> 00:25:56,779
but

641
00:26:02,049 --> 00:26:01,020
and and we applied a kind of a

642
00:26:04,909 --> 00:26:02,059
scientific

643
00:26:06,649 --> 00:26:04,919
methodology to it so we reported those

644
00:26:09,409 --> 00:26:06,659
results and they are replicable if

645
00:26:11,450 --> 00:26:09,419
somebody else applies this to that

646
00:26:13,370 --> 00:26:11,460
sample they will find exactly the same

647
00:26:16,850 --> 00:26:13,380
thing but that doesn't mean that this

648
00:26:23,870 --> 00:26:20,029
so uh yeah

649
00:26:26,269 --> 00:26:23,880
if if I was really convinced that this

650
00:26:27,649 --> 00:26:26,279
was Hebrew I think the next thing I

651
00:26:29,990 --> 00:26:27,659
would have to do is to actually learn

652
00:26:32,450 --> 00:26:30,000
Hebrew because

653
00:26:36,669 --> 00:26:32,460
that would be the only way to decipher

654
00:26:40,130 --> 00:26:36,679
that complicated uh manuscript followers

655
00:26:43,190 --> 00:26:40,140
but of course uh I have a lot of other

656
00:26:45,409 --> 00:26:43,200
projects to do so I'm not going to study

657
00:26:48,230 --> 00:26:45,419
Hebrew for that purpose but there are

658
00:26:50,090 --> 00:26:48,240
many people that know Hebrew and uh

659
00:26:52,490 --> 00:26:50,100
and I'm sure if if this was really

660
00:26:53,390 --> 00:26:52,500
Hebrew they would be able to to decipher

661
00:26:56,230 --> 00:26:53,400
it

662
00:26:58,730 --> 00:26:56,240
and themselves so if someone watching

663
00:27:00,710 --> 00:26:58,740
speaks Hebrew and is a computer

664
00:27:03,230 --> 00:27:00,720
scientist and they also want to help

665
00:27:07,570 --> 00:27:03,240
what should they do contact you or is

666
00:27:11,990 --> 00:27:10,130
because I've been contacted by so many

667
00:27:14,390 --> 00:27:12,000
people I'm just gonna put on screen

668
00:27:17,409 --> 00:27:14,400
please contact crondrak here's his email

669
00:27:20,870 --> 00:27:17,419
address and his private phone number

670
00:27:24,049 --> 00:27:20,880
yeah I uh some of those people

671
00:27:26,330 --> 00:27:24,059
uh were actually experts in Hebrew and

672
00:27:28,909 --> 00:27:26,340
in computers and in ciphers

673
00:27:30,950 --> 00:27:28,919
and even they they could not make any

674
00:27:32,570 --> 00:27:30,960
progression so then why do you think

675
00:27:34,250 --> 00:27:32,580
that if you were to learn Hebrew would

676
00:27:38,269 --> 00:27:34,260
help

677
00:27:40,250 --> 00:27:38,279
no I said that if I really was 100 sure

678
00:27:43,370 --> 00:27:40,260
that it's Hebrew then that would

679
00:27:45,769 --> 00:27:43,380
definitely help to know Hebrew right my

680
00:27:48,289 --> 00:27:45,779
work is from a point of view of a

681
00:27:52,310 --> 00:27:48,299
computer scientist not from a point of

682
00:27:54,890 --> 00:27:52,320
view of a linguist or or a cryptographer

683
00:27:57,470 --> 00:27:54,900
so it's not as simple as saying identify

684
00:28:00,549 --> 00:27:57,480
the language then suggest the different

685
00:28:03,950 --> 00:28:00,559
rules so that is the substitution slash

686
00:28:07,549 --> 00:28:03,960
transposition combination and once you

687
00:28:10,610 --> 00:28:07,559
have that feed a text any text that is

688
00:28:12,529 --> 00:28:10,620
in that language and is some unknown

689
00:28:15,230 --> 00:28:12,539
substitution slash transposition

690
00:28:18,710 --> 00:28:15,240
combination and outputs it tells you yes

691
00:28:22,250 --> 00:28:18,720
or no it's not as simple as that

692
00:28:24,230 --> 00:28:22,260
so you know as I said the the main value

693
00:28:28,010 --> 00:28:24,240
of vines manuscript is that it forces

694
00:28:30,590 --> 00:28:28,020
you to come up with new methods that

695
00:28:31,730 --> 00:28:30,600
later may turn out to be useful for

696
00:28:34,669 --> 00:28:31,740
other things

697
00:28:37,970 --> 00:28:34,679
what we came up with is is a methodology

698
00:28:39,350 --> 00:28:37,980
for doing this and it we proved it in

699
00:28:40,970 --> 00:28:39,360
our paper

700
00:28:43,750 --> 00:28:40,980
that it works

701
00:28:47,390 --> 00:28:43,760
maybe at kind of 95

702
00:28:49,310 --> 00:28:47,400
accuracy if you take a language whatever

703
00:28:53,830 --> 00:28:49,320
language speak any language

704
00:28:58,730 --> 00:28:56,690
substitute letters for other symbols

705
00:29:01,430 --> 00:28:58,740
scramble them

706
00:29:03,890 --> 00:29:01,440
give it to our program it will decipher

707
00:29:06,649 --> 00:29:03,900
it with 95 percent like

708
00:29:08,630 --> 00:29:06,659
so that so that is proven and that is a

709
00:29:12,110 --> 00:29:08,640
replicable thing that

710
00:29:13,370 --> 00:29:12,120
was published in the paper

711
00:29:19,010 --> 00:29:13,380
but

712
00:29:20,210 --> 00:29:19,020
that it is actually a actual human

713
00:29:26,090 --> 00:29:20,220
language

714
00:29:27,950 --> 00:29:26,100
is being used for that purpose

715
00:29:31,970 --> 00:29:27,960
the fact that it doesn't work with

716
00:29:34,190 --> 00:29:31,980
Voynich suggests that Vonage is not

717
00:29:35,269 --> 00:29:34,200
written in Hebrew or any of those 400

718
00:29:37,430 --> 00:29:35,279
languages

719
00:29:39,230 --> 00:29:37,440
so you ended up testing it on all 400

720
00:29:43,010 --> 00:29:39,240
languages

721
00:29:46,730 --> 00:29:43,020
no we tested it on a smaller subset I

722
00:29:50,389 --> 00:29:48,710
um is it just computationally too

723
00:29:52,549 --> 00:29:50,399
difficult to do all of them like it

724
00:29:54,710 --> 00:29:52,559
takes up too much time can you not just

725
00:29:56,750 --> 00:29:54,720
tell the computer to run with it

726
00:29:58,909 --> 00:29:56,760
the problem is that you need to build

727
00:30:00,169 --> 00:29:58,919
What's called the language model for

728
00:30:03,529 --> 00:30:00,179
each language

729
00:30:06,130 --> 00:30:03,539
and for that you need a lot of texts and

730
00:30:08,810 --> 00:30:06,140
the European languages

731
00:30:11,570 --> 00:30:08,820
usually all of them have a lot of text

732
00:30:14,149 --> 00:30:11,580
like people write newspapers in them but

733
00:30:16,850 --> 00:30:14,159
if you pick languages that are very

734
00:30:19,430 --> 00:30:16,860
small or very exotic and it's very

735
00:30:21,970 --> 00:30:19,440
difficult to find any electronic text

736
00:30:25,970 --> 00:30:21,980
written in those languages

737
00:30:28,370 --> 00:30:25,980
so so then it's very difficult to

738
00:30:30,409 --> 00:30:28,380
to derive a language model from those

739
00:30:33,350 --> 00:30:30,419
texts because they are too small

740
00:30:36,470 --> 00:30:33,360
that was the reason

741
00:30:38,269 --> 00:30:36,480
yeah it's quite the conundrum

742
00:30:40,190 --> 00:30:38,279
at least you were able to develop some

743
00:30:42,470 --> 00:30:40,200
new techniques that can be applied to

744
00:30:43,730 --> 00:30:42,480
other problems have you made any other

745
00:30:46,130 --> 00:30:43,740
progress other than what you've just

746
00:30:49,990 --> 00:30:46,140
indicated

747
00:30:52,010 --> 00:30:50,000
so I mentioned that we worked on another

748
00:30:53,570 --> 00:30:52,020
undeciphered text how about we just

749
00:30:55,430 --> 00:30:53,580
transition to that and I'll come back

750
00:30:56,990 --> 00:30:55,440
and forth to the Voynich at different

751
00:30:58,970 --> 00:30:57,000
points why don't you tell us about the

752
00:31:01,730 --> 00:30:58,980
Dora Bella cipherite

753
00:31:03,470 --> 00:31:01,740
you know for me like I know there are

754
00:31:06,049 --> 00:31:03,480
people that just spend their all their

755
00:31:08,090 --> 00:31:06,059
lives on Vonage right they're like

756
00:31:10,130 --> 00:31:08,100
obsessed with Voynich but for me it was

757
00:31:12,649 --> 00:31:10,140
just one projects of many

758
00:31:14,870 --> 00:31:12,659
so after running chapter we decided that

759
00:31:17,990 --> 00:31:14,880
we've done everything we could with it

760
00:31:21,889 --> 00:31:18,000
we left it to other people to to puzzle

761
00:31:23,630 --> 00:31:21,899
over and there was another uh Cipher

762
00:31:25,549 --> 00:31:23,640
that caught my attention which is called

763
00:31:29,029 --> 00:31:25,559
the durabella cipher

764
00:31:30,649 --> 00:31:29,039
and this was written in 20th century we

765
00:31:35,690 --> 00:31:30,659
know who wrote it

766
00:31:40,430 --> 00:31:35,700
it was an English composer

767
00:31:43,370 --> 00:31:40,440
who wrote a postcard to his friend

768
00:31:46,310 --> 00:31:43,380
and that first part was was the cipher

769
00:31:47,750 --> 00:31:46,320
it included the cipher which is about 80

770
00:31:50,810 --> 00:31:47,760
characters

771
00:31:54,710 --> 00:31:50,820
in a kind of a strange script

772
00:31:56,810 --> 00:31:54,720
and that that postcard survived and was

773
00:32:01,370 --> 00:31:56,820
published after his death

774
00:32:05,870 --> 00:32:03,470
um and nobody has been able to decipher

775
00:32:07,909 --> 00:32:05,880
that short text

776
00:32:09,889 --> 00:32:07,919
so that's the durable a Cypher another

777
00:32:11,810 --> 00:32:09,899
undeciphered

778
00:32:15,769 --> 00:32:11,820
safer

779
00:32:19,190 --> 00:32:15,779
and now our approach was that maybe you

780
00:32:20,269 --> 00:32:19,200
know this is not a text any language

781
00:32:22,909 --> 00:32:20,279
text

782
00:32:26,330 --> 00:32:22,919
maybe this is just music because that

783
00:32:28,190 --> 00:32:26,340
guy was a composer so what happens what

784
00:32:29,269 --> 00:32:28,200
will happen if we try to decipher into

785
00:32:31,669 --> 00:32:29,279
music

786
00:32:35,090 --> 00:32:31,679
so we came up with algorithms and

787
00:32:38,090 --> 00:32:35,100
implemented programs that can take

788
00:32:41,269 --> 00:32:38,100
a short text like that

789
00:32:43,909 --> 00:32:41,279
uh well even I mean not tax but a short

790
00:32:48,130 --> 00:32:43,919
piece of music that is encoded in some

791
00:32:55,970 --> 00:32:51,830
uh and uh and that's what happened and

792
00:32:58,490 --> 00:32:55,980
we we published a paper that

793
00:33:01,190 --> 00:32:58,500
at the end produces a kind of a

794
00:33:02,690 --> 00:33:01,200
reconstruction of a Melody that is our

795
00:33:06,409 --> 00:33:02,700
best guess

796
00:33:07,669 --> 00:33:06,419
at the decipherment of that Cipher into

797
00:33:10,010 --> 00:33:07,679
music

798
00:33:12,230 --> 00:33:10,020
you said A peculiar statement about the

799
00:33:14,570 --> 00:33:12,240
Voynich manuscript that it may not be a

800
00:33:16,190 --> 00:33:14,580
human language now do you mean to say a

801
00:33:18,409 --> 00:33:16,200
language that large groups of people

802
00:33:21,470 --> 00:33:18,419
speak or that is an alien language like

803
00:33:24,769 --> 00:33:21,480
it's not a homo sapien

804
00:33:26,149 --> 00:33:24,779
it could be a made-up language right so

805
00:33:28,970 --> 00:33:26,159
you know that

806
00:33:31,669 --> 00:33:28,980
actually there exists languages that

807
00:33:35,210 --> 00:33:31,679
were invented like Esperanto

808
00:33:37,250 --> 00:33:35,220
and many languages like hundreds of

809
00:33:39,950 --> 00:33:37,260
languages have been invented this could

810
00:33:42,769 --> 00:33:39,960
be one of those a language that was

811
00:33:44,870 --> 00:33:42,779
never spoken by any Community but

812
00:33:47,570 --> 00:33:44,880
somebody just kind of made up

813
00:33:50,330 --> 00:33:47,580
a language and and it's possible anybody

814
00:33:52,190 --> 00:33:50,340
can do that in learn their own language

815
00:33:54,529 --> 00:33:52,200
I see so it's still it's a human

816
00:33:56,570 --> 00:33:54,539
language in the sense that it's made by

817
00:33:58,490 --> 00:33:56,580
a human but it's not a human language in

818
00:34:00,409 --> 00:33:58,500
the sense that it's not spoken by many

819
00:34:02,149 --> 00:34:00,419
people or even known about

820
00:34:03,409 --> 00:34:02,159
so it's not as if the alligator made up

821
00:34:04,970 --> 00:34:03,419
this language or some other extra

822
00:34:08,089 --> 00:34:04,980
dimensional entity made up the language

823
00:34:10,490 --> 00:34:08,099
or divinely inspired

824
00:34:12,530 --> 00:34:10,500
that's right the better word is probably

825
00:34:14,869 --> 00:34:12,540
natural language we say natural

826
00:34:19,310 --> 00:34:14,879
languages the languages that occur

827
00:34:22,310 --> 00:34:19,320
in in uh on the on the planet spoken by

828
00:34:25,190 --> 00:34:22,320
some community of people I see Stephen

829
00:34:27,050 --> 00:34:25,200
Bax is another professor who is no

830
00:34:29,510 --> 00:34:27,060
longer with us but he studied this

831
00:34:31,849 --> 00:34:29,520
manuscript and I'm curious if you can go

832
00:34:34,790 --> 00:34:31,859
through what his theory is on it are and

833
00:34:38,030 --> 00:34:34,800
then also your commentary on them

834
00:34:41,450 --> 00:34:38,040
well actually I'm not an expert in in

835
00:34:43,629 --> 00:34:41,460
history or or any other theories

836
00:34:46,550 --> 00:34:43,639
you know the the

837
00:34:48,770 --> 00:34:46,560
ultimate test of a theory is that it

838
00:34:50,389 --> 00:34:48,780
produces a decipherment

839
00:34:53,570 --> 00:34:50,399
right so

840
00:34:56,810 --> 00:34:53,580
uh as far as I know no reasonable

841
00:34:58,970 --> 00:34:56,820
decipherment has been produced by Dr Bax

842
00:35:03,050 --> 00:34:58,980
or anybody else

843
00:35:04,970 --> 00:35:03,060
so uh it's it's not a huge motivation to

844
00:35:07,970 --> 00:35:04,980
study somebody's method

845
00:35:10,069 --> 00:35:07,980
if that method has not actually worked

846
00:35:11,569 --> 00:35:10,079
but how does one go about the process of

847
00:35:14,450 --> 00:35:11,579
learning a language from a computational

848
00:35:20,329 --> 00:35:17,510
so you know everybody speaks a language

849
00:35:22,670 --> 00:35:20,339
that's the universal thing every human

850
00:35:24,950 --> 00:35:22,680
being they have their own native

851
00:35:26,270 --> 00:35:24,960
language plus the image speak other

852
00:35:28,970 --> 00:35:26,280
languages

853
00:35:30,710 --> 00:35:28,980
but the majority of people I I think

854
00:35:32,569 --> 00:35:30,720
they just learn very well their own

855
00:35:34,450 --> 00:35:32,579
native language and they learn it as

856
00:35:37,310 --> 00:35:34,460
children

857
00:35:40,130 --> 00:35:37,320
if you try to learn a language

858
00:35:42,470 --> 00:35:40,140
after you're like something somehow like

859
00:35:44,510 --> 00:35:42,480
10 years old then you'll find out that

860
00:35:46,910 --> 00:35:44,520
it actually becomes a different process

861
00:35:48,530 --> 00:35:46,920
it becomes more difficult and you

862
00:35:52,190 --> 00:35:48,540
actually have to school you have to go

863
00:35:53,329 --> 00:35:52,200
to school or study books or go on the

864
00:35:56,569 --> 00:35:53,339
internet

865
00:35:58,730 --> 00:35:56,579
and somebody teaches you a language this

866
00:36:00,650 --> 00:35:58,740
is not how children language

867
00:36:02,270 --> 00:36:00,660
so this is a big difference between the

868
00:36:05,390 --> 00:36:02,280
native language and the second language

869
00:36:06,890 --> 00:36:05,400
that we learn for example when I speak

870
00:36:09,650 --> 00:36:06,900
different you can probably tell that

871
00:36:12,530 --> 00:36:09,660
English is not my first language

872
00:36:14,870 --> 00:36:12,540
my first language is Polish

873
00:36:17,089 --> 00:36:14,880
so

874
00:36:19,730 --> 00:36:17,099
because it's not my native language and

875
00:36:22,609 --> 00:36:19,740
because I learned it as a teenager

876
00:36:26,210 --> 00:36:22,619
you can tell from my accent that uh that

877
00:36:28,390 --> 00:36:26,220
I'm not an native speaker right so uh so

878
00:36:30,890 --> 00:36:28,400
this already tells you something about

879
00:36:33,650 --> 00:36:30,900
what people call the language Instinct

880
00:36:35,089 --> 00:36:33,660
the the ability of people to acquire

881
00:36:40,010 --> 00:36:35,099
language

882
00:36:42,290 --> 00:36:40,020
now Linguistics is is uh is a

883
00:36:45,349 --> 00:36:42,300
science of

884
00:36:47,089 --> 00:36:45,359
of the language which deals with various

885
00:36:49,910 --> 00:36:47,099
aspects of the language and those

886
00:36:51,069 --> 00:36:49,920
include things like formatics and

887
00:36:54,670 --> 00:36:51,079
morphology

888
00:36:57,349 --> 00:36:54,680
grammar syntax semantics pragmatics

889
00:36:59,210 --> 00:36:57,359
acquisition many things what are

890
00:37:02,450 --> 00:36:59,220
pragmatics briefly sorry

891
00:37:04,970 --> 00:37:02,460
yeah pragmatics is probably what you are

892
00:37:08,270 --> 00:37:04,980
most interested yourself is

893
00:37:11,930 --> 00:37:08,280
is uh basically for example a sentiment

894
00:37:13,790 --> 00:37:11,940
analysis is pragmatics right if you if

895
00:37:15,349 --> 00:37:13,800
you deal with sentiment analysis you're

896
00:37:18,349 --> 00:37:15,359
not really interested in finding out

897
00:37:20,750 --> 00:37:18,359
what people say but what they feel about

898
00:37:22,609 --> 00:37:20,760
what they say right so that's that's

899
00:37:25,310 --> 00:37:22,619
what we call pragmatics it's not just

900
00:37:27,829 --> 00:37:25,320
about the message it's about all the

901
00:37:30,109 --> 00:37:27,839
other stuff upon it how do we feel about

902
00:37:32,829 --> 00:37:30,119
the message that sounds terribly

903
00:37:40,550 --> 00:37:37,010
is it no it does well it's it is

904
00:37:42,349 --> 00:37:40,560
difficult but uh but uh but it's doable

905
00:37:44,810 --> 00:37:42,359
and it's not the hardest part it's it is

906
00:37:46,790 --> 00:37:44,820
one of the tasks that people do and and

907
00:37:49,490 --> 00:37:46,800
we have programs now that are very good

908
00:37:50,990 --> 00:37:49,500
at it okay so continue on where you were

909
00:37:53,510 --> 00:37:51,000
please

910
00:37:55,370 --> 00:37:53,520
yeah when you say that it sounds uh very

911
00:37:58,190 --> 00:37:55,380
complicated is because it's hard to

912
00:38:00,650 --> 00:37:58,200
Define exactly what we mean by things

913
00:38:03,410 --> 00:38:00,660
like sentiment like me you mean like

914
00:38:05,870 --> 00:38:03,420
what do you mean sentiment like and then

915
00:38:07,010 --> 00:38:05,880
people say well are you angry are you

916
00:38:10,130 --> 00:38:07,020
happy

917
00:38:12,950 --> 00:38:10,140
are you sad

918
00:38:15,349 --> 00:38:12,960
and then how many feelings do we have

919
00:38:18,290 --> 00:38:15,359
well we have eight feelings

920
00:38:19,970 --> 00:38:18,300
really eight no maybe 12 right so this

921
00:38:22,490 --> 00:38:19,980
these are things that are very difficult

922
00:38:25,550 --> 00:38:22,500
to Define it's much easier to deal with

923
00:38:27,589 --> 00:38:25,560
things like letters or phonemes where we

924
00:38:29,210 --> 00:38:27,599
know exactly how many letters or phonies

925
00:38:32,390 --> 00:38:29,220
we have in the language

926
00:38:35,630 --> 00:38:32,400
and it's easier to write programs that

927
00:38:38,030 --> 00:38:35,640
deal with that yeah so I'll I'll give it

928
00:38:41,390 --> 00:38:38,040
a try so so you can imagine it's like a

929
00:38:44,450 --> 00:38:41,400
pipeline so you start with when you hear

930
00:38:46,069 --> 00:38:44,460
somebody speaking you start with what

931
00:38:47,930 --> 00:38:46,079
are the sounds of the language right

932
00:38:50,630 --> 00:38:47,940
that's phonetics

933
00:38:52,849 --> 00:38:50,640
and now once you've done that then

934
00:38:55,730 --> 00:38:52,859
you've tried to figure out where one

935
00:38:58,250 --> 00:38:55,740
word starts the the other ends right you

936
00:39:00,230 --> 00:38:58,260
want to see you want to identify the

937
00:39:02,270 --> 00:39:00,240
words because there is only a limited

938
00:39:05,930 --> 00:39:02,280
number of words

939
00:39:08,030 --> 00:39:05,940
and that's the what we call lexicon or

940
00:39:10,430 --> 00:39:08,040
lexicos

941
00:39:13,310 --> 00:39:10,440
and then you when you look at the words

942
00:39:15,230 --> 00:39:13,320
you see that they are made up of sounds

943
00:39:17,270 --> 00:39:15,240
or letters

944
00:39:19,089 --> 00:39:17,280
but they also made us something bigger

945
00:39:21,890 --> 00:39:19,099
which is called morphemes

946
00:39:24,349 --> 00:39:21,900
and uh and that's the stuff of

947
00:39:28,730 --> 00:39:24,359
morphology that's the study of

948
00:39:31,069 --> 00:39:28,740
morphology for example if uh if I say a

949
00:39:32,870 --> 00:39:31,079
word like ungrammaticality

950
00:39:35,510 --> 00:39:32,880
and then you can say well there are

951
00:39:37,010 --> 00:39:35,520
three parts of it the an the grammar and

952
00:39:40,130 --> 00:39:37,020
the ality

953
00:39:43,190 --> 00:39:40,140
and that's the morphology

954
00:39:46,370 --> 00:39:43,200
so so these are considered the kind of a

955
00:39:49,190 --> 00:39:46,380
low level low levels of language and as

956
00:39:51,829 --> 00:39:49,200
you go up it becomes more interesting so

957
00:39:53,630 --> 00:39:51,839
first of all how are words put together

958
00:39:56,810 --> 00:39:53,640
into sentences

959
00:39:59,450 --> 00:39:56,820
how is it that you can have sentences

960
00:40:01,250 --> 00:39:59,460
that you ask somebody is that a proper

961
00:40:03,589 --> 00:40:01,260
English sentence and they say yes or no

962
00:40:06,410 --> 00:40:03,599
they can tell even though they have no

963
00:40:08,569 --> 00:40:06,420
idea they haven't studied linguistics

964
00:40:10,849 --> 00:40:08,579
every native speaker can tell you if a

965
00:40:13,609 --> 00:40:10,859
sentence is grammatical or not

966
00:40:15,170 --> 00:40:13,619
that's the study that Noam Chomsky did

967
00:40:17,450 --> 00:40:15,180
in the 50s

968
00:40:18,829 --> 00:40:17,460
can we write a program that can tell a

969
00:40:21,589 --> 00:40:18,839
grammatical sentence from an

970
00:40:24,470 --> 00:40:21,599
ungrammatical sentence

971
00:40:26,270 --> 00:40:24,480
and on top of that on top of syntax is

972
00:40:29,089 --> 00:40:26,280
semantics which is about the meaning of

973
00:40:31,490 --> 00:40:29,099
words we can have perfectly Dramatical

974
00:40:33,650 --> 00:40:31,500
sentences that are meaningless

975
00:40:37,310 --> 00:40:33,660
and vice versa we can have meaningless

976
00:40:39,530 --> 00:40:37,320
utterances that are not grammatical

977
00:40:41,569 --> 00:40:39,540
and then on top of cement it is

978
00:40:44,210 --> 00:40:41,579
pragmatics which has all these things

979
00:40:47,390 --> 00:40:44,220
that are difficult to Define and that

980
00:40:49,790 --> 00:40:47,400
appear to you very complicated this

981
00:40:51,829 --> 00:40:49,800
Universal grammar of chomsky's it's true

982
00:40:53,150 --> 00:40:51,839
in the sense that you can create a

983
00:40:54,650 --> 00:40:53,160
program that can identify which

984
00:40:57,470 --> 00:40:54,660
sentences are grammatically correct and

985
00:41:02,210 --> 00:41:00,230
actually I don't think so I think that's

986
00:41:03,589 --> 00:41:02,220
what Chomsky tried to do all his life

987
00:41:08,150 --> 00:41:03,599
but

988
00:41:09,410 --> 00:41:08,160
it has not it has not been done as far

989
00:41:12,530 --> 00:41:09,420
as I know

990
00:41:15,770 --> 00:41:12,540
but at least that was the state of the

991
00:41:18,589 --> 00:41:15,780
earth about 10 years ago now the the the

992
00:41:21,650 --> 00:41:18,599
last few years we have seen the the

993
00:41:23,810 --> 00:41:21,660
neural language models appearing which

994
00:41:26,089 --> 00:41:23,820
are extremely effective

995
00:41:28,490 --> 00:41:26,099
and which as you know can produce a

996
00:41:31,130 --> 00:41:28,500
completely grammatical and text that

997
00:41:33,710 --> 00:41:31,140
also makes sense yeah so so by extension

998
00:41:35,930 --> 00:41:33,720
that means that these these uh programs

999
00:41:37,910 --> 00:41:35,940
can tell the difference between a

1000
00:41:39,530 --> 00:41:37,920
grammatical and ungrammatical centers

1001
00:41:41,089 --> 00:41:39,540
because they only produce grammatical

1002
00:41:43,490 --> 00:41:41,099
sentence

1003
00:41:49,370 --> 00:41:43,500
are there other Universal Concepts in

1004
00:41:55,670 --> 00:41:52,569
uh so the universal Concepts in language

1005
00:41:56,829 --> 00:41:55,680
are the things that are in every

1006
00:42:02,450 --> 00:41:56,839
language

1007
00:42:04,550 --> 00:42:02,460
there is something that almost all

1008
00:42:06,950 --> 00:42:04,560
languages possess but some languages

1009
00:42:09,290 --> 00:42:06,960
don't then it's not Universal

1010
00:42:12,410 --> 00:42:09,300
there is a whole area of linguistics

1011
00:42:15,230 --> 00:42:12,420
that is dealing with finding things that

1012
00:42:18,230 --> 00:42:15,240
are Universal in human languages

1013
00:42:20,990 --> 00:42:18,240
and I as far as I know there is a long

1014
00:42:22,910 --> 00:42:21,000
list of those things have you used any

1015
00:42:25,190 --> 00:42:22,920
machine learning or neural language

1016
00:42:27,430 --> 00:42:25,200
processing in the decipherment of the

1017
00:42:30,530 --> 00:42:27,440
Voynich

1018
00:42:33,950 --> 00:42:30,540
we did use machine learning but not

1019
00:42:37,130 --> 00:42:33,960
neural methods no the the

1020
00:42:40,069 --> 00:42:37,140
reason we didn't try we didn't use the

1021
00:42:42,170 --> 00:42:40,079
neural methods for decipherment

1022
00:42:44,810 --> 00:42:42,180
I I think you have some experience

1023
00:42:48,710 --> 00:42:44,820
already with these neural Bots is that

1024
00:42:50,750 --> 00:42:48,720
they can make sense of everything so for

1025
00:42:53,450 --> 00:42:50,760
example Google translate if you give it

1026
00:42:55,190 --> 00:42:53,460
a something that doesn't make sense it

1027
00:42:58,250 --> 00:42:55,200
will still translate it

1028
00:42:59,990 --> 00:42:58,260
into something that does obviously we

1029
00:43:02,750 --> 00:43:00,000
don't want something like that to be

1030
00:43:04,750 --> 00:43:02,760
applied to voynish manuscript because we

1031
00:43:08,089 --> 00:43:04,760
want to really know what's really there

1032
00:43:10,309 --> 00:43:08,099
not how to make sense out of it

1033
00:43:12,950 --> 00:43:10,319
in some way right

1034
00:43:14,510 --> 00:43:12,960
there isn't some way of identifying what

1035
00:43:16,670 --> 00:43:14,520
makes sense and what doesn't in the same

1036
00:43:18,050 --> 00:43:16,680
way that for some sentences you can

1037
00:43:20,329 --> 00:43:18,060
identify if it's grammatically correct

1038
00:43:22,069 --> 00:43:20,339
or incorrect like that program has not

1039
00:43:24,230 --> 00:43:22,079
been completely explicated like you

1040
00:43:25,910 --> 00:43:24,240
mentioned with Chomsky but maybe there's

1041
00:43:28,069 --> 00:43:25,920
huge progress there is there not

1042
00:43:29,870 --> 00:43:28,079
progress in saying the sentence makes

1043
00:43:33,170 --> 00:43:29,880
sense or not

1044
00:43:34,910 --> 00:43:33,180
that is much harder to do there is

1045
00:43:35,870 --> 00:43:34,920
progress yeah every year there is

1046
00:43:39,050 --> 00:43:35,880
progress

1047
00:43:40,670 --> 00:43:39,060
but we are still far from reaching that

1048
00:43:44,030 --> 00:43:40,680
point

1049
00:43:47,329 --> 00:43:44,040
you've seen that there's chat GPT and

1050
00:43:48,950 --> 00:43:47,339
there's open AIS gpt3 what's your

1051
00:43:52,730 --> 00:43:48,960
opinion of them are you excited by them

1052
00:43:59,809 --> 00:43:56,270
I am excited that those tools become

1053
00:44:02,030 --> 00:43:59,819
available but I'm also kind of worried

1054
00:44:03,109 --> 00:44:02,040
that people are too enthusiastic about

1055
00:44:06,829 --> 00:44:03,119
them

1056
00:44:10,130 --> 00:44:06,839
and they for me the problem is that

1057
00:44:13,010 --> 00:44:10,140
they are basically what somebody called

1058
00:44:15,230 --> 00:44:13,020
parrots right they they're parrots that

1059
00:44:18,349 --> 00:44:15,240
have heard a lot of language being

1060
00:44:20,750 --> 00:44:18,359
spoken everything that was ever written

1061
00:44:22,849 --> 00:44:20,760
and they are very good at repeating

1062
00:44:24,109 --> 00:44:22,859
putting together those sentences and

1063
00:44:26,450 --> 00:44:24,119
words together

1064
00:44:28,250 --> 00:44:26,460
but there is no really under no real

1065
00:44:32,210 --> 00:44:28,260
understanding underneath

1066
00:44:32,990 --> 00:44:32,220
those systems cannot tell us why they

1067
00:44:35,329 --> 00:44:33,000
think

1068
00:44:37,970 --> 00:44:35,339
the these things that they say are true

1069
00:44:39,829 --> 00:44:37,980
they basically repeating

1070
00:44:42,829 --> 00:44:39,839
the words that have been written

1071
00:44:45,829 --> 00:44:42,839
somewhere and rearranging

1072
00:44:46,970 --> 00:44:45,839
to be fair most people when they're

1073
00:44:48,470 --> 00:44:46,980
putting out something that's creative

1074
00:44:49,790 --> 00:44:48,480
they're just repeating what they've seen

1075
00:44:52,609 --> 00:44:49,800
and they're mixing it up and they

1076
00:44:54,349 --> 00:44:52,619
believe it to be absolutely new and also

1077
00:44:55,849 --> 00:44:54,359
just so you know there is something

1078
00:44:59,270 --> 00:44:55,859
creative about mixing up and then

1079
00:45:01,250 --> 00:44:59,280
presenting it and furthermore most

1080
00:45:02,630 --> 00:45:01,260
people maybe even all of us we don't

1081
00:45:04,430 --> 00:45:02,640
know the motivations like will

1082
00:45:06,230 --> 00:45:04,440
confabulate some reason for why we

1083
00:45:08,630 --> 00:45:06,240
created so and so like that's why the

1084
00:45:10,430 --> 00:45:08,640
whole field of psychoanalysis came about

1085
00:45:13,730 --> 00:45:10,440
because we don't know why we do what we

1086
00:45:15,230 --> 00:45:13,740
do we make up some reason so why does it

1087
00:45:17,809 --> 00:45:15,240
matter that the computer doesn't know

1088
00:45:19,450 --> 00:45:17,819
why it does what it's doing and that

1089
00:45:22,010 --> 00:45:19,460
it's quote unquote

1090
00:45:24,650 --> 00:45:22,020
repeating while mixing let's say mixing

1091
00:45:28,490 --> 00:45:26,450
I don't think it matters if you're

1092
00:45:30,829 --> 00:45:28,500
interested in a computer producing art

1093
00:45:31,730 --> 00:45:30,839
like writing a song or painting a

1094
00:45:34,069 --> 00:45:31,740
picture

1095
00:45:36,230 --> 00:45:34,079
but it doesn't matter if you rely on the

1096
00:45:37,690 --> 00:45:36,240
computer to tell you what the truth is

1097
00:45:41,690 --> 00:45:37,700
right

1098
00:45:43,250 --> 00:45:41,700
because if you don't if somebody cannot

1099
00:45:44,930 --> 00:45:43,260
explain to you why they believe

1100
00:45:48,770 --> 00:45:44,940
something is true

1101
00:45:50,750 --> 00:45:48,780
then how can you trust them

1102
00:45:54,710 --> 00:45:50,760
these are deep questions

1103
00:45:57,050 --> 00:45:54,720
so yeah what I find remarkable is that

1104
00:45:59,870 --> 00:45:57,060
you can just even a simple program

1105
00:46:01,790 --> 00:45:59,880
asking it to code this in Python code

1106
00:46:03,349 --> 00:46:01,800
something that does this in Python code

1107
00:46:06,109 --> 00:46:03,359
something that does this in autohotkey

1108
00:46:09,589 --> 00:46:06,119
or whatever it may be and it does it or

1109
00:46:11,690 --> 00:46:09,599
does it 90 the way there who the heck I

1110
00:46:13,609 --> 00:46:11,700
didn't think that that would be possible

1111
00:46:15,470 --> 00:46:13,619
for quite some time there's something

1112
00:46:17,089 --> 00:46:15,480
truthful about that in the sense that it

1113
00:46:19,069 --> 00:46:17,099
works like you can actually test if the

1114
00:46:21,530 --> 00:46:19,079
code works so that's a test of Truth

1115
00:46:23,089 --> 00:46:21,540
more so than a statement are you happy

1116
00:46:24,530 --> 00:46:23,099
about that or you feel like even that

1117
00:46:28,010 --> 00:46:24,540
old technology that could have been done

1118
00:46:34,790 --> 00:46:32,510
so well the program programming is a bit

1119
00:46:37,790 --> 00:46:34,800
different story right because

1120
00:46:40,390 --> 00:46:37,800
you can actually test programs right

1121
00:46:43,309 --> 00:46:40,400
so if some if you ask

1122
00:46:45,050 --> 00:46:43,319
whether it's a human or it's a bot to

1123
00:46:47,390 --> 00:46:45,060
write a program you can

1124
00:46:50,150 --> 00:46:47,400
you provide a specification then you can

1125
00:46:51,829 --> 00:46:50,160
go through the testing the test

1126
00:46:54,470 --> 00:46:51,839
procedure and find out if that program

1127
00:46:56,650 --> 00:46:54,480
really does what it does right so we

1128
00:47:00,470 --> 00:46:56,660
don't actually have to trust

1129
00:47:03,230 --> 00:47:00,480
the pro the trust anything we can just

1130
00:47:05,569 --> 00:47:03,240
test it right but if we don't if we

1131
00:47:07,250 --> 00:47:05,579
don't if if we don't have time to test

1132
00:47:10,370 --> 00:47:07,260
it then I would be wondering whether

1133
00:47:12,589 --> 00:47:10,380
it's a good idea to depend on such a

1134
00:47:15,109 --> 00:47:12,599
program

1135
00:47:18,170 --> 00:47:15,119
so going back to the voyage have you

1136
00:47:20,150 --> 00:47:18,180
thought about if it's composed of at

1137
00:47:22,990 --> 00:47:20,160
least one language like maybe there are

1138
00:47:30,349 --> 00:47:27,230
you know uh it could be it could be a

1139
00:47:34,190 --> 00:47:30,359
lot of things there you know the the you

1140
00:47:36,290 --> 00:47:34,200
can make these uh encryption systems as

1141
00:47:38,990 --> 00:47:36,300
complicated as you wish

1142
00:47:41,089 --> 00:47:39,000
so so it's all possible there is no

1143
00:47:42,950 --> 00:47:41,099
there is no limit there will be no limit

1144
00:47:45,470 --> 00:47:42,960
where we can say well we tried

1145
00:47:48,530 --> 00:47:45,480
everything and now we know it doesn't

1146
00:47:53,569 --> 00:47:48,540
make sense so it must be some kind of a

1147
00:47:56,990 --> 00:47:53,579
joke or some kind of random generator

1148
00:48:01,010 --> 00:47:57,000
but uh what is fascinating about Voynich

1149
00:48:04,630 --> 00:48:01,020
is that we can use it to actually

1150
00:48:08,750 --> 00:48:04,640
create new things right so

1151
00:48:09,950 --> 00:48:08,760
we take decipher and we create a Melody

1152
00:48:13,309 --> 00:48:09,960
right

1153
00:48:16,309 --> 00:48:13,319
and and many people take advantage and

1154
00:48:20,930 --> 00:48:16,319
they produce decipherments that

1155
00:48:23,809 --> 00:48:20,940
are like their own pieces of of art like

1156
00:48:25,930 --> 00:48:23,819
their own books the only problem is that

1157
00:48:29,450 --> 00:48:25,940
everybody produces a different one so

1158
00:48:32,089 --> 00:48:29,460
none of them can be actually correct but

1159
00:48:34,309 --> 00:48:32,099
it is still a creation

1160
00:48:37,130 --> 00:48:34,319
so I think that is that is very good

1161
00:48:39,470 --> 00:48:37,140
about varnish that it exists

1162
00:48:41,930 --> 00:48:39,480
have you thought about Voynich from less

1163
00:48:44,809 --> 00:48:41,940
of a computational perspective and more

1164
00:48:46,309 --> 00:48:44,819
just from a human motivation one what

1165
00:48:49,130 --> 00:48:46,319
the heck is this about why would someone

1166
00:48:50,990 --> 00:48:49,140
go through such lengths to decipher this

1167
00:48:52,190 --> 00:48:51,000
or maybe it's not even lengths like you

1168
00:48:53,990 --> 00:48:52,200
mentioned it could be something trivial

1169
00:48:57,410 --> 00:48:54,000
we're just overlooking like what are

1170
00:49:00,470 --> 00:48:57,420
their theories come up in your mind just

1171
00:49:03,650 --> 00:49:00,480
surmising just conjecture

1172
00:49:06,109 --> 00:49:03,660
yeah so one of of the more interesting

1173
00:49:09,829 --> 00:49:06,119
theories that I've encountered actually

1174
00:49:12,010 --> 00:49:09,839
comes from this U.S expert

1175
00:49:15,050 --> 00:49:12,020
on decipherment

1176
00:49:18,710 --> 00:49:15,060
at in the end he said

1177
00:49:21,890 --> 00:49:18,720
that he thinks this is a an artificial

1178
00:49:23,270 --> 00:49:21,900
language somebody created an artificial

1179
00:49:29,870 --> 00:49:23,280
language

1180
00:49:31,370 --> 00:49:29,880
well if that's the case then it's it's

1181
00:49:33,109 --> 00:49:31,380
very difficult it would be very

1182
00:49:35,809 --> 00:49:33,119
difficult to decipher it because we

1183
00:49:37,010 --> 00:49:35,819
don't know the principles of that

1184
00:49:41,210 --> 00:49:37,020
language

1185
00:49:43,790 --> 00:49:41,220
completely unpronounceable it's just a

1186
00:49:46,250 --> 00:49:43,800
sequence of symbols

1187
00:49:50,329 --> 00:49:46,260
um yeah so I think that's that that is

1188
00:49:53,930 --> 00:49:50,339
one theory that that is for me the most

1189
00:49:58,190 --> 00:49:56,809
what else have you heard that is at

1190
00:50:04,250 --> 00:49:58,200
least somewhat convincing maybe this

1191
00:50:09,530 --> 00:50:06,170
um

1192
00:50:11,089 --> 00:50:09,540
you know when I look anybody can look at

1193
00:50:14,329 --> 00:50:11,099
those illustrations they are on the web

1194
00:50:15,770 --> 00:50:14,339
right and if if you look for them for a

1195
00:50:19,370 --> 00:50:15,780
long time

1196
00:50:21,470 --> 00:50:19,380
sometimes I think this this somebody was

1197
00:50:24,349 --> 00:50:21,480
not quite

1198
00:50:26,809 --> 00:50:24,359
it wasn't it was a work of an expert it

1199
00:50:28,849 --> 00:50:26,819
was a work of somebody who actually

1200
00:50:30,650 --> 00:50:28,859
didn't know what they were doing and

1201
00:50:33,470 --> 00:50:30,660
just try to

1202
00:50:36,410 --> 00:50:33,480
create something like what they saw

1203
00:50:38,089 --> 00:50:36,420
before in other codices in other books

1204
00:50:40,010 --> 00:50:38,099
a little bit like a neural language

1205
00:50:41,510 --> 00:50:40,020
model that just looks a lot of things

1206
00:50:43,550 --> 00:50:41,520
yeah

1207
00:50:46,130 --> 00:50:43,560
she's a lot of reads a lot and then

1208
00:50:49,190 --> 00:50:46,140
produces something that looks like

1209
00:50:51,109 --> 00:50:49,200
it should make sense but it doesn't huh

1210
00:50:52,970 --> 00:50:51,119
that's interesting

1211
00:50:55,250 --> 00:50:52,980
we know that those language models can

1212
00:50:57,049 --> 00:50:55,260
be tricked to produce texts that just

1213
00:50:59,990 --> 00:50:57,059
seem to make sense about complete

1214
00:51:02,510 --> 00:51:00,000
nonsense right for example why it's good

1215
00:51:04,609 --> 00:51:02,520
to eat crushed glass right

1216
00:51:06,470 --> 00:51:04,619
it will give you all the all the reasons

1217
00:51:07,490 --> 00:51:06,480
for that why it is good to eat first

1218
00:51:09,530 --> 00:51:07,500
class

1219
00:51:11,270 --> 00:51:09,540
when it comes to the doorbell Cipher

1220
00:51:12,589 --> 00:51:11,280
there were some other people who came up

1221
00:51:15,290 --> 00:51:12,599
with decipherments I'm going to read

1222
00:51:18,049 --> 00:51:15,300
some right now starts Larks it's chaotic

1223
00:51:20,569 --> 00:51:18,059
but a cloak obscures my new letters a b

1224
00:51:22,910 --> 00:51:20,579
Alpha Beta that is the Greek letters

1225
00:51:24,650 --> 00:51:22,920
below I own the dark

1226
00:51:26,450 --> 00:51:24,660
and then there's others like why am I

1227
00:51:27,530 --> 00:51:26,460
very sad and so on I'm sure you've heard

1228
00:51:29,329 --> 00:51:27,540
these

1229
00:51:30,770 --> 00:51:29,339
yeah what do you mean yeah okay what do

1230
00:51:32,569 --> 00:51:30,780
you make of these

1231
00:51:35,630 --> 00:51:32,579
I've seen this before but it always

1232
00:51:37,970 --> 00:51:35,640
makes me laugh when I hear it's this for

1233
00:51:39,890 --> 00:51:37,980
me is complete nonsense

1234
00:51:42,349 --> 00:51:39,900
why why

1235
00:51:46,069 --> 00:51:42,359
well it is it is nonsense to imagine

1236
00:51:48,049 --> 00:51:46,079
that uh a distinguished uh English

1237
00:51:52,010 --> 00:51:48,059
composer would write something like that

1238
00:51:55,430 --> 00:51:52,020
to his uh love interest

1239
00:51:57,890 --> 00:51:55,440
if you were to decipher Voynich what

1240
00:52:02,109 --> 00:51:57,900
would be next for you no more ciphers or

1241
00:52:08,510 --> 00:52:05,690
you know I I think

1242
00:52:10,250 --> 00:52:08,520
did this happen I mean varnish has not

1243
00:52:13,190 --> 00:52:10,260
been decrypted but there was a very

1244
00:52:17,150 --> 00:52:13,200
interesting uh deciphering recently of

1245
00:52:19,010 --> 00:52:17,160
uh of actual Cipher which was called the

1246
00:52:20,809 --> 00:52:19,020
Zodiac Cypher I don't know issue right

1247
00:52:23,630 --> 00:52:20,819
right yeah

1248
00:52:26,150 --> 00:52:23,640
and that that is actually correct right

1249
00:52:29,270 --> 00:52:26,160
that that decide for me is not fake

1250
00:52:32,390 --> 00:52:29,280
it is actually a correct decipherment

1251
00:52:34,730 --> 00:52:32,400
so I I would probably ask that that

1252
00:52:36,049 --> 00:52:34,740
person about their feelings like what

1253
00:52:39,470 --> 00:52:36,059
how they feel

1254
00:52:41,630 --> 00:52:39,480
about cracking that Cipher is it like a

1255
00:52:43,430 --> 00:52:41,640
complete Bliss or is it like some kind

1256
00:52:46,130 --> 00:52:43,440
of disappointment that

1257
00:52:48,650 --> 00:52:46,140
and I put so much work into it and then

1258
00:52:51,349 --> 00:52:48,660
I find that this text is actually

1259
00:52:53,930 --> 00:52:51,359
kind of you know not interesting at all

1260
00:52:56,329 --> 00:52:53,940
it's like some kind of deranged mind

1261
00:53:00,650 --> 00:52:56,339
writing it

1262
00:53:03,170 --> 00:53:00,660
so uh yes you know there are one one

1263
00:53:04,790 --> 00:53:03,180
kind of tragedy is if you don't achieve

1264
00:53:07,549 --> 00:53:04,800
your goal and the other tragedy is you

1265
00:53:09,349 --> 00:53:07,559
if you do it you should go

1266
00:53:11,630 --> 00:53:09,359
yeah that's interesting hey let's get

1267
00:53:13,910 --> 00:53:11,640
philosophical here to me that means that

1268
00:53:16,250 --> 00:53:13,920
you have to enjoy the process more than

1269
00:53:18,109 --> 00:53:16,260
the state even though there's some end

1270
00:53:19,730 --> 00:53:18,119
States and that's supposedly driving the

1271
00:53:22,130 --> 00:53:19,740
process you have to fall in love with

1272
00:53:25,730 --> 00:53:22,140
the process because you may if you're

1273
00:53:28,430 --> 00:53:25,740
lucky and maybe unlucky reach that state

1274
00:53:32,569 --> 00:53:28,440
yeah absolutely and this is something

1275
00:53:34,670 --> 00:53:32,579
that I do feel about problems in

1276
00:53:39,410 --> 00:53:34,680
computational linguistics that

1277
00:53:41,809 --> 00:53:39,420
and that I love doing this stuff and uh

1278
00:53:43,690 --> 00:53:41,819
I just would be able to do this you know

1279
00:53:47,270 --> 00:53:43,700
for free

1280
00:53:49,370 --> 00:53:47,280
and if because it's it's such such huge

1281
00:53:53,150 --> 00:53:49,380
fun to do this

1282
00:53:56,210 --> 00:53:53,160
but the Vonage was just one of the

1283
00:53:59,569 --> 00:53:56,220
projects that I got interested in and uh

1284
00:54:02,030 --> 00:53:59,579
I learned from the project I I got some

1285
00:54:04,430 --> 00:54:02,040
experience from that project that I

1286
00:54:07,250 --> 00:54:04,440
think made me a better scientist

1287
00:54:09,710 --> 00:54:07,260
so that I can apply this experience to

1288
00:54:11,690 --> 00:54:09,720
the projects that are not that actually

1289
00:54:14,030 --> 00:54:11,700
do have a solution

1290
00:54:16,069 --> 00:54:14,040
well right now and we are very excited

1291
00:54:18,049 --> 00:54:16,079
to be working on semantics on lexical

1292
00:54:21,770 --> 00:54:18,059
semantics and

1293
00:54:24,950 --> 00:54:21,780
and we are proposing uh we you know we

1294
00:54:27,290 --> 00:54:24,960
are finding things that other scientists

1295
00:54:29,809 --> 00:54:27,300
find Maybe

1296
00:54:33,530 --> 00:54:29,819
uh controversial right

1297
00:54:35,270 --> 00:54:33,540
but the huge the huge advantage of of

1298
00:54:38,390 --> 00:54:35,280
the work that we do is that we can

1299
00:54:40,790 --> 00:54:38,400
actually provide proofs mathematical

1300
00:54:43,010 --> 00:54:40,800
proofs of of what we do

1301
00:54:45,890 --> 00:54:43,020
and this is this gives us the

1302
00:54:47,870 --> 00:54:45,900
satisfaction of of actually being

1303
00:54:50,690 --> 00:54:47,880
certain that we are doing something

1304
00:54:53,510 --> 00:54:50,700
right because we can prove it

1305
00:54:55,609 --> 00:54:53,520
going back to loving the road more than

1306
00:54:57,589 --> 00:54:55,619
where you're going I feel the same with

1307
00:54:59,690 --> 00:54:57,599
this podcast it's about theories of

1308
00:55:01,849 --> 00:54:59,700
everything and the physics sense so my

1309
00:55:04,069 --> 00:55:01,859
background is in mathematical physics

1310
00:55:07,490 --> 00:55:04,079
and a part of me I feel like I'll be

1311
00:55:10,250 --> 00:55:07,500
extremely disappointed if I encounter or

1312
00:55:12,650 --> 00:55:10,260
if we discover as people as scientists

1313
00:55:14,630 --> 00:55:12,660
the theory of everything there is

1314
00:55:16,730 --> 00:55:14,640
something that's terribly fun about

1315
00:55:19,510 --> 00:55:16,740
learning it and investigating so I don't

1316
00:55:23,170 --> 00:55:21,829
I don't think you have to worry about

1317
00:55:26,150 --> 00:55:23,180
that

1318
00:55:28,790 --> 00:55:26,160
personally I think

1319
00:55:30,950 --> 00:55:28,800
you know looking at how the universe is

1320
00:55:32,809 --> 00:55:30,960
constructed I'm pretty sure it has some

1321
00:55:35,569 --> 00:55:32,819
built-in mechanisms

1322
00:55:37,549 --> 00:55:35,579
so that we can never actually figure it

1323
00:55:41,030 --> 00:55:37,559
out completely

1324
00:55:43,430 --> 00:55:41,040
what gives you that intuition

1325
00:55:46,849 --> 00:55:43,440
well you know the the

1326
00:55:49,370 --> 00:55:46,859
yeah yeah you know like this when you

1327
00:55:52,730 --> 00:55:49,380
talk about theories of errors in you you

1328
00:55:56,750 --> 00:55:52,740
obviously talk to physicists so uh and

1329
00:55:58,190 --> 00:55:56,760
that deal with quantum mechanics and

1330
00:56:00,650 --> 00:55:58,200
things like that

1331
00:56:04,490 --> 00:56:00,660
and how there are certain principles

1332
00:56:06,950 --> 00:56:04,500
that we can prove that we'll never know

1333
00:56:10,849 --> 00:56:06,960
the truth right like we'll never know

1334
00:56:13,730 --> 00:56:10,859
where the particular particle is what is

1335
00:56:14,510 --> 00:56:13,740
the its exact location and speed and so

1336
00:56:18,349 --> 00:56:14,520
on

1337
00:56:21,710 --> 00:56:18,359
and this uh this just this is for me an

1338
00:56:24,710 --> 00:56:21,720
indication that these things are

1339
00:56:30,530 --> 00:56:24,720
constructed in such a way that we will

1340
00:56:34,270 --> 00:56:32,510
all right well that's hopeful but also

1341
00:56:36,410 --> 00:56:34,280
dismaying

1342
00:56:40,010 --> 00:56:36,420
at least it's both and not one without

1343
00:56:42,470 --> 00:56:40,020
the other so about the zodiac

1344
00:56:44,510 --> 00:56:42,480
if that's a substitution Cipher was it a

1345
00:56:45,950 --> 00:56:44,520
substitution and transposition or just

1346
00:56:50,630 --> 00:56:45,960
substitution

1347
00:56:53,450 --> 00:56:50,640
position and it was a very tricky dance

1348
00:56:55,190 --> 00:56:53,460
position too yeah why is that and would

1349
00:56:58,809 --> 00:56:55,200
your method have worked on the Zodiac

1350
00:57:04,670 --> 00:57:01,849
no the the battle would not work on

1351
00:57:06,829 --> 00:57:04,680
zodiac because the Assumption of our

1352
00:57:09,290 --> 00:57:06,839
methods was that we know where the words

1353
00:57:10,609 --> 00:57:09,300
are so in varnish there are spaces

1354
00:57:12,829 --> 00:57:10,619
between words

1355
00:57:15,230 --> 00:57:12,839
and we made this assumption that this is

1356
00:57:17,030 --> 00:57:15,240
not just to confuse but they are

1357
00:57:19,609 --> 00:57:17,040
actually words right

1358
00:57:22,069 --> 00:57:19,619
now in the Zodiac Cipher there was no

1359
00:57:23,390 --> 00:57:22,079
spaces between words

1360
00:57:25,910 --> 00:57:23,400
so

1361
00:57:28,609 --> 00:57:25,920
although it is possible to kind of

1362
00:57:30,530 --> 00:57:28,619
hypothesis where the spaces are that

1363
00:57:33,470 --> 00:57:30,540
method that particular method would not

1364
00:57:36,410 --> 00:57:33,480
work on zodiac

1365
00:57:40,069 --> 00:57:36,420
and the method used to crack the Zodiac

1366
00:57:45,470 --> 00:57:40,079
Cipher can that method or methods be

1367
00:57:50,690 --> 00:57:48,349
I actually you know I don't think so I I

1368
00:57:53,630 --> 00:57:50,700
think the the the key of the

1369
00:57:57,049 --> 00:57:53,640
decipherment in that case was just

1370
00:57:59,690 --> 00:57:57,059
finding the a specific pattern right

1371
00:58:02,510 --> 00:57:59,700
of transposition

1372
00:58:05,030 --> 00:58:02,520
so it was not any kind of cool new

1373
00:58:07,430 --> 00:58:05,040
Theory that is general and can be

1374
00:58:08,750 --> 00:58:07,440
applied to various things it was just

1375
00:58:12,349 --> 00:58:08,760
kind of a

1376
00:58:13,309 --> 00:58:12,359
stroke of lack there that like trial and

1377
00:58:15,349 --> 00:58:13,319
error

1378
00:58:17,390 --> 00:58:15,359
it is always a child in there when you

1379
00:58:18,589 --> 00:58:17,400
do actual decipherment but what I mean

1380
00:58:22,309 --> 00:58:18,599
is that

1381
00:58:26,150 --> 00:58:22,319
there is no method behind it that can be

1382
00:58:28,370 --> 00:58:26,160
generalized and applied to other things

1383
00:58:29,990 --> 00:58:28,380
how it's Ai and maybe this is a term

1384
00:58:32,150 --> 00:58:30,000
that you don't want to use but how has

1385
00:58:34,250 --> 00:58:32,160
AI aided your field so instead of saying

1386
00:58:36,109 --> 00:58:34,260
AI then reference the a specific model

1387
00:58:37,849 --> 00:58:36,119
like Gans I've changed my field because

1388
00:58:42,470 --> 00:58:37,859
of social supervised learning in the

1389
00:58:50,809 --> 00:58:47,329
yeah so you cannot uh you cannot can you

1390
00:58:52,809 --> 00:58:50,819
avoid the word neural nowadays when you

1391
00:58:54,650 --> 00:58:52,819
talk about language

1392
00:58:58,910 --> 00:58:54,660
understanding

1393
00:59:01,309 --> 00:58:58,920
uh it's it's a it's a powerful uh new

1394
00:59:04,130 --> 00:59:01,319
tool and everybody is very very excited

1395
00:59:06,589 --> 00:59:04,140
about it including myself

1396
00:59:08,230 --> 00:59:06,599
so it of course it changed everything

1397
00:59:11,990 --> 00:59:08,240
because

1398
00:59:14,630 --> 00:59:12,000
what the story of language processing is

1399
00:59:17,089 --> 00:59:14,640
that it started from a kind of a

1400
00:59:19,910 --> 00:59:17,099
symbolic processing

1401
00:59:22,730 --> 00:59:19,920
and then moved into the machine learning

1402
00:59:25,789 --> 00:59:22,740
stage and then evolved into the the

1403
00:59:28,549 --> 00:59:25,799
neural methods which we use nowadays so

1404
00:59:31,309 --> 00:59:28,559
what is exciting about it is that every

1405
00:59:34,010 --> 00:59:31,319
few years you have a new Revolution and

1406
00:59:36,289 --> 00:59:34,020
a new methods and and we make constant

1407
00:59:37,670 --> 00:59:36,299
progress to the point that some people

1408
00:59:39,770 --> 00:59:37,680
think

1409
00:59:42,170 --> 00:59:39,780
that the problem of language has been

1410
00:59:44,210 --> 00:59:42,180
solved but it's not the case

1411
00:59:46,309 --> 00:59:44,220
sorry that the problem of language has

1412
00:59:48,349 --> 00:59:46,319
been solved the program of language

1413
00:59:50,930 --> 00:59:48,359
understanding has been solved that that

1414
00:59:53,510 --> 00:59:50,940
we can basically now have programs that

1415
00:59:55,130 --> 00:59:53,520
will do every language related task that

1416
00:59:57,829 --> 00:59:55,140
we want

1417
01:00:01,190 --> 00:59:57,839
and it's not true who thinks that that's

1418
01:00:06,710 --> 01:00:04,010
well you know when I read these uh

1419
01:00:08,809 --> 01:00:06,720
articles about the the neural Bots that

1420
01:00:13,210 --> 01:00:08,819
can you know write

1421
01:00:16,789 --> 01:00:13,220
uh newspaper articles or or

1422
01:00:18,650 --> 01:00:16,799
compose Symphonies or something that

1423
01:00:21,410 --> 01:00:18,660
sometimes you get an impression that

1424
01:00:23,270 --> 01:00:21,420
well we're done right we can just leave

1425
01:00:25,309 --> 01:00:23,280
it all to the computers

1426
01:00:26,390 --> 01:00:25,319
and they will they will do everything

1427
01:00:30,170 --> 01:00:26,400
for us

1428
01:00:33,829 --> 01:00:30,180
but what I tell my students is that you

1429
01:00:36,230 --> 01:00:33,839
really want to become to be educated to

1430
01:00:37,130 --> 01:00:36,240
be somebody who cannot be replaced by a

1431
01:00:39,530 --> 01:00:37,140
computer

1432
01:00:42,289 --> 01:00:39,540
and I guarantee you that they will never

1433
01:00:49,309 --> 01:00:42,299
be able to replace the most important

1434
01:00:54,829 --> 01:00:51,710
are you sure about that what is it about

1435
01:00:56,630 --> 01:00:54,839
human creativity that a machine can't

1436
01:00:58,789 --> 01:00:56,640
replicate by the way I'm not being

1437
01:01:00,410 --> 01:00:58,799
skeptical I just don't know I'm curious

1438
01:01:02,329 --> 01:01:00,420
what your thoughts are because since

1439
01:01:04,569 --> 01:01:02,339
you're in this field

1440
01:01:08,809 --> 01:01:04,579
well that's exactly what you said

1441
01:01:11,289 --> 01:01:08,819
machine cannot replicate creativity and

1442
01:01:13,609 --> 01:01:11,299
replication are are opposite things

1443
01:01:15,170 --> 01:01:13,619
creativity is doing something that has

1444
01:01:16,849 --> 01:01:15,180
not been done before

1445
01:01:19,789 --> 01:01:16,859
of course you can say well it's just

1446
01:01:23,569 --> 01:01:19,799
kind of building on what was before but

1447
01:01:26,270 --> 01:01:23,579
it's not replicating it's not parroting

1448
01:01:29,930 --> 01:01:26,280
it's creating something new based on a

1449
01:01:34,430 --> 01:01:32,809
well there's this old joke of if you

1450
01:01:36,230 --> 01:01:34,440
want to create an apple pie from scratch

1451
01:01:38,089 --> 01:01:36,240
you have to first create the universe

1452
01:01:39,650 --> 01:01:38,099
it's like well did you get it from the

1453
01:01:41,150 --> 01:01:39,660
farm no I bought it from this okay but

1454
01:01:43,069 --> 01:01:41,160
even if you had it from the farm that

1455
01:01:44,750 --> 01:01:43,079
you grow the dirt did you well yes okay

1456
01:01:45,530 --> 01:01:44,760
but did you make the cow and so on and

1457
01:01:48,890 --> 01:01:45,540
so on

1458
01:01:50,870 --> 01:01:48,900
in a sense whatever we think of as new

1459
01:01:52,549 --> 01:01:50,880
it's so tricky like it depends on what

1460
01:01:55,549 --> 01:01:52,559
what the heck are we defining as novel

1461
01:01:57,289 --> 01:01:55,559
as creative and I'm sure if we could

1462
01:01:58,910 --> 01:01:57,299
look into our brain with a certain

1463
01:02:00,470 --> 01:01:58,920
amount of resolution and we had the

1464
01:02:02,809 --> 01:02:00,480
correct model if it even could be

1465
01:02:04,490 --> 01:02:02,819
modeled computationally but regardless

1466
01:02:06,349 --> 01:02:04,500
maybe there's some non-computational

1467
01:02:08,990 --> 01:02:06,359
model if it can even be modeled

1468
01:02:11,450 --> 01:02:09,000
quote-unquote model the point is that I

1469
01:02:13,130 --> 01:02:11,460
imagine it's conceivable to me that what

1470
01:02:15,170 --> 01:02:13,140
we think of as outputting something

1471
01:02:17,329 --> 01:02:15,180
creative is something that is

1472
01:02:19,730 --> 01:02:17,339
algorithmic like I'm not set on this but

1473
01:02:21,890 --> 01:02:19,740
it's conceivable and if that's the case

1474
01:02:23,630 --> 01:02:21,900
then I don't see why a computer can't do

1475
01:02:25,609 --> 01:02:23,640
it now whether or not a computer can

1476
01:02:27,890 --> 01:02:25,619
feel and understand what it's doing like

1477
01:02:30,230 --> 01:02:27,900
that's a separate problem but the actual

1478
01:02:31,970 --> 01:02:30,240
output I don't see an in principle

1479
01:02:33,650 --> 01:02:31,980
reason why it can't be done and I'm

1480
01:02:35,210 --> 01:02:33,660
telling you this as a romantic like I

1481
01:02:37,010 --> 01:02:35,220
don't want this to be done but I see

1482
01:02:39,170 --> 01:02:37,020
more and more like aspects that we

1483
01:02:41,270 --> 01:02:39,180
thought computers could not do

1484
01:02:43,010 --> 01:02:41,280
it couldn't beat us a chest and then but

1485
01:02:45,410 --> 01:02:43,020
well I can't create art oh my gosh can

1486
01:02:47,750 --> 01:02:45,420
it create art and can't compose music oh

1487
01:02:50,510 --> 01:02:47,760
my gosh can it confuse different musical

1488
01:02:52,010 --> 01:02:50,520
styles so it keeps encroaching

1489
01:02:53,870 --> 01:02:52,020
encroaching into these areas that we

1490
01:02:55,849 --> 01:02:53,880
thought this were just exclusive to

1491
01:02:57,470 --> 01:02:55,859
humans well what's left for us is like

1492
01:02:59,390 --> 01:02:57,480
physical dexterity and that seems to be

1493
01:03:01,670 --> 01:02:59,400
it so far my question is are you

1494
01:03:03,530 --> 01:03:01,680
confident in the statement that there's

1495
01:03:05,690 --> 01:03:03,540
something special about human creativity

1496
01:03:08,510 --> 01:03:05,700
I would like that to be the cake I want

1497
01:03:12,309 --> 01:03:08,520
to be convinced of that

1498
01:03:15,170 --> 01:03:12,319
well first of all when you said uh

1499
01:03:17,210 --> 01:03:15,180
that something cannot be done you cannot

1500
01:03:20,089 --> 01:03:17,220
demonstrate that something cannot be

1501
01:03:22,910 --> 01:03:20,099
done you can only demonstrate that

1502
01:03:27,410 --> 01:03:22,920
something can be done by doing it right

1503
01:03:29,750 --> 01:03:27,420
so I I will not be able or I don't think

1504
01:03:33,589 --> 01:03:29,760
anybody would be able to demonstrate

1505
01:03:37,490 --> 01:03:33,599
that computers cannot do something

1506
01:03:40,490 --> 01:03:37,500
but I am a computer scientist I I've you

1507
01:03:43,190 --> 01:03:40,500
know I programmed a lot

1508
01:03:47,049 --> 01:03:43,200
I work with computers a lot and I know

1509
01:03:49,910 --> 01:03:47,059
that the computers are good at doing

1510
01:03:52,250 --> 01:03:49,920
repeatedly certain things

1511
01:03:54,289 --> 01:03:52,260
and repeating patterns that already

1512
01:03:56,870 --> 01:03:54,299
exist right

1513
01:03:59,450 --> 01:03:56,880
you you cannot have an algorithm to

1514
01:04:00,170 --> 01:03:59,460
create something that does not exist I

1515
01:04:02,809 --> 01:04:00,180
mean

1516
01:04:05,510 --> 01:04:02,819
there is novel that is Meaningful of

1517
01:04:07,490 --> 01:04:05,520
course you can create novel things you

1518
01:04:10,549 --> 01:04:07,500
can create chaos right

1519
01:04:12,410 --> 01:04:10,559
you can create a random generator then

1520
01:04:15,289 --> 01:04:12,420
this sequence of randomly generated

1521
01:04:18,829 --> 01:04:15,299
numbers is unique is it novel

1522
01:04:22,069 --> 01:04:18,839
no because it doesn't make sense

1523
01:04:24,650 --> 01:04:22,079
are you afraid of where AI may be or are

1524
01:04:27,470 --> 01:04:24,660
you more hopeful

1525
01:04:30,109 --> 01:04:27,480
I think it's a serious issue and we have

1526
01:04:32,630 --> 01:04:30,119
to think about it you know because

1527
01:04:35,510 --> 01:04:32,640
the danger I see is that people will

1528
01:04:39,710 --> 01:04:35,520
trust those programs too much

1529
01:04:41,930 --> 01:04:39,720
and they uh we build them and we are

1530
01:04:43,430 --> 01:04:41,940
responsible for telling them what we

1531
01:04:46,190 --> 01:04:43,440
want them to do

1532
01:04:48,829 --> 01:04:46,200
if we don't do this right they may do

1533
01:04:50,630 --> 01:04:48,839
surprising things that we never actually

1534
01:04:53,270 --> 01:04:50,640
anticipated

1535
01:04:56,450 --> 01:04:53,280
I think the key thing is that we want

1536
01:04:58,910 --> 01:04:56,460
these things to be transparent we want

1537
01:05:02,089 --> 01:04:58,920
to know if they tell us

1538
01:05:04,370 --> 01:05:02,099
a statement then we want to know why

1539
01:05:07,370 --> 01:05:04,380
they think the statement is true

1540
01:05:09,230 --> 01:05:07,380
we want them to provide a proof of

1541
01:05:12,410 --> 01:05:09,240
something that they state

1542
01:05:13,609 --> 01:05:12,420
obviously they are not at this level yet

1543
01:05:15,349 --> 01:05:13,619
right

1544
01:05:18,950 --> 01:05:15,359
for example

1545
01:05:20,450 --> 01:05:18,960
they can write basically history books

1546
01:05:22,370 --> 01:05:20,460
right let's

1547
01:05:25,010 --> 01:05:22,380
we don't know whether they are

1548
01:05:27,190 --> 01:05:25,020
hallucinating or is it actual facts

1549
01:05:29,630 --> 01:05:27,200
they're talking about

1550
01:05:31,430 --> 01:05:29,640
so there must be some way of them

1551
01:05:34,549 --> 01:05:31,440
providing evidence

1552
01:05:36,829 --> 01:05:34,559
of what they are saying is true

1553
01:05:40,609 --> 01:05:36,839
like put references when you make a

1554
01:05:42,349 --> 01:05:40,619
statement like exactly exactly yeah so

1555
01:05:43,970 --> 01:05:42,359
we I've been talking to students

1556
01:05:46,130 --> 01:05:43,980
recently

1557
01:05:50,030 --> 01:05:46,140
about

1558
01:05:53,450 --> 01:05:50,040
what is true right like if

1559
01:05:56,349 --> 01:05:53,460
how can we decide if if a sentence is

1560
01:05:59,450 --> 01:05:56,359
true or false me

1561
01:06:01,849 --> 01:05:59,460
and and the fact is that

1562
01:06:04,190 --> 01:06:01,859
you know some people say Everything's

1563
01:06:05,630 --> 01:06:04,200
Relative you know like some people think

1564
01:06:08,630 --> 01:06:05,640
this is true and the other people you

1565
01:06:12,230 --> 01:06:08,640
think this is true I what I what I want

1566
01:06:15,410 --> 01:06:12,240
to the students to do is to decide

1567
01:06:17,510 --> 01:06:15,420
first what is the speaker what's the

1568
01:06:20,569 --> 01:06:17,520
author of the utterance if they think

1569
01:06:23,109 --> 01:06:20,579
it's true or not and this is non-trivial

1570
01:06:25,970 --> 01:06:23,119
but if they can establish that the the

1571
01:06:27,950 --> 01:06:25,980
author of the utterance or sentence

1572
01:06:30,650 --> 01:06:27,960
believes it's true

1573
01:06:32,990 --> 01:06:30,660
then it is true with respect to that

1574
01:06:35,109 --> 01:06:33,000
person right so we can say this is a

1575
01:06:38,930 --> 01:06:35,119
true statement according to this person

1576
01:06:41,750 --> 01:06:38,940
and it it is then kind of clear

1577
01:06:44,930 --> 01:06:41,760
that this is some kind of evidence based

1578
01:06:47,990 --> 01:06:44,940
on somebody's belief

1579
01:06:51,289 --> 01:06:48,000
so I do believe we can tell whether a

1580
01:06:54,109 --> 01:06:51,299
statement is true or false modular

1581
01:06:56,089 --> 01:06:54,119
the author of the statement

1582
01:06:57,589 --> 01:06:56,099
except in the case of AI like in the

1583
01:07:00,230 --> 01:06:57,599
case of people we can because they have

1584
01:07:02,450 --> 01:07:00,240
intentions but AI is no

1585
01:07:04,069 --> 01:07:02,460
currently no is there a subfield in

1586
01:07:05,990 --> 01:07:04,079
computer science that's dedicated to

1587
01:07:08,029 --> 01:07:06,000
this problem how did the machine come

1588
01:07:09,289 --> 01:07:08,039
upon this decision can it explain the

1589
01:07:11,210 --> 01:07:09,299
reasons

1590
01:07:13,190 --> 01:07:11,220
yeah many people are working on that

1591
01:07:15,170 --> 01:07:13,200
because many people have realized that

1592
01:07:17,510 --> 01:07:15,180
this is what we need in order to be able

1593
01:07:21,049 --> 01:07:17,520
to use those tools

1594
01:07:22,910 --> 01:07:21,059
and what's that field called or subfield

1595
01:07:24,950 --> 01:07:22,920
well the field that I worked in is

1596
01:07:28,789 --> 01:07:24,960
called natural language processing

1597
01:07:32,809 --> 01:07:28,799
uh and that's sometimes considered to be

1598
01:07:35,569 --> 01:07:32,819
a synonym of computational linguistics

1599
01:07:38,390 --> 01:07:35,579
is there a name for when you're

1600
01:07:40,910 --> 01:07:38,400
specifically trying to

1601
01:07:42,890 --> 01:07:40,920
pry open that black box and then pull

1602
01:07:45,710 --> 01:07:42,900
out something that is understandable to

1603
01:07:50,529 --> 01:07:48,589
like how did it make the decision

1604
01:07:53,510 --> 01:07:50,539
the word that I've heard uses

1605
01:07:56,329 --> 01:07:53,520
interpretability right so you want to

1606
01:07:59,270 --> 01:07:56,339
have a program that not just does the

1607
01:08:03,049 --> 01:07:59,280
job but is also interpretable so it can

1608
01:08:06,250 --> 01:08:03,059
be we can interpret why it does the job

1609
01:08:10,250 --> 01:08:06,260
as it does so the current

1610
01:08:12,470 --> 01:08:10,260
non-interpretability of AI is that what

1611
01:08:14,690 --> 01:08:12,480
you see as its greatest threat or do you

1612
01:08:16,430 --> 01:08:14,700
see that like you've heard strong Ai and

1613
01:08:19,249 --> 01:08:16,440
you've heard of The Singularity and that

1614
01:08:21,470 --> 01:08:19,259
machines may turn on humans or that

1615
01:08:23,269 --> 01:08:21,480
other people may use like if you invert

1616
01:08:26,749 --> 01:08:23,279
certain parameters then a drug that was

1617
01:08:28,970 --> 01:08:26,759
that a machine developed to produce a

1618
01:08:30,530 --> 01:08:28,980
drug that was helpful can be turned to

1619
01:08:32,269 --> 01:08:30,540
produce a drug that's extremely potent

1620
01:08:35,150 --> 01:08:32,279
and deleterious do you see the

1621
01:08:36,890 --> 01:08:35,160
non-interpretability of machines as the

1622
01:08:38,209 --> 01:08:36,900
greatest issue that we have right now or

1623
01:08:39,829 --> 01:08:38,219
is somehow connected to all those other

1624
01:08:41,930 --> 01:08:39,839
issues

1625
01:08:44,209 --> 01:08:41,940
I don't know if it's the greatest issue

1626
01:08:46,490 --> 01:08:44,219
but it's an important issue another

1627
01:08:49,370 --> 01:08:46,500
important issue is the so-called bias

1628
01:08:51,890 --> 01:08:49,380
right the these language models are

1629
01:08:54,349 --> 01:08:51,900
trained on texts that have been written

1630
01:08:57,110 --> 01:08:54,359
by people that are biased right and they

1631
01:08:59,930 --> 01:08:57,120
become biased themselves obviously we

1632
01:09:02,390 --> 01:08:59,940
don't want that to be guided by such

1633
01:09:04,249 --> 01:09:02,400
kind of texts

1634
01:09:06,890 --> 01:09:04,259
there's a phrase that you wrote down

1635
01:09:09,829 --> 01:09:06,900
English orthography is not close to

1636
01:09:11,930 --> 01:09:09,839
Optimal correct can you explain firstly

1637
01:09:14,269 --> 01:09:11,940
what orthography is and then take us

1638
01:09:17,390 --> 01:09:14,279
through that phrase

1639
01:09:21,110 --> 01:09:17,400
yes so orthography is the way we write

1640
01:09:22,490 --> 01:09:21,120
language so English exists primarily as

1641
01:09:25,070 --> 01:09:22,500
the spoken thing

1642
01:09:26,390 --> 01:09:25,080
but we also write it down like as every

1643
01:09:29,090 --> 01:09:26,400
language and

1644
01:09:32,150 --> 01:09:29,100
the orthograph is the way we write down

1645
01:09:35,930 --> 01:09:32,160
the sounds and as you may know English

1646
01:09:38,030 --> 01:09:35,940
doesn't have a very good orthography

1647
01:09:40,729 --> 01:09:38,040
well it doesn't seem to be good because

1648
01:09:42,349 --> 01:09:40,739
it's very hard to learn and people that

1649
01:09:44,570 --> 01:09:42,359
learn English they make a lot of

1650
01:09:47,090 --> 01:09:44,580
spelling errors and even native speakers

1651
01:09:49,370 --> 01:09:47,100
find it difficult to write down words

1652
01:09:54,050 --> 01:09:49,380
that they speak

1653
01:09:57,290 --> 01:09:54,060
so a Noam Chomsky had that a kind of a

1654
01:10:00,530 --> 01:09:57,300
statement that English orthography is

1655
01:10:03,530 --> 01:10:00,540
near optimal right it's close to Optimal

1656
01:10:07,550 --> 01:10:03,540
even though it appears not to be

1657
01:10:09,709 --> 01:10:07,560
so we had the projects when we when we

1658
01:10:12,050 --> 01:10:09,719
kind of showed that

1659
01:10:16,070 --> 01:10:12,060
it actually is not optimal it's not

1660
01:10:26,030 --> 01:10:19,729
and so that's the essence of that paper

1661
01:10:33,590 --> 01:10:29,290
Chomsky had ever had very good reasons

1662
01:10:35,350 --> 01:10:33,600
for saying what what he said but uh you

1663
01:10:39,470 --> 01:10:35,360
know in science

1664
01:10:42,410 --> 01:10:39,480
our job is to question everything right

1665
01:10:46,430 --> 01:10:42,420
and that's what we did in that project

1666
01:10:49,970 --> 01:10:46,440
we we wanted to question that statement

1667
01:10:52,689 --> 01:10:49,980
which seems to be nowadays accepted as

1668
01:10:56,270 --> 01:10:52,699
Truth by everybody

1669
01:10:59,090 --> 01:10:56,280
and to show that to to provide evidence

1670
01:11:00,709 --> 01:10:59,100
for that we we wrote programs and we did

1671
01:11:04,070 --> 01:11:00,719
such simulations

1672
01:11:07,130 --> 01:11:04,080
and we published this to show that

1673
01:11:10,430 --> 01:11:07,140
it is not actually optimal it is it is

1674
01:11:11,450 --> 01:11:10,440
not close to Optima could be much better

1675
01:11:14,450 --> 01:11:11,460
um

1676
01:11:15,229 --> 01:11:14,460
Yeah so basically that that's the point

1677
01:11:17,090 --> 01:11:15,239
here

1678
01:11:18,890 --> 01:11:17,100
what was chomsky's reasons for

1679
01:11:20,870 --> 01:11:18,900
suggesting it was optimal because as you

1680
01:11:23,270 --> 01:11:20,880
pointed out it seems on the face that

1681
01:11:26,090 --> 01:11:23,280
it's clear it's not like the word tough

1682
01:11:28,070 --> 01:11:26,100
is with an F but it ends with GH it

1683
01:11:29,570 --> 01:11:28,080
seems clear that it's not so Chomsky

1684
01:11:31,070 --> 01:11:29,580
must have had some reasons and like you

1685
01:11:32,990 --> 01:11:31,080
mentioned he had good reasons what were

1686
01:11:37,850 --> 01:11:33,000
they and then what was his response to

1687
01:11:44,390 --> 01:11:41,810
yes so Chomsky was when he wrote this in

1688
01:11:47,209 --> 01:11:44,400
the 60s he was going against the the

1689
01:11:49,310 --> 01:11:47,219
consensus right which was that English

1690
01:11:51,530 --> 01:11:49,320
orthography is is

1691
01:11:54,110 --> 01:11:51,540
bad okay

1692
01:11:57,290 --> 01:11:54,120
and he questioned that and he said no

1693
01:11:59,630 --> 01:11:57,300
it's actually near optimal

1694
01:12:01,910 --> 01:11:59,640
and it would take a lot of time to go

1695
01:12:04,310 --> 01:12:01,920
into those arguments which are which are

1696
01:12:06,770 --> 01:12:04,320
reasonable

1697
01:12:09,229 --> 01:12:06,780
however

1698
01:12:12,709 --> 01:12:09,239
every every there is more to it right

1699
01:12:15,350 --> 01:12:12,719
every everything can be interpreted in

1700
01:12:19,570 --> 01:12:15,360
different ways and

1701
01:12:23,510 --> 01:12:19,580
the the main assumption uh that is

1702
01:12:27,290 --> 01:12:23,520
not spoken is that our writing system in

1703
01:12:29,270 --> 01:12:27,300
English is based on a history of English

1704
01:12:33,530 --> 01:12:29,280
and other languages

1705
01:12:36,169 --> 01:12:33,540
for example a lot of English at some

1706
01:12:39,410 --> 01:12:36,179
point was very influenced by French

1707
01:12:41,649 --> 01:12:39,420
about a thousand years ago and and that

1708
01:12:46,430 --> 01:12:41,659
influenced the spelling of English

1709
01:12:48,890 --> 01:12:46,440
now even if we if we could change the

1710
01:12:51,290 --> 01:12:48,900
orthography of English

1711
01:12:53,689 --> 01:12:51,300
to something better if if there is

1712
01:12:56,330 --> 01:12:53,699
something better then that wouldn't be

1713
01:12:59,209 --> 01:12:56,340
practically possible because people are

1714
01:13:00,229 --> 01:12:59,219
just used to the way as it is written

1715
01:13:02,390 --> 01:13:00,239
right now

1716
01:13:04,790 --> 01:13:02,400
and besides English is spoken in many

1717
01:13:06,649 --> 01:13:04,800
different languages in countries and

1718
01:13:09,169 --> 01:13:06,659
those countries would never agree on on

1719
01:13:12,770 --> 01:13:09,179
a new system

1720
01:13:14,330 --> 01:13:12,780
so so in a sense uh Chomsky was right

1721
01:13:15,770 --> 01:13:14,340
about so-called morphological

1722
01:13:19,330 --> 01:13:15,780
consistency

1723
01:13:22,189 --> 01:13:19,340
that words that have the same morphemes

1724
01:13:24,050 --> 01:13:22,199
which are pronounced differently should

1725
01:13:26,750 --> 01:13:24,060
have the same representation for the

1726
01:13:27,890 --> 01:13:26,760
Murphy that representation shouldn't

1727
01:13:29,510 --> 01:13:27,900
change

1728
01:13:31,490 --> 01:13:29,520
but there is also something called

1729
01:13:33,229 --> 01:13:31,500
phonetic consistency and you gave

1730
01:13:36,290 --> 01:13:33,239
example of that

1731
01:13:37,810 --> 01:13:36,300
and that is just not good right there

1732
01:13:40,610 --> 01:13:37,820
are just too many

1733
01:13:43,310 --> 01:13:40,620
arbitrary solutions that reflect the

1734
01:13:44,390 --> 01:13:43,320
pronunciation that as it was 500 years

1735
01:13:46,550 --> 01:13:44,400
ago

1736
01:13:48,470 --> 01:13:46,560
for example they were tough as you said

1737
01:13:51,770 --> 01:13:48,480
was actually pronounced with a consonant

1738
01:13:54,350 --> 01:13:51,780
at the end 500 years ago

1739
01:13:56,750 --> 01:13:54,360
no not tough but there are things like

1740
01:13:58,610 --> 01:13:56,760
though for example which also ends with

1741
01:14:00,890 --> 01:13:58,620
GH there's morphological consistency

1742
01:14:03,890 --> 01:14:00,900
phonetic consistency and then there's

1743
01:14:06,709 --> 01:14:03,900
orthographical optimality can you place

1744
01:14:09,950 --> 01:14:06,719
numbers on those like you can say this

1745
01:14:12,370 --> 01:14:09,960
language is 90 optimal orthographically

1746
01:14:14,689 --> 01:14:12,380
and at 50

1747
01:14:17,209 --> 01:14:14,699
morphologically consistent can you

1748
01:14:20,570 --> 01:14:17,219
actually Place numbers on them

1749
01:14:23,750 --> 01:14:20,580
yeah so let me give it right so for

1750
01:14:25,149 --> 01:14:23,760
example finish is considered an

1751
01:14:28,490 --> 01:14:25,159
extremely good

1752
01:14:33,110 --> 01:14:28,500
orthography is completely consistent you

1753
01:14:34,610 --> 01:14:33,120
know in all uh kind of aspects

1754
01:14:37,870 --> 01:14:34,620
um

1755
01:14:40,189 --> 01:14:37,880
some languages are like

1756
01:14:43,189 --> 01:14:40,199
Croatian for example

1757
01:14:46,910 --> 01:14:43,199
is a the orthography was created under

1758
01:14:49,669 --> 01:14:46,920
the principle right as you speak so that

1759
01:14:52,610 --> 01:14:49,679
has this consistency that

1760
01:14:54,229 --> 01:14:52,620
you can just uh you could you don't you

1761
01:14:55,790 --> 01:14:54,239
never make spelling mistakes you just

1762
01:14:57,770 --> 01:14:55,800
write as you speak

1763
01:14:59,649 --> 01:14:57,780
sorry which language was based like that

1764
01:15:03,169 --> 01:14:59,659
that sounds interesting

1765
01:15:04,910 --> 01:15:03,179
that's uh the the server it used to be

1766
01:15:07,570 --> 01:15:04,920
server creation now these are separate

1767
01:15:11,689 --> 01:15:07,580
languages but it still applies to it

1768
01:15:14,290 --> 01:15:11,699
Croatia Knight now Spanish which many

1769
01:15:18,110 --> 01:15:14,300
people are familiar with this is another

1770
01:15:21,590 --> 01:15:18,120
type of language where you know always

1771
01:15:24,169 --> 01:15:21,600
know how to read something you may you

1772
01:15:27,229 --> 01:15:24,179
may still make spelling mistakes

1773
01:15:29,090 --> 01:15:27,239
but you will never pronounce a written

1774
01:15:31,850 --> 01:15:29,100
word in a wrong way

1775
01:15:34,490 --> 01:15:31,860
so that's another type of consistency

1776
01:15:35,990 --> 01:15:34,500
the English doesn't have either of those

1777
01:15:38,450 --> 01:15:36,000
right you can

1778
01:15:40,250 --> 01:15:38,460
you as a native speaker will probably

1779
01:15:43,070 --> 01:15:40,260
make mistakes unless you have a spell

1780
01:15:46,370 --> 01:15:43,080
checker even though you know perfectly

1781
01:15:49,189 --> 01:15:46,380
well how to pronounce a word

1782
01:15:51,470 --> 01:15:49,199
and me as a second language learner of

1783
01:15:54,530 --> 01:15:51,480
English I will encounter words that I

1784
01:15:57,290 --> 01:15:54,540
just don't know how to pronounce right

1785
01:16:00,050 --> 01:15:57,300
so it is definitely a problem in English

1786
01:16:01,689 --> 01:16:00,060
but other languages are even more

1787
01:16:03,770 --> 01:16:01,699
difficult like the Japanese

1788
01:16:06,290 --> 01:16:03,780
orthographic system is even more

1789
01:16:09,530 --> 01:16:06,300
difficult than English

1790
01:16:12,590 --> 01:16:09,540
I'm curious if English stands out as

1791
01:16:15,350 --> 01:16:12,600
best or worse in some Metric and if so

1792
01:16:18,350 --> 01:16:15,360
which for instance I heard that English

1793
01:16:20,870 --> 01:16:18,360
can convey a complex sentence second

1794
01:16:22,729 --> 01:16:20,880
best something like that and Mandarin is

1795
01:16:24,590 --> 01:16:22,739
first you can think of it as a simple

1796
01:16:26,209 --> 01:16:24,600
language as one that a child may just

1797
01:16:28,250 --> 01:16:26,219
come up with on their own and it goes

1798
01:16:30,470 --> 01:16:28,260
ooh and ah and the complexity of what

1799
01:16:31,850 --> 01:16:30,480
they can convey is small and so somehow

1800
01:16:33,470 --> 01:16:31,860
there's some way of measuring that I

1801
01:16:35,149 --> 01:16:33,480
don't know I don't know the the actual

1802
01:16:36,890 --> 01:16:35,159
terminology I just heard this and I

1803
01:16:38,810 --> 01:16:36,900
heard that English is actually pretty

1804
01:16:40,910 --> 01:16:38,820
great it's second in the world and

1805
01:16:42,229 --> 01:16:40,920
Chinese or sorry or Mandarin is best at

1806
01:16:44,149 --> 01:16:42,239
that but anyway the point is I just

1807
01:16:47,030 --> 01:16:44,159
heard this so what is English great ad

1808
01:16:52,130 --> 01:16:50,149
yes so so English English and Chinese

1809
01:16:53,930 --> 01:16:52,140
have something in common which is that

1810
01:16:57,470 --> 01:16:53,940
they are analytic languages so

1811
01:16:59,930 --> 01:16:57,480
morphology in English is is very basic

1812
01:17:01,430 --> 01:16:59,940
compared to languages like Spanish or

1813
01:17:04,610 --> 01:17:01,440
polish

1814
01:17:06,290 --> 01:17:04,620
oh and in in uh in Chinese it's even

1815
01:17:08,510 --> 01:17:06,300
more simple basically there is no

1816
01:17:11,330 --> 01:17:08,520
morphology at all

1817
01:17:15,110 --> 01:17:11,340
so in that in that sense these

1818
01:17:16,030 --> 01:17:15,120
analytical languages uh reach some kind

1819
01:17:18,310 --> 01:17:16,040
of

1820
01:17:23,689 --> 01:17:18,320
maximum at some

1821
01:17:31,070 --> 01:17:27,530
I I know that um

1822
01:17:32,750 --> 01:17:31,080
English is if you compare things written

1823
01:17:34,970 --> 01:17:32,760
in different languages

1824
01:17:37,850 --> 01:17:34,980
sometimes you see on products like 20

1825
01:17:40,910 --> 01:17:37,860
languages with the same message

1826
01:17:42,649 --> 01:17:40,920
the English text will probably be one of

1827
01:17:44,930 --> 01:17:42,659
the shortest ones

1828
01:17:47,090 --> 01:17:44,940
so I I think this is maybe something

1829
01:17:49,090 --> 01:17:47,100
you're referring to that

1830
01:17:52,790 --> 01:17:49,100
that it can actually

1831
01:17:56,330 --> 01:17:52,800
convey the same message with fewer

1832
01:17:58,070 --> 01:17:56,340
letters or fewer symbols

1833
01:17:59,689 --> 01:17:58,080
reminds me of this joke someone was

1834
01:18:01,370 --> 01:17:59,699
translating I think it's I think this

1835
01:18:03,770 --> 01:18:01,380
actually happened I think it was from

1836
01:18:05,330 --> 01:18:03,780
hideo Kojima who's a video game creator

1837
01:18:07,090 --> 01:18:05,340
and he was on stage she speaks Japanese

1838
01:18:09,709 --> 01:18:07,100
and he says he goes

1839
01:18:12,350 --> 01:18:09,719
It goes on for like 20 seconds 30

1840
01:18:13,370 --> 01:18:12,360
seconds and the translator comes he says

1841
01:18:14,990 --> 01:18:13,380
thank you

1842
01:18:16,669 --> 01:18:15,000
laughs

1843
01:18:17,570 --> 01:18:16,679
you're like that's not what he's like

1844
01:18:19,250 --> 01:18:17,580
just

1845
01:18:20,750 --> 01:18:19,260
if you don't if you're lazy or you've

1846
01:18:23,209 --> 01:18:20,760
forgotten that's fine but just there's

1847
01:18:26,510 --> 01:18:23,219
no way that's all of what he said yeah

1848
01:18:29,090 --> 01:18:26,520
yeah well I I actually lived in Japan

1849
01:18:31,250 --> 01:18:29,100
for a while so this is actually the

1850
01:18:33,709 --> 01:18:31,260
issue of pragmatics right

1851
01:18:36,290 --> 01:18:33,719
uh language human language is not just

1852
01:18:39,050 --> 01:18:36,300
exchanging messages it's there's a lot

1853
01:18:42,950 --> 01:18:39,060
of for example related to politeness

1854
01:18:46,430 --> 01:18:42,960
uh and in in in Japanese you spend a lot

1855
01:18:49,250 --> 01:18:46,440
of time just being polite in addition to

1856
01:18:51,229 --> 01:18:49,260
passing a message ah like sand sand at

1857
01:18:53,510 --> 01:18:51,239
the end of a person's name is that take

1858
01:18:56,030 --> 01:18:53,520
a note I Am lower than you or respect

1859
01:18:58,250 --> 01:18:56,040
yeah there is a lot more tools for

1860
01:18:59,450 --> 01:18:58,260
expressing this kind of relationships in

1861
01:19:02,510 --> 01:18:59,460
Japanese

1862
01:19:05,270 --> 01:19:02,520
uh to give a simpler example in in

1863
01:19:07,790 --> 01:19:05,280
Spanish for example you can refer to

1864
01:19:10,850 --> 01:19:07,800
somebody as two or you

1865
01:19:13,550 --> 01:19:10,860
or usted which is like sir

1866
01:19:16,430 --> 01:19:13,560
but in English it's that that doesn't

1867
01:19:18,830 --> 01:19:16,440
exist you just call you everyone

1868
01:19:20,810 --> 01:19:18,840
so that makes things simpler

1869
01:19:23,169 --> 01:19:20,820
do you know who Larry David is from

1870
01:19:28,910 --> 01:19:23,179
Seinfeld

1871
01:19:30,770 --> 01:19:28,920
David the Creator he said that when

1872
01:19:32,870 --> 01:19:30,780
Caesar was being assassinated by Brutus

1873
01:19:34,729 --> 01:19:32,880
that Brutus said something with the two

1874
01:19:36,290 --> 01:19:34,739
and then then Larry David said that was

1875
01:19:38,510 --> 01:19:36,300
too informal for an assassination you

1876
01:19:40,870 --> 01:19:38,520
should be saying instead

1877
01:19:46,130 --> 01:19:43,910
to end this you did your Master's thesis

1878
01:19:48,709 --> 01:19:46,140
on a theoretical evaluation on selected

1879
01:19:50,870 --> 01:19:48,719
backtracking algorithms so how has your

1880
01:19:53,030 --> 01:19:50,880
perspective on that subject since the

1881
01:19:54,830 --> 01:19:53,040
writing of that thesis changed how is it

1882
01:19:57,830 --> 01:19:54,840
developed

1883
01:20:00,110 --> 01:19:57,840
yeah so I I did this this is part of

1884
01:20:04,610 --> 01:20:00,120
what's called artificial intelligence

1885
01:20:06,110 --> 01:20:04,620
but it's a very formal thing like uh

1886
01:20:08,990 --> 01:20:06,120
constraint called constraint

1887
01:20:11,090 --> 01:20:09,000
satisfaction and what I liked about it

1888
01:20:12,669 --> 01:20:11,100
is that you can actually prove something

1889
01:20:15,470 --> 01:20:12,679
I mean

1890
01:20:17,330 --> 01:20:15,480
unlike in pure Linguistics you can never

1891
01:20:20,290 --> 01:20:17,340
prove anything you can just argue about

1892
01:20:23,390 --> 01:20:20,300
it and then some people will disagree

1893
01:20:25,790 --> 01:20:23,400
but I didn't stay in that area because I

1894
01:20:26,930 --> 01:20:25,800
wanted to work with language that I love

1895
01:20:31,070 --> 01:20:26,940
language

1896
01:20:33,290 --> 01:20:31,080
it's very hard to prove anything because

1897
01:20:36,290 --> 01:20:33,300
there are always exceptions

1898
01:20:37,910 --> 01:20:36,300
but now after though all those years I'm

1899
01:20:40,669 --> 01:20:37,920
coming back to the point that I think

1900
01:20:43,550 --> 01:20:40,679
that I can actually use the language of

1901
01:20:47,090 --> 01:20:43,560
mathematics to describe human language

1902
01:20:49,910 --> 01:20:47,100
and I find this very exciting so I I

1903
01:20:53,750 --> 01:20:49,920
hope to be able to prove things

1904
01:20:56,870 --> 01:20:53,760
and then be actually

1905
01:20:58,850 --> 01:20:56,880
safe in saying that

1906
01:21:00,649 --> 01:20:58,860
I'm I'm saying the right thing I'm

1907
01:21:02,209 --> 01:21:00,659
saying the truth because it has been

1908
01:21:04,130 --> 01:21:02,219
proven

1909
01:21:06,410 --> 01:21:04,140
what's one of the more out there

1910
01:21:07,790 --> 01:21:06,420
theories of the Voynich manuscript as to

1911
01:21:09,530 --> 01:21:07,800
what it's about what it contains

1912
01:21:11,030 --> 01:21:09,540
information on that you don't believe in

1913
01:21:13,310 --> 01:21:11,040
but you find interesting maybe even

1914
01:21:16,070 --> 01:21:13,320
plausible

1915
01:21:17,149 --> 01:21:16,080
so there was this hilarious paper and

1916
01:21:19,729 --> 01:21:17,159
somebody

1917
01:21:22,070 --> 01:21:19,739
trying to show that the language of

1918
01:21:23,810 --> 01:21:22,080
vines is actually Lodge Bank

1919
01:21:28,490 --> 01:21:23,820
I don't know if you've heard about it

1920
01:21:32,030 --> 01:21:28,500
it's it's a it's an invented language

1921
01:21:34,430 --> 01:21:32,040
and and this paper showed to me that you

1922
01:21:37,010 --> 01:21:34,440
can actually uh

1923
01:21:39,649 --> 01:21:37,020
provide evidence for anything for any

1924
01:21:42,290 --> 01:21:39,659
language if it's a large band that was

1925
01:21:44,810 --> 01:21:42,300
invented in the 20th century somebody

1926
01:21:46,729 --> 01:21:44,820
wrote Vonage manuscript in the 15th

1927
01:21:49,370 --> 01:21:46,739
century in that language

1928
01:21:50,750 --> 01:21:49,380
then that means you can basically argue

1929
01:21:54,110 --> 01:21:50,760
for anything

1930
01:21:56,270 --> 01:21:54,120
and that again shows the value of if you

1931
01:21:58,189 --> 01:21:56,280
can actually prove something and in the

1932
01:22:00,350 --> 01:21:58,199
case of the orange manuscript the proof

1933
01:22:01,390 --> 01:22:00,360
would be actually in the pudding which

1934
01:22:04,550 --> 01:22:01,400
means

1935
01:22:07,430 --> 01:22:04,560
deciphering it into some kind of

1936
01:22:09,830 --> 01:22:07,440
text that made sense

1937
01:22:12,530 --> 01:22:09,840
do you think it will be deciphered in

1938
01:22:18,649 --> 01:22:16,310
no I I hope it will be I hope it will

1939
01:22:22,130 --> 01:22:18,659
buy wooden Bet On It

1940
01:22:23,870 --> 01:22:22,140
you know the people said uh before in

1941
01:22:25,970 --> 01:22:23,880
the in the history people often said

1942
01:22:26,990 --> 01:22:25,980
something will never be done and it was

1943
01:22:29,870 --> 01:22:27,000
done

1944
01:22:31,490 --> 01:22:29,880
when I first heard about the Zodiac

1945
01:22:35,090 --> 01:22:31,500
Cipher

1946
01:22:37,669 --> 01:22:35,100
I thought no that's not ever never gonna

1947
01:22:41,030 --> 01:22:37,679
be deciphered because it's probably just

1948
01:22:43,310 --> 01:22:41,040
random night and then it turns out that

1949
01:22:45,950 --> 01:22:43,320
it was deciphered so

1950
01:22:47,990 --> 01:22:45,960
that's a lesson for us

1951
01:22:49,669 --> 01:22:48,000
meaning in the case of the zodiac you

1952
01:22:51,709 --> 01:22:49,679
thought that it was gibberish but he

1953
01:22:53,030 --> 01:22:51,719
didn't actually write anything it's not

1954
01:22:54,770 --> 01:22:53,040
something that was deciphered it's just

1955
01:22:56,930 --> 01:22:54,780
symbols

1956
01:22:58,970 --> 01:22:56,940
yeah I thought it was just the

1957
01:23:01,370 --> 01:22:58,980
intentional gibberish to confuse people

1958
01:23:05,169 --> 01:23:01,380
this is similar to the people that say

1959
01:23:07,610 --> 01:23:05,179
that Vines is a joke right it's uh uh

1960
01:23:10,850 --> 01:23:07,620
they they make the same assumption that

1961
01:23:13,729 --> 01:23:10,860
somebody just did it to confuse people

1962
01:23:16,070 --> 01:23:13,739
well thank you for spending about two

1963
01:23:18,410 --> 01:23:16,080
hours with me or an hour and a half on

1964
01:23:20,090 --> 01:23:18,420
what is potentially a joke but we

1965
01:23:22,610 --> 01:23:20,100
hopefully not

1966
01:23:24,470 --> 01:23:22,620
take care man it's good to speak with

1967
01:23:25,970 --> 01:23:24,480
you thank you it was it was fun talking

1968
01:23:27,470 --> 01:23:25,980
to you

1969
01:23:29,630 --> 01:23:27,480
bye

1970
01:23:31,310 --> 01:23:29,640
the podcast is now concluded thank you

1971
01:23:33,229 --> 01:23:31,320
for watching if you haven't subscribed

1972
01:23:35,209 --> 01:23:33,239
or clicked on that like button now would

1973
01:23:38,030 --> 01:23:35,219
be a great time to do so as each

1974
01:23:40,070 --> 01:23:38,040
subscribe and like helps YouTube push

1975
01:23:41,870 --> 01:23:40,080
this content to more people also I

1976
01:23:44,270 --> 01:23:41,880
recently found out that external links

1977
01:23:46,790 --> 01:23:44,280
count plenty toward the algorithm which

1978
01:23:48,050 --> 01:23:46,800
means that when you share on Twitter on

1979
01:23:50,209 --> 01:23:48,060
Facebook on Reddit

1980
01:23:52,070 --> 01:23:50,219
Etc it shows YouTube that people are

1981
01:23:54,110 --> 01:23:52,080
talking about this outside of YouTube

1982
01:23:55,970 --> 01:23:54,120
which in turn greatly AIDS the

1983
01:23:57,649 --> 01:23:55,980
Distribution on YouTube as well if you'd

1984
01:23:59,950 --> 01:23:57,659
like to support more conversations like

1985
01:24:02,030 --> 01:23:59,960
this then do consider visiting

1986
01:24:04,250 --> 01:24:02,040
theoriesofeverything.org again it's

1987
01:24:06,709 --> 01:24:04,260
support from the sponsors and you that

1988
01:24:08,330 --> 01:24:06,719
allow me to work on toe full-time you

1989
01:24:10,430 --> 01:24:08,340
get early access to ad-free audio

1990
01:24:12,590 --> 01:24:10,440
episodes there as well every dollar

1991
01:24:14,630 --> 01:24:12,600
helps far more than you may think either