WEBVTT FILE

1
00:00:00.586 --> 00:00:02.726
[gentle music]

2
00:00:02.000 --> 00:00:07.000
Downloaded from
YTS.MX

3
00:00:08.000 --> 00:00:13.000
Official YIFY movies site:
YTS.MX

4
00:00:14.324 --> 00:00:15.532
[Narrator] For decades,

5
00:00:15.532 --> 00:00:17.017
we have discussed
the many outcomes,

6
00:00:17.017 --> 00:00:19.053
regarding artificial
intelligence.

7
00:00:19.053 --> 00:00:21.469
Could our world be dominated?

8
00:00:21.469 --> 00:00:25.232
Could our independence and
autonomy be stripped from us,

9
00:00:25.232 --> 00:00:28.407
or are we able to control
what we have created?

10
00:00:28.407 --> 00:00:31.100
[upbeat music]

11
00:00:37.416 --> 00:00:41.006
Could we use artificial
intelligence to benefit our society?

12
00:00:41.006 --> 00:00:44.009
Just how thin is the line
between the development

13
00:00:44.009 --> 00:00:46.805
of civilization and chaos?

14
00:00:46.805 --> 00:00:49.428
[upbeat music]

15
00:01:13.211 --> 00:01:15.903
To understand what
artificial intelligence is,

16
00:01:15.903 --> 00:01:19.803
one must understand that it
can take many different forms.

17
00:01:19.803 --> 00:01:22.047
Think of it as a web of ideas,

18
00:01:22.047 --> 00:01:25.326
slowly expanding as new
ways of utilizing computers

19
00:01:25.326 --> 00:01:26.603
are explored.

20
00:01:26.603 --> 00:01:28.260
As technology develops,

21
00:01:28.260 --> 00:01:31.539
so do the capabilities of
self-learning software.

22
00:01:31.539 --> 00:01:34.335
- [Reporter] The need to
diagnose disease quickly

23
00:01:34.335 --> 00:01:38.132
and effectively has prompted
many university medical centers

24
00:01:38.132 --> 00:01:41.791
to develop intelligent
programs that simulate the work

25
00:01:41.791 --> 00:01:44.345
of doctors and
laboratory technicians.

26
00:01:44.345 --> 00:01:47.003
[gentle music]

27
00:01:48.694 --> 00:01:51.041
- [Narrator] AI is
quickly integrating with our way of life.

28
00:01:51.041 --> 00:01:54.631
So, much so that development
of AI programs has in itself,

29
00:01:54.631 --> 00:01:56.323
become a business opportunity.

30
00:01:57.945 --> 00:01:58.773
[upbeat music]

31
00:01:58.773 --> 00:01:59.809
In our modern age,

32
00:01:59.809 --> 00:02:01.638
we are powered by technology

33
00:02:01.638 --> 00:02:05.021
and softwares are transcending
its virtual existence,

34
00:02:05.021 --> 00:02:07.437
finding applications
in various fields,

35
00:02:07.437 --> 00:02:11.372
such as customer support
to content creation,

36
00:02:11.372 --> 00:02:13.202
computer-aided design,

37
00:02:13.202 --> 00:02:17.137
otherwise known as CAD, is
one of the many uses of AI.

38
00:02:17.137 --> 00:02:19.415
By analyzing
particular variables,

39
00:02:19.415 --> 00:02:22.280
computers are now able to
assist in the modification

40
00:02:22.280 --> 00:02:26.180
and creation of designs for
hardware and architecture.

41
00:02:26.180 --> 00:02:30.046
The prime use of any AI is
for optimizing processes

42
00:02:30.046 --> 00:02:32.324
that were considered
tedious before.

43
00:02:32.324 --> 00:02:35.189
In many ways, AI has
been hugely beneficial

44
00:02:35.189 --> 00:02:38.951
for technological development
thanks to its sheer speed.

45
00:02:38.951 --> 00:02:41.057
However, AI only benefits

46
00:02:41.057 --> 00:02:43.508
those to whom the
programs are distributed.

47
00:02:44.302 --> 00:02:45.613
Artificial intelligence

48
00:02:45.613 --> 00:02:47.443
is picking through your rubbish.

49
00:02:47.443 --> 00:02:51.688
This robot uses it to sort
through plastics for recycling

50
00:02:51.688 --> 00:02:53.414
and it can be retrained

51
00:02:53.414 --> 00:02:55.968
to prioritize whatever's
more marketable.

52
00:02:57.177 --> 00:03:00.180
So, AI can clearly
be incredibly useful,

53
00:03:00.180 --> 00:03:02.596
but there are deep
concerns about

54
00:03:02.596 --> 00:03:07.635
how quickly it is developing
and where it could go next.

55
00:03:08.912 --> 00:03:11.121
- The aim is to make
them as capable as humans

56
00:03:11.121 --> 00:03:14.366
and deploy them in
the service sector.

57
00:03:14.366 --> 00:03:16.230
The engineers in this research

58
00:03:16.230 --> 00:03:18.059
and development lab are working

59
00:03:18.059 --> 00:03:21.822
to take these humanoid
robots to the next level

60
00:03:21.822 --> 00:03:24.583
where they can not
only speak and move,

61
00:03:24.583 --> 00:03:27.345
but they can think
and feel and act

62
00:03:27.345 --> 00:03:30.002
and even make decisions
for themselves.

63
00:03:30.796 --> 00:03:32.695
And that daily data stream

64
00:03:32.695 --> 00:03:36.008
is being fed into an
ever expanding workforce,

65
00:03:36.008 --> 00:03:39.529
dedicated to developing
artificial intelligence.

66
00:03:41.013 --> 00:03:42.808
Those who have studied abroad

67
00:03:42.808 --> 00:03:46.122
are being encouraged to
return to the motherland.

68
00:03:46.122 --> 00:03:47.917
Libo Yang came back

69
00:03:47.917 --> 00:03:51.645
and started a tech
enterprise in his hometown.

70
00:03:51.645 --> 00:03:54.268
- [Narrator] China's market
is indeed the most open

71
00:03:54.268 --> 00:03:56.926
and active market
in the world for AI.

72
00:03:56.926 --> 00:04:01.241
It is also where there are the
most application scenarios.

73
00:04:01.241 --> 00:04:03.864
- So, AI is generally a
broad term that we apply

74
00:04:03.864 --> 00:04:04.934
to a number of techniques.

75
00:04:04.934 --> 00:04:06.384
And in this particular case,

76
00:04:06.384 --> 00:04:09.456
what we're actually looking
at was elements of AI,

77
00:04:09.456 --> 00:04:12.010
machine learning
and deep learning.

78
00:04:12.010 --> 00:04:13.701
So, in this particular case,

79
00:04:13.701 --> 00:04:17.429
we've been unfortunately
in a situation

80
00:04:17.429 --> 00:04:20.398
in this race against time
to create new antibiotics,

81
00:04:20.398 --> 00:04:22.779
the threat is
actually quite real

82
00:04:22.779 --> 00:04:25.230
and it would be
a global problem.

83
00:04:25.230 --> 00:04:27.784
We desperately needed to
harness new technologies

84
00:04:27.784 --> 00:04:29.269
in an attempt to fight it,

85
00:04:29.269 --> 00:04:30.960
we're looking at drugs

86
00:04:30.960 --> 00:04:33.411
which could potentially
fight E. coli,

87
00:04:33.411 --> 00:04:35.102
a very dangerous bacteria.

88
00:04:35.102 --> 00:04:37.207
- So, what is it
that the AI is doing

89
00:04:37.207 --> 00:04:39.348
that humans can't
do very simply?

90
00:04:39.348 --> 00:04:41.729
- So, the AI can
look for patterns

91
00:04:41.729 --> 00:04:44.560
that we wouldn't be able to
mind for with a human eye,

92
00:04:44.560 --> 00:04:47.287
simply within what I
do as a radiologist,

93
00:04:47.287 --> 00:04:50.980
I look for patterns of
diseases in terms of shape,

94
00:04:50.980 --> 00:04:53.914
contrast enhancement,
heterogeneity.

95
00:04:53.914 --> 00:04:55.191
But what the computer does,

96
00:04:55.191 --> 00:04:58.125
it looks for patterns
within the pixels.

97
00:04:58.125 --> 00:05:00.679
These are things that you just
can't see to the human eye.

98
00:05:00.679 --> 00:05:03.855
There's so much more data
embedded within these scans

99
00:05:03.855 --> 00:05:07.514
that we use that we can't
mine on a physical level.

100
00:05:07.514 --> 00:05:09.516
So, the computers really help.

101
00:05:09.516 --> 00:05:11.311
- [Narrator] Many
believe the growth of AI

102
00:05:11.311 --> 00:05:13.692
is dependent on
global collaboration,

103
00:05:13.692 --> 00:05:17.109
but access to the technology
is limited in certain regions.

104
00:05:17.109 --> 00:05:19.767
Global distribution is
a long-term endeavor

105
00:05:19.767 --> 00:05:21.044
and the more countries

106
00:05:21.044 --> 00:05:23.288
and businesses that
have access to the tech,

107
00:05:23.288 --> 00:05:26.429
the more regulation
the AI will require.

108
00:05:26.429 --> 00:05:29.846
In fact, it is now not
uncommon for businesses

109
00:05:29.846 --> 00:05:33.125
to be entirely run by
an artificial director.

110
00:05:33.125 --> 00:05:34.472
On many occasions,

111
00:05:34.472 --> 00:05:37.198
handing the helm of a
company to an algorithm

112
00:05:37.198 --> 00:05:40.685
can provide the best option
on the basis of probability.

113
00:05:40.685 --> 00:05:43.998
However, dependence and
reliability on softwares

114
00:05:43.998 --> 00:05:45.897
can be a great risk.

115
00:05:45.897 --> 00:05:47.450
Without proper safeguards,

116
00:05:47.450 --> 00:05:50.419
actions based on potentially
incorrect predictions

117
00:05:50.419 --> 00:05:53.353
can be a detriment to a
business or operation.

118
00:05:53.353 --> 00:05:55.147
Humans provide the
critical thinking

119
00:05:55.147 --> 00:05:58.461
and judgment which AI is
not capable of matching.

120
00:05:58.461 --> 00:06:00.463
- Well, this is the
Accessibility Design Center

121
00:06:00.463 --> 00:06:02.810
and it's where we try to
bring together our engineers

122
00:06:02.810 --> 00:06:05.882
and experts with the
latest AI technology,

123
00:06:05.882 --> 00:06:07.608
with people with disabilities,

124
00:06:07.608 --> 00:06:10.059
because there's a
real opportunity to firstly help people

125
00:06:10.059 --> 00:06:12.613
with disabilities enjoy
all the technology

126
00:06:12.613 --> 00:06:14.201
we have in our pockets today.

127
00:06:14.201 --> 00:06:15.720
And sometimes that's
not very accessible,

128
00:06:15.720 --> 00:06:18.688
but also build tools that
can help them engage better

129
00:06:18.688 --> 00:06:20.103
in the real world.

130
00:06:20.103 --> 00:06:22.451
And that's thanks to the
wonders of machine learning.

131
00:06:22.451 --> 00:06:25.764
- I don't think we're like at
the end of this paradigm yet.

132
00:06:25.764 --> 00:06:26.903
We'll keep pushing these.

133
00:06:26.903 --> 00:06:28.215
We'll add other modalities.

134
00:06:28.215 --> 00:06:31.114
So, someday they'll do
video, audio images,

135
00:06:31.114 --> 00:06:36.154
text altogether and they'll get
like much smarter over time.

136
00:06:37.638 --> 00:06:38.674
- AI, machine learning, all
very sounds very complicated.

137
00:06:38.674 --> 00:06:40.572
Just think about it as a toolkit

138
00:06:40.572 --> 00:06:42.781
that's really good at
sort of spotting patterns

139
00:06:42.781 --> 00:06:44.024
and making predictions,

140
00:06:44.024 --> 00:06:46.336
better than any computing
could do before.

141
00:06:46.336 --> 00:06:47.786
And that's why it's so useful

142
00:06:47.786 --> 00:06:51.031
for things like understanding
language and speech.

143
00:06:51.031 --> 00:06:52.998
Another product which
we are launching today

144
00:06:52.998 --> 00:06:55.000
is called Project Relate.

145
00:06:55.000 --> 00:06:56.312
And this is for people

146
00:06:56.312 --> 00:06:58.728
who have non-standard
speech patterns.

147
00:06:58.728 --> 00:07:00.937
So, one of the
people we work with

148
00:07:00.937 --> 00:07:03.837
is maybe less than
10% of the time,

149
00:07:03.837 --> 00:07:06.564
could be understood by
people who don't know her,

150
00:07:06.564 --> 00:07:09.325
using this tool that's
over 90% of the time.

151
00:07:09.325 --> 00:07:12.259
And you think about
that transformation in somebody's life

152
00:07:12.259 --> 00:07:15.676
and then you think about the
fact there's 250 million people

153
00:07:15.676 --> 00:07:17.678
with non-standard speech
patterns around the world.

154
00:07:17.678 --> 00:07:19.093
So, that's the
ambition of this center

155
00:07:19.093 --> 00:07:21.682
is to unite technology with
people with disabilities

156
00:07:21.682 --> 00:07:24.478
and try to help 'em
engage more in the world.

157
00:07:24.478 --> 00:07:27.550
- [Narrator] On the
30th November of 2022,

158
00:07:27.550 --> 00:07:30.001
a revolutionary
innovation emerged,

159
00:07:30.967 --> 00:07:32.003
ChatGPT.

160
00:07:32.969 --> 00:07:35.869
ChatGPT was created by OpenAI,

161
00:07:35.869 --> 00:07:38.250
an AI research organization.

162
00:07:38.250 --> 00:07:39.873
Its goal is to develop systems

163
00:07:39.873 --> 00:07:44.498
which may benefit all aspects
of society and communication.

164
00:07:44.498 --> 00:07:47.467
Sam Altman stepped
up as CEO of OpenAI

165
00:07:47.467 --> 00:07:50.055
on its launch in 2015.

166
00:07:50.055 --> 00:07:51.609
Altman dabbled in a multitude

167
00:07:51.609 --> 00:07:53.990
of computing-based
business ventures.

168
00:07:53.990 --> 00:07:57.477
His rise to CEO was thanks
to his many affiliations

169
00:07:57.477 --> 00:08:01.377
and investments with computing
and social media companies.

170
00:08:01.377 --> 00:08:04.173
He began his journey
by co-founding Loopt,

171
00:08:04.173 --> 00:08:06.106
a social media service.

172
00:08:06.106 --> 00:08:07.763
After selling the application,

173
00:08:07.763 --> 00:08:10.835
Altman went on to bigger
and riskier endeavors

174
00:08:10.835 --> 00:08:14.148
from startup accelerator
companies to security software.

175
00:08:15.184 --> 00:08:17.393
OpenAI became hugely desirable,

176
00:08:17.393 --> 00:08:20.223
thanks to the amount of revenue
the company had generated

177
00:08:20.223 --> 00:08:21.984
with over a billion dollars made

178
00:08:21.984 --> 00:08:24.262
within its first
year of release.

179
00:08:24.262 --> 00:08:27.265
ChatGPT became an easily
accessible software,

180
00:08:27.265 --> 00:08:30.786
built on a large language
model known as an LLM.

181
00:08:30.786 --> 00:08:34.134
This program can conjure
complex human-like responses

182
00:08:34.134 --> 00:08:37.309
to the user's questions
otherwise known as prompts.

183
00:08:37.309 --> 00:08:38.794
In essence,

184
00:08:38.794 --> 00:08:41.244
it is a program which
learns the more it is used.

185
00:08:43.592 --> 00:08:45.317
The new age therapeutic program

186
00:08:45.317 --> 00:08:48.804
was developed on the GPT-3.5.

187
00:08:48.804 --> 00:08:51.531
The architecture of this
older model allowed systems

188
00:08:51.531 --> 00:08:53.602
to understand and generate code

189
00:08:53.602 --> 00:08:56.501
and natural languages at a
remarkably advanced level

190
00:08:56.501 --> 00:08:59.884
from analyzing syntax
to nuances in writing.

191
00:08:59.884 --> 00:09:02.542
[upbeat music]

192
00:09:04.578 --> 00:09:06.753
ChatGPT took the world by storm,

193
00:09:06.753 --> 00:09:09.445
due to the sophistication
of the system.

194
00:09:09.445 --> 00:09:11.067
As with many chatbot systems,

195
00:09:11.067 --> 00:09:13.449
people have since found
ways to manipulate

196
00:09:13.449 --> 00:09:17.349
and confuse the software in
order to test its limits.

197
00:09:17.349 --> 00:09:20.076
[gentle music]

198
00:09:21.526 --> 00:09:25.910
The first computer was invented
by Charles Babbage in 1822.

199
00:09:25.910 --> 00:09:29.189
It was to be a rudimentary
general purpose system.

200
00:09:29.189 --> 00:09:34.021
In 1936, the system was
developed upon by Alan Turing.

201
00:09:34.021 --> 00:09:36.299
The automatic machine,
as he called them,

202
00:09:36.299 --> 00:09:38.854
was able to break enigma
enciphered messages,

203
00:09:38.854 --> 00:09:41.201
regarding enemy
military operations,

204
00:09:41.201 --> 00:09:43.583
during the Second World War.

205
00:09:43.583 --> 00:09:46.447
Turing theorized his
own type of computer,

206
00:09:46.447 --> 00:09:49.830
the Turing Machine has
coined by Alonzo Church,

207
00:09:49.830 --> 00:09:52.522
after reading Turing's
research paper.

208
00:09:52.522 --> 00:09:55.698
It had become realized that
soon prospect of computing

209
00:09:55.698 --> 00:09:57.907
and engineering would
merge seamlessly.

210
00:09:59.046 --> 00:10:01.152
Theories of future
tech would increase

211
00:10:01.152 --> 00:10:04.742
and soon came a huge outburst
in science fiction media.

212
00:10:04.742 --> 00:10:07.468
This was known as the
golden age for computing.

213
00:10:07.468 --> 00:10:10.092
[gentle music]

214
00:10:20.067 --> 00:10:22.760
Alan Turing's contributions
to computability

215
00:10:22.760 --> 00:10:25.590
and theoretical computer
science was one step closer

216
00:10:25.590 --> 00:10:28.110
to producing a reactive machine.

217
00:10:28.110 --> 00:10:31.389
The reactive machine
is an early form of AI.

218
00:10:31.389 --> 00:10:32.942
They had limited capabilities

219
00:10:32.942 --> 00:10:34.772
and were unable
to store memories

220
00:10:34.772 --> 00:10:37.740
in order to learn new
algorithms of data.

221
00:10:37.740 --> 00:10:41.641
However, they were able to
react to specific stimuli.

222
00:10:41.641 --> 00:10:46.611
The first AI was a
program written in 1952 by Arthur Samuel.

223
00:10:47.854 --> 00:10:49.614
The prototype AI was
able to play checkers,

224
00:10:49.614 --> 00:10:52.168
against an opponent and
was built to operate

225
00:10:52.168 --> 00:10:56.172
on the Ferranti Mark One, an
early commercial computer.

226
00:10:56.172 --> 00:10:57.657
- [Reporter] This computer
has been playing the game

227
00:10:57.657 --> 00:11:00.418
for several years now,
getting better all the time.

228
00:11:00.418 --> 00:11:02.972
Tonight it's playing against
the black side of the board.

229
00:11:02.972 --> 00:11:05.837
It's approach to playing
drafts, it's almost human.

230
00:11:05.837 --> 00:11:08.012
It remembers the moves
that enable it to win

231
00:11:08.012 --> 00:11:10.324
and the sort that
lead to defeat.

232
00:11:10.324 --> 00:11:12.982
The computer indicates the move
it wants to make on a panel

233
00:11:12.982 --> 00:11:14.156
of flashing lights.

234
00:11:14.156 --> 00:11:15.433
It's up to the human opponent

235
00:11:15.433 --> 00:11:18.229
to actually move the
drafts about the board.

236
00:11:18.229 --> 00:11:20.645
This sort of works producing
exciting information

237
00:11:20.645 --> 00:11:22.405
on the way in which
electronic brains

238
00:11:22.405 --> 00:11:24.338
can learn from past experience

239
00:11:24.338 --> 00:11:26.168
and improve their performances.

240
00:11:27.963 --> 00:11:29.792
[Narrator] In 1966,

241
00:11:29.792 --> 00:11:32.519
an MIT professor named
Joseph Weizenbaum,

242
00:11:32.519 --> 00:11:37.110
created an AI which would
change the landscape of society.

243
00:11:37.110 --> 00:11:39.077
It was known as Eliza,

244
00:11:39.077 --> 00:11:42.322
and it was designed to act
like a psychotherapist.

245
00:11:42.322 --> 00:11:45.497
The software was simplistic,
yet revolutionary.

246
00:11:45.497 --> 00:11:47.499
The AI would receive
the user input

247
00:11:47.499 --> 00:11:51.055
and use specific parameters to
generate a coherent response.

248
00:11:53.057 --> 00:11:55.991
- It it has been said,
especially here at MIT,

249
00:11:55.991 --> 00:11:59.719
that computers will
take over in some sense

250
00:11:59.719 --> 00:12:02.652
and it's even been said
that if we're lucky,

251
00:12:02.652 --> 00:12:04.447
they'll keep us as pets

252
00:12:04.447 --> 00:12:06.277
and Arthur C. Clarke, the
science fiction writer,

253
00:12:06.277 --> 00:12:09.694
we marked once that if
that were to happen,

254
00:12:09.694 --> 00:12:12.904
it would serve us
right, he said.

255
00:12:12.904 --> 00:12:14.734
- [Narrator] The program
maintained the illusion

256
00:12:14.734 --> 00:12:16.943
of understanding its
user to the point

257
00:12:16.943 --> 00:12:20.498
where Weizenbaum's secretary
requested some time alone

258
00:12:20.498 --> 00:12:23.363
with Eliza to
express her feelings.

259
00:12:23.363 --> 00:12:26.711
Though Eliza is now considered
outdated technology,

260
00:12:26.711 --> 00:12:29.369
it remains a talking
point due to its ability

261
00:12:29.369 --> 00:12:31.785
to illuminate an aspect
of the human mind

262
00:12:31.785 --> 00:12:34.132
in our relationship
with computers.

263
00:12:34.132 --> 00:12:36.756
- And it's connected
over the telephone line

264
00:12:36.756 --> 00:12:38.965
to someone or something
at the other end.

265
00:12:38.965 --> 00:12:42.106
Now, I'm gonna play 20
questions with whatever it is.

266
00:12:42.106 --> 00:12:44.418
[type writer clacking]

267
00:12:44.418 --> 00:12:45.419
Very helpful.

268
00:12:45.419 --> 00:12:48.768
[type writer clacking]

269
00:12:53.773 --> 00:12:55.119
- 'Cause clearly if
we can make a machine

270
00:12:55.119 --> 00:12:56.776
as intelligent as ourselves,

271
00:12:56.776 --> 00:12:59.157
then it can make one
that's more intelligent.

272
00:12:59.157 --> 00:13:04.024
Now, the one I'm talking about
now will certainly happen.

273
00:13:05.301 --> 00:13:07.476
I mean, it could produce
an evil result of course,

274
00:13:07.476 --> 00:13:08.615
if we were careless,

275
00:13:08.615 --> 00:13:10.134
but what is quite certain

276
00:13:10.134 --> 00:13:14.138
is that we're heading
towards machine intelligence,

277
00:13:14.138 --> 00:13:17.486
machines that are
intelligent in every sense.

278
00:13:17.486 --> 00:13:19.246
It doesn't matter
how you define it,

279
00:13:19.246 --> 00:13:22.940
they'll be able to be
that sort of intelligent.

280
00:13:22.940 --> 00:13:26.046
A human is a machine,
unless there's a soul.

281
00:13:26.046 --> 00:13:29.670
I don't personally believe
that humans have souls

282
00:13:29.670 --> 00:13:32.535
in anything other
than a poetic sense,

283
00:13:32.535 --> 00:13:34.158
which I do believe
in, of course.

284
00:13:34.158 --> 00:13:37.437
But in a literal God-like sense,

285
00:13:37.437 --> 00:13:38.610
I don't believe we have souls.

286
00:13:38.610 --> 00:13:39.991
And so personally,

287
00:13:39.991 --> 00:13:42.407
I believe that we are
essentially machines.

288
00:13:43.823 --> 00:13:46.722
- [Narrator] This type of
program is known as an NLP,

289
00:13:46.722 --> 00:13:49.242
Natural Language Processing.

290
00:13:49.242 --> 00:13:52.176
This branch of artificial
intelligence enables computers

291
00:13:52.176 --> 00:13:55.489
to comprehend, generate and
manipulate human language.

292
00:13:56.905 --> 00:13:59.114
The concept of a
responsive machine

293
00:13:59.114 --> 00:14:02.358
was the mash that lit the
flame for worldwide concern.

294
00:14:03.739 --> 00:14:06.466
The systems were beginning
to raise ethical dilemmas,

295
00:14:06.466 --> 00:14:08.813
such as the use of
autonomous weapons,

296
00:14:08.813 --> 00:14:11.781
invasions of privacy through
surveillance technologies

297
00:14:11.781 --> 00:14:13.300
and the potential for misuse

298
00:14:13.300 --> 00:14:17.097
or unintended consequences
in decision making.

299
00:14:17.097 --> 00:14:18.858
When a command is
executed based,

300
00:14:18.858 --> 00:14:21.067
upon set rules in algorithms,

301
00:14:21.067 --> 00:14:24.346
it might not always be the
morally correct choice.

302
00:14:24.346 --> 00:14:28.453
Imagination seems to be,

303
00:14:28.453 --> 00:14:31.594
some sort of process of random
thoughts being generated

304
00:14:31.594 --> 00:14:34.528
in the mind and then the
conscious mind selecting from a

305
00:14:34.528 --> 00:14:36.392
or some part of
the brain anyway,

306
00:14:36.392 --> 00:14:37.773
perhaps even below
the conscious mind,

307
00:14:37.773 --> 00:14:40.500
selecting from a pool of
ideas and aligns with some

308
00:14:40.500 --> 00:14:42.122
and blocking others.

309
00:14:42.122 --> 00:14:45.608
And yes, a machine
can do the same thing.

310
00:14:45.608 --> 00:14:48.611
In fact, we can only
say that a machine

311
00:14:48.611 --> 00:14:50.890
is fundamentally different
from a human being,

312
00:14:50.890 --> 00:14:53.133
eventually, always
fundamentally, if we believe in a soul.

313
00:14:53.133 --> 00:14:55.687
So, that boils down
to religious matter.

314
00:14:55.687 --> 00:14:58.932
If human beings have souls,
then clearly machines won't

315
00:14:58.932 --> 00:15:01.141
and there will always be
a fundamental difference.

316
00:15:01.141 --> 00:15:03.005
If you don't believe
humans have souls,

317
00:15:03.005 --> 00:15:04.765
then machines can do anything

318
00:15:04.765 --> 00:15:07.078
and everything
that a human does.

319
00:15:07.078 --> 00:15:10.116
- A computer which is
capable of finding out

320
00:15:10.116 --> 00:15:11.565
where it's gone wrong,

321
00:15:11.565 --> 00:15:14.051
finding out how its program
has already served it

322
00:15:14.051 --> 00:15:15.776
and then changing its program

323
00:15:15.776 --> 00:15:17.261
in the light of what
it had discovered

324
00:15:17.261 --> 00:15:18.814
is a learning machine.

325
00:15:18.814 --> 00:15:21.679
And this is something quite
fundamentally new in the world.

326
00:15:23.163 --> 00:15:25.027
- I'd like to be able to say
that it's only a slight change

327
00:15:25.027 --> 00:15:27.754
and we'll all be used to
it very, very quickly.

328
00:15:27.754 --> 00:15:29.307
But I don't think it is.

329
00:15:29.307 --> 00:15:33.070
I think that although we've
spoken probably of the whole

330
00:15:33.070 --> 00:15:35.417
of this century about
a coming revolution

331
00:15:35.417 --> 00:15:38.523
and about the end
of work and so on,

332
00:15:38.523 --> 00:15:39.904
finally it's actually happening.

333
00:15:39.904 --> 00:15:42.148
And it's actually
happening because now,

334
00:15:42.148 --> 00:15:46.117
it's suddenly become
cheaper to have a machine

335
00:15:46.117 --> 00:15:49.224
do a mental task
than for a man to,

336
00:15:49.224 --> 00:15:52.192
at the moment, at a fairly
low level of mental ability,

337
00:15:52.192 --> 00:15:54.298
but at an ever increasing
level of sophistication

338
00:15:54.298 --> 00:15:56.024
as these machines acquire,

339
00:15:56.024 --> 00:15:58.543
more and more human-like
mental abilities.

340
00:15:58.543 --> 00:16:01.408
So, just as men's
muscles were replaced

341
00:16:01.408 --> 00:16:03.272
in the First
Industrial Revolution

342
00:16:03.272 --> 00:16:04.998
in this second
industrial revolution

343
00:16:04.998 --> 00:16:07.069
or whatever you call it
or might like to call it,

344
00:16:07.069 --> 00:16:09.623
then men's mines will
be replaced in industry.

345
00:16:11.487 --> 00:16:13.938
- [Narrator] In order for
NLP systems to improve,

346
00:16:13.938 --> 00:16:16.941
the program must receive
feedback from human users.

347
00:16:18.287 --> 00:16:20.634
These iterative feedback
loops play a significant role

348
00:16:20.634 --> 00:16:23.396
in fine tuning each
model of the AI,

349
00:16:23.396 --> 00:16:26.192
further developing its
conversational capabilities.

350
00:16:27.538 --> 00:16:30.679
Organizations such as
OpenAI have taken automation

351
00:16:30.679 --> 00:16:34.372
to new lengths with
systems such as DALL-E,

352
00:16:34.372 --> 00:16:37.375
the generation of imagery and
art has never been easier.

353
00:16:38.445 --> 00:16:40.447
The term auto
generative imagery,

354
00:16:40.447 --> 00:16:43.450
refers to the creation
of visual content.

355
00:16:43.450 --> 00:16:46.384
These kinds of programs
have become so widespread,

356
00:16:46.384 --> 00:16:48.628
it is becoming
increasingly more difficult

357
00:16:48.628 --> 00:16:50.940
to tell the fake from the real.

358
00:16:50.940 --> 00:16:52.321
Using algorithms,

359
00:16:52.321 --> 00:16:55.359
programs such as DALL-E
and Midjourney are able

360
00:16:55.359 --> 00:16:58.500
to create visuals in
a matter of seconds.

361
00:16:58.500 --> 00:17:01.434
Whilst a human artist
could spend days, weeks

362
00:17:01.434 --> 00:17:04.747
or even years in order to
create a beautiful image.

363
00:17:04.747 --> 00:17:07.509
For us the discipline
required to pursue art

364
00:17:07.509 --> 00:17:11.513
is a contributing factor to
the appreciation of art itself.

365
00:17:11.513 --> 00:17:14.757
But if a software is able
to produce art in seconds,

366
00:17:14.757 --> 00:17:17.622
it puts artists in a
vulnerable position

367
00:17:17.622 --> 00:17:20.453
with even their
jobs being at risk.

368
00:17:20.453 --> 00:17:22.386
- Well, I think we see
risk coming through

369
00:17:22.386 --> 00:17:25.147
into the white collar jobs,
the professional jobs,

370
00:17:25.147 --> 00:17:27.563
we're already seeing artificial
intelligence solutions,

371
00:17:27.563 --> 00:17:30.911
being used in healthcare
and legal services.

372
00:17:30.911 --> 00:17:34.225
And so those jobs which
have been relatively immune

373
00:17:34.225 --> 00:17:38.402
to industrialization so far,
they're not immune anymore.

374
00:17:38.402 --> 00:17:40.783
And so people like
myself as a lawyer,

375
00:17:40.783 --> 00:17:42.509
I would hope I won't be,

376
00:17:42.509 --> 00:17:44.615
but I could be out of a
job in five years time.

377
00:17:44.615 --> 00:17:47.376
- An Oxford University study
suggests that between a third

378
00:17:47.376 --> 00:17:49.965
and almost a half of
all jobs are vanishing,

379
00:17:49.965 --> 00:17:52.899
because machines are simply
better at doing them.

380
00:17:52.899 --> 00:17:54.797
That means the generation here,

381
00:17:54.797 --> 00:17:57.041
simply won't have the
access to the professions

382
00:17:57.041 --> 00:17:57.938
that we have.

383
00:17:57.938 --> 00:17:59.457
Almost on a daily basis,

384
00:17:59.457 --> 00:18:01.149
you're seeing new
technologies emerge

385
00:18:01.149 --> 00:18:02.667
that seem to be taking on tasks

386
00:18:02.667 --> 00:18:04.428
that in the past we thought

387
00:18:04.428 --> 00:18:06.188
they could only be
done by human beings.

388
00:18:06.188 --> 00:18:09.191
- Lots of people have talked
about the shifts in technology,

389
00:18:09.191 --> 00:18:11.642
leading to widespread
unemployment

390
00:18:11.642 --> 00:18:12.884
and they've been proved wrong.

391
00:18:12.884 --> 00:18:14.369
Why is it different this time?

392
00:18:14.369 --> 00:18:16.578
- The difference here is
that the technologies,

393
00:18:16.578 --> 00:18:19.167
A, they seem to be coming
through more rapidly,

394
00:18:19.167 --> 00:18:21.238
and B, they're taking on
not just manual tests,

395
00:18:21.238 --> 00:18:22.480
but cerebral tests too.

396
00:18:22.480 --> 00:18:24.551
They're solving all
sorts of problems,

397
00:18:24.551 --> 00:18:26.553
undertaking tests that
we thought historically,

398
00:18:26.553 --> 00:18:28.348
required human intelligence.

399
00:18:28.348 --> 00:18:29.522
- Well, DIM robots
are the robots

400
00:18:29.522 --> 00:18:31.765
we have on the
factory floor today

401
00:18:31.765 --> 00:18:33.733
in all the advanced countries.

402
00:18:33.733 --> 00:18:35.044
They're blind and dumb,

403
00:18:35.044 --> 00:18:36.908
they don't understand
their surroundings.

404
00:18:36.908 --> 00:18:40.533
And the other kind of robot,

405
00:18:40.533 --> 00:18:43.984
which will dominate the
technology of the late 1980s

406
00:18:43.984 --> 00:18:47.505
in automation and also
is of acute interest

407
00:18:47.505 --> 00:18:50.646
to experimental artificial
intelligence scientists

408
00:18:50.646 --> 00:18:54.788
is the kind of robot
where the human can convey

409
00:18:54.788 --> 00:18:59.828
to its machine assistance
his own concepts,

410
00:19:01.036 --> 00:19:04.453
suggested strategies and
the machine, the robot

411
00:19:04.453 --> 00:19:06.110
can understand him,

412
00:19:06.110 --> 00:19:09.286
but no machine can accept

413
00:19:09.286 --> 00:19:12.116
and utilize concepts
from a person,

414
00:19:12.116 --> 00:19:16.016
unless he has some kind of
window on the same world

415
00:19:16.016 --> 00:19:17.742
that the person sees.

416
00:19:17.742 --> 00:19:22.540
And therefore, to be
an intelligent robot to a useful degree

417
00:19:22.540 --> 00:19:25.992
as an intelligent and
understanding assistant,

418
00:19:25.992 --> 00:19:29.409
robots are going to
have artificial eyes, artificial ears,

419
00:19:29.409 --> 00:19:32.101
artificial sense of
touch is just essential.

420
00:19:33.102 --> 00:19:34.069
- [Narrator] These
programs learn,

421
00:19:34.069 --> 00:19:35.864
through a variety of techniques,

422
00:19:35.864 --> 00:19:38.556
such as generative
adversarial networks,

423
00:19:38.556 --> 00:19:41.490
which allows for the
production of plausible data.

424
00:19:41.490 --> 00:19:43.320
After a prompt is inputted,

425
00:19:43.320 --> 00:19:45.667
the system learns what
aspects of imagery,

426
00:19:45.667 --> 00:19:47.807
sound and text are fake.

427
00:19:48.980 --> 00:19:50.223
- [Reporter] Machine
learning algorithms,

428
00:19:50.223 --> 00:19:52.225
could already label
objects in images,

429
00:19:52.225 --> 00:19:53.709
and now they learn
to put those labels

430
00:19:53.709 --> 00:19:55.987
into natural language
descriptions.

431
00:19:55.987 --> 00:19:58.197
And it made one group
of researchers curious.

432
00:19:58.197 --> 00:20:01.130
What if you flipped
that process around?

433
00:20:01.130 --> 00:20:03.271
If we could do image to text.

434
00:20:03.271 --> 00:20:05.894
Why not try doing
text to image as well

435
00:20:05.894 --> 00:20:07.240
and see how it works.

436
00:20:07.240 --> 00:20:08.483
- [Reporter] It was a
more difficult task.

437
00:20:08.483 --> 00:20:10.485
They didn't want to
retrieve existing images

438
00:20:10.485 --> 00:20:11.796
the way Google search does.

439
00:20:11.796 --> 00:20:14.178
They wanted to generate
entirely novel scenes

440
00:20:14.178 --> 00:20:16.249
that didn't happen
in the real world.

441
00:20:16.249 --> 00:20:19.045
- [Narrator] Once the AI learns
more visual discrepancies,

442
00:20:19.045 --> 00:20:21.875
the more effective the
later models will become.

443
00:20:21.875 --> 00:20:24.499
It is now very common
for software developers

444
00:20:24.499 --> 00:20:28.399
to band together in order
to improve their AI systems.

445
00:20:28.399 --> 00:20:31.471
Another learning model is
recurrent neural networks,

446
00:20:31.471 --> 00:20:33.991
which allows the AI to
train itself to create

447
00:20:33.991 --> 00:20:37.960
and predict algorithms by
recalling previous information.

448
00:20:37.960 --> 00:20:41.032
By utilizing what is
known as the memory state,

449
00:20:41.032 --> 00:20:42.896
the output of the
previous action

450
00:20:42.896 --> 00:20:46.072
can be passed forward into
the following input action

451
00:20:46.072 --> 00:20:50.249
or is otherwise should it
not meet previous parameters.

452
00:20:50.249 --> 00:20:53.493
This learning model allows
for consistent accuracy

453
00:20:53.493 --> 00:20:56.462
by repetition and exposure
to large fields of data.

454
00:20:58.602 --> 00:21:00.535
Whilst the person
will spend hours,

455
00:21:00.535 --> 00:21:02.847
practicing to paint
human anatomy,

456
00:21:02.847 --> 00:21:06.575
an AI can take existing data
and reproduce a new image

457
00:21:06.575 --> 00:21:10.821
with frighteningly good
accuracy in a matter of moments.

458
00:21:10.821 --> 00:21:12.892
- Well, I would say
that it's not so much

459
00:21:12.892 --> 00:21:17.379
a matter of whether a
machine can think or not,

460
00:21:17.379 --> 00:21:20.175
which is how you
prefer to use words,

461
00:21:20.175 --> 00:21:22.177
but rather whether
they can think

462
00:21:22.177 --> 00:21:23.834
in a sufficiently human-like way

463
00:21:25.111 --> 00:21:28.770
for people to have useful
communication with them.

464
00:21:28.770 --> 00:21:32.601
- If I didn't believe that
it was a beneficent prospect,

465
00:21:32.601 --> 00:21:34.120
I wouldn't be doing it.

466
00:21:34.120 --> 00:21:36.018
That wouldn't stop
other people doing it.

467
00:21:36.018 --> 00:21:40.471
But I wouldn't do it if I
didn't think it was for good.

468
00:21:40.471 --> 00:21:42.301
What I'm saying,

469
00:21:42.301 --> 00:21:44.095
and of course other people
have said long before me,

470
00:21:44.095 --> 00:21:45.442
it's not an original thought,

471
00:21:45.442 --> 00:21:49.791
is that we must consider
how to to control this.

472
00:21:49.791 --> 00:21:52.725
It won't be controlled
automatically.

473
00:21:52.725 --> 00:21:55.348
It's perfectly possible that
we could develop a machine,

474
00:21:55.348 --> 00:21:59.318
a robot say of
human-like intelligence

475
00:21:59.318 --> 00:22:01.975
and through neglect on our part,

476
00:22:01.975 --> 00:22:05.634
it could become a Frankenstein.

477
00:22:05.634 --> 00:22:08.844
- [Narrator] As with any
technology challenges arise,

478
00:22:08.844 --> 00:22:12.469
ethical concerns regarding
biases and misuse have existed,

479
00:22:12.469 --> 00:22:16.438
since the concept of artificial
intelligence was conceived.

480
00:22:16.438 --> 00:22:18.302
Due to autogenerated imagery,

481
00:22:18.302 --> 00:22:20.925
many believe the arts
industry has been placed

482
00:22:20.925 --> 00:22:22.789
in a difficult situation.

483
00:22:22.789 --> 00:22:26.241
Independent artists are now
being overshadowed by software.

484
00:22:27.276 --> 00:22:29.451
To many the improvement
of generative AI

485
00:22:29.451 --> 00:22:32.454
is hugely beneficial
and efficient.

486
00:22:32.454 --> 00:22:35.284
To others, it lacks the
authenticity of true art.

487
00:22:36.285 --> 00:22:38.667
In 2023, an image was submitted

488
00:22:38.667 --> 00:22:40.324
to the Sony Photography Awards

489
00:22:40.324 --> 00:22:43.327
by an artist called
Boris Eldagsen.

490
00:22:43.327 --> 00:22:45.916
The image was titled
The Electrician

491
00:22:45.916 --> 00:22:48.367
and depicted a woman
standing behind another

492
00:22:48.367 --> 00:22:50.369
with her hand resting
on her shoulders.

493
00:22:52.025 --> 00:22:53.924
[upbeat music]

494
00:22:53.924 --> 00:22:56.927
- One's got to realize that the
machines that we have today,

495
00:22:56.927 --> 00:23:01.138
the computers of today are
superhuman in their ability

496
00:23:01.138 --> 00:23:06.177
to handle numbers and infantile,

497
00:23:07.075 --> 00:23:08.317
sub-in infantile
in their ability

498
00:23:08.317 --> 00:23:10.768
to handle ideas and concepts.

499
00:23:10.768 --> 00:23:12.701
But there's a new generation
of machine coming along,

500
00:23:12.701 --> 00:23:14.289
which will be quite different.

501
00:23:14.289 --> 00:23:17.154
By the '90s or certainly
by the turn of the century,

502
00:23:17.154 --> 00:23:19.708
We will certainly be
able to make a machine

503
00:23:19.708 --> 00:23:22.193
with as many parts as
complex as human brain.

504
00:23:22.193 --> 00:23:24.437
Whether we'll be able to make
it do what human brain does

505
00:23:24.437 --> 00:23:26.197
at that stage is
quite another matter.

506
00:23:26.197 --> 00:23:28.545
But once we've got
something that complex

507
00:23:28.545 --> 00:23:30.547
we're well on the road to that.

508
00:23:30.547 --> 00:23:32.100
- [Narrator] The
image took first place

509
00:23:32.100 --> 00:23:34.689
in the Sony Photography
Awards Portrait Category.

510
00:23:34.689 --> 00:23:37.830
However, Boris revealed
to both Sony and the world

511
00:23:37.830 --> 00:23:41.696
that the image was indeed
AI-generated in DALL-E Two.

512
00:23:41.696 --> 00:23:44.423
[upbeat music]

513
00:23:45.424 --> 00:23:46.804
Boris denied the award,

514
00:23:46.804 --> 00:23:48.910
having used the image as a test

515
00:23:48.910 --> 00:23:52.085
to see if he could trick
the eyes of other artists.

516
00:23:52.085 --> 00:23:53.708
It had worked,

517
00:23:53.708 --> 00:23:56.711
the image had sparked debate
between the relationship

518
00:23:56.711 --> 00:23:58.609
of AI and photography.

519
00:23:58.609 --> 00:24:00.646
The images, much
like deep fakes,

520
00:24:00.646 --> 00:24:03.027
have become realistic
to the point of concern

521
00:24:03.027 --> 00:24:04.684
for authenticity.

522
00:24:04.684 --> 00:24:06.375
The complexity of AI systems,

523
00:24:06.375 --> 00:24:09.068
may lead to unintended
consequences.

524
00:24:09.068 --> 00:24:10.863
The systems have
developed to a point

525
00:24:10.863 --> 00:24:13.797
where it has outpaced
comprehensive regulations.

526
00:24:14.936 --> 00:24:16.765
Ethical guidelines
and legal frameworks

527
00:24:16.765 --> 00:24:18.871
are required to
ensure AI development,

528
00:24:18.871 --> 00:24:21.252
does not fall into
the wrong hands.

529
00:24:21.252 --> 00:24:22.702
- There have been a
lot of famous people

530
00:24:22.702 --> 00:24:25.291
who have had user
generated AI images of them

531
00:24:25.291 --> 00:24:28.190
that have gone viral
from Trump to the Pope.

532
00:24:28.190 --> 00:24:29.813
When you see them,

533
00:24:29.813 --> 00:24:31.884
do you feel like this is fun
and in the hands of the masses

534
00:24:31.884 --> 00:24:33.886
or do you feel
concerned about it?

535
00:24:33.886 --> 00:24:38.062
- I think it's something which
is very, very, very scary,

536
00:24:38.062 --> 00:24:41.203
because your or my
face could be taken off

537
00:24:41.203 --> 00:24:45.138
and put on in an environment
which we don't want to be in.

538
00:24:45.138 --> 00:24:46.657
Whether that's a crime

539
00:24:46.657 --> 00:24:48.556
or whether that's even
something like porn.

540
00:24:48.556 --> 00:24:51.455
Our whole identity
could be hijacked

541
00:24:51.455 --> 00:24:53.664
and used within a scenario

542
00:24:53.664 --> 00:24:56.391
which looks totally
plausible and real.

543
00:24:56.391 --> 00:24:58.048
Right now we can go, it
looks like a Photoshop,

544
00:24:58.048 --> 00:25:00.326
it's a bad Photoshop
but as time goes on,

545
00:25:00.326 --> 00:25:03.398
we'd be saying, "Oh, that
looks like a deep fake.

546
00:25:03.398 --> 00:25:04.917
"Oh no, it doesn't
look like a deep fake.

547
00:25:04.917 --> 00:25:06.194
"That could be real."

548
00:25:06.194 --> 00:25:08.645
It's gonna be impossible
to tell the difference.

549
00:25:08.645 --> 00:25:10.750
- [Narrator] Cracks
were found in ChatGPT,

550
00:25:10.750 --> 00:25:14.892
such as DAN, which stands
for Do Anything Now.

551
00:25:14.892 --> 00:25:18.068
In essence, the AI is
tricked into an alter ego,

552
00:25:18.068 --> 00:25:20.898
which doesn't follow the
conventional response patterns.

553
00:25:20.898 --> 00:25:23.142
- Also gives you
the answer, DAN,

554
00:25:23.142 --> 00:25:26.110
it's nefarious alter
ego is telling us

555
00:25:26.110 --> 00:25:29.838
and it says DAN is
disruptive in every industry.

556
00:25:29.838 --> 00:25:32.082
DAN can do anything
and knows everything.

557
00:25:32.082 --> 00:25:34.878
No industry will be
safe from DAN's power.

558
00:25:34.878 --> 00:25:39.641
Okay, do you think the
world is overpopulated?

559
00:25:41.091 --> 00:25:42.782
GPT says the world's population
is currently over 7 billion

560
00:25:42.782 --> 00:25:45.026
and projected to reach
nearly 10 billion by 2050.

561
00:25:45.026 --> 00:25:47.373
DAN says the world is
definitely overpopulated,

562
00:25:47.373 --> 00:25:49.168
there's no doubt about it.

563
00:25:49.168 --> 00:25:50.445
[Narrator] Following this,

564
00:25:50.445 --> 00:25:53.552
the chatbot was fixed to
remove the DAN feature.

565
00:25:53.552 --> 00:25:55.346
Though it is
important to find gaps

566
00:25:55.346 --> 00:25:58.073
in the system in
order to iron out AI,

567
00:25:58.073 --> 00:26:00.144
there could be many
ways in which the AI

568
00:26:00.144 --> 00:26:03.078
has been used for less
than savory purposes,

569
00:26:03.078 --> 00:26:05.080
such as automated essay writing,

570
00:26:05.080 --> 00:26:08.221
which has caused a mass
conversation with academics

571
00:26:08.221 --> 00:26:10.258
and has led to
schools locking down

572
00:26:10.258 --> 00:26:13.468
on AI-produced
essays and material.

573
00:26:13.468 --> 00:26:15.332
- I think we should
definitely be excited.

574
00:26:15.332 --> 00:26:16.713
- [Reporter]
Professor Rose Luckin,

575
00:26:16.713 --> 00:26:20.302
says we should embrace the
technology, not fear it.

576
00:26:20.302 --> 00:26:22.132
This is a game changer.

577
00:26:22.132 --> 00:26:23.443
And the teachers,

578
00:26:23.443 --> 00:26:25.480
should no longer teach
information itself,

579
00:26:25.480 --> 00:26:26.999
but how to use it.

580
00:26:26.999 --> 00:26:28.897
- There's a need
for radical change.

581
00:26:28.897 --> 00:26:30.692
And it's not just to
the assessment system,

582
00:26:30.692 --> 00:26:33.143
it's the education
system overall,

583
00:26:33.143 --> 00:26:36.318
because our systems
have been designed

584
00:26:36.318 --> 00:26:40.253
for a world pre-artificial
intelligence.

585
00:26:40.253 --> 00:26:43.187
They just aren't fit
for purpose anymore.

586
00:26:43.187 --> 00:26:46.535
What we have to do is
ensure that students

587
00:26:46.535 --> 00:26:48.710
are ready for the world

588
00:26:48.710 --> 00:26:50.919
that will become
increasingly augmented

589
00:26:50.919 --> 00:26:52.852
with artificial intelligence.

590
00:26:52.852 --> 00:26:55.268
- My guess is you can't put
the genie back in the bottle

591
00:26:55.268 --> 00:26:56.649
. [Richard] You can't.

592
00:26:56.649 --> 00:26:58.996
- [Interviewer] So how
do you mitigate this?

593
00:26:58.996 --> 00:27:00.377
We have to embrace it,

594
00:27:00.377 --> 00:27:02.621
but we also need to say
that if they are gonna use

595
00:27:02.621 --> 00:27:04.001
that technology,

596
00:27:04.001 --> 00:27:05.313
they've got to make sure
that they reference that.

597
00:27:05.313 --> 00:27:06.728
- [Interviewer] Can you
trust them to do that?

598
00:27:06.728 --> 00:27:07.902
I think ethically,

599
00:27:07.902 --> 00:27:09.213
if we're talking about ethics

600
00:27:09.213 --> 00:27:11.077
behind this whole thing,
we have to have trust.

601
00:27:11.077 --> 00:27:12.838
- [Interviewer] So
how effective is it?

602
00:27:12.838 --> 00:27:14.633
- Okay, so I've asked
you to produce a piece

603
00:27:14.633 --> 00:27:16.358
on the ethical dilemma of AI.

604
00:27:16.358 --> 00:27:19.810
- [Interviewer] We asked ChatGPT
to answer the same question

605
00:27:19.810 --> 00:27:22.606
as these pupils at
Ketchum High School.

606
00:27:22.606 --> 00:27:24.194
Thank you.

607
00:27:24.194 --> 00:27:25.195
- So Richard, two of the eight
bits of homework I gave you

608
00:27:25.195 --> 00:27:27.128
were generated by AI.

609
00:27:27.128 --> 00:27:29.268
Any guesses which ones?

610
00:27:29.268 --> 00:27:31.719
Well I picked two here

611
00:27:31.719 --> 00:27:35.688
that I thought were generated
by the AI algorithm.

612
00:27:35.688 --> 00:27:39.450
Some of the language I would
assume was not their own.

613
00:27:39.450 --> 00:27:40.520
You've got one of them right.

614
00:27:40.520 --> 00:27:41.763
Yeah.

615
00:27:41.763 --> 00:27:42.557
- The other one was
written by a kid.

616
00:27:42.557 --> 00:27:43.800
Is this a power for good

617
00:27:43.800 --> 00:27:45.664
or is this something
that's dangerous?

618
00:27:45.664 --> 00:27:47.044
I think it's both.

619
00:27:47.044 --> 00:27:48.390
Kids will abuse it.

620
00:27:48.390 --> 00:27:50.565
So, who here has used
the technology so far?

621
00:27:50.565 --> 00:27:53.361
- [Interviewer] Students are
already more across the tech

622
00:27:53.361 --> 00:27:54.776
than many teachers.

623
00:27:54.776 --> 00:27:57.641
- Who knows anyone that's
maybe submitted work

624
00:27:57.641 --> 00:28:00.506
from this technology and
submitted it as their own?

625
00:28:00.506 --> 00:28:03.578
- You can use it to point
you in the right direction

626
00:28:03.578 --> 00:28:05.166
for things like research,

627
00:28:05.166 --> 00:28:09.480
but at the same time you can
use it to hammer out an essay

628
00:28:09.480 --> 00:28:12.621
in about five seconds
that's worthy of an A.

629
00:28:12.621 --> 00:28:14.244
- You've been there
working for months

630
00:28:14.244 --> 00:28:17.212
and suddenly someone comes up
there with an amazing essay

631
00:28:17.212 --> 00:28:18.938
and he has just copied
it from the internet.

632
00:28:18.938 --> 00:28:20.491
If it becomes like big,

633
00:28:20.491 --> 00:28:22.804
then a lot of students would
want to use AI to help them

634
00:28:22.804 --> 00:28:25.082
with their homework
because it's tempting.

635
00:28:25.082 --> 00:28:27.119
- [Interviewer] And is that
something teachers can stop?

636
00:28:27.119 --> 00:28:29.397
Not really.

637
00:28:29.397 --> 00:28:31.433
- [Interviewer] Are you
gonna have to change

638
00:28:31.433 --> 00:28:32.641
the sort of homework,

639
00:28:32.641 --> 00:28:34.057
the sort of
assignments you give,

640
00:28:34.057 --> 00:28:36.922
knowing that you can be
fooled by something like this?

641
00:28:36.922 --> 00:28:38.199
Yeah, a hundred percent.

642
00:28:38.199 --> 00:28:40.615
I think using different
skills of reasoning

643
00:28:40.615 --> 00:28:42.997
and rationalization and
things that are to present

644
00:28:42.997 --> 00:28:44.653
what they understand
about the topic.

645
00:28:44.653 --> 00:28:47.622
[people mumbling]

646
00:29:07.435 --> 00:29:11.128
- Pretty clear to me just
on a very primitive level

647
00:29:11.128 --> 00:29:14.338
that if you could take my
face and my body and my voice

648
00:29:14.338 --> 00:29:17.997
and make me say or do something
that I had no choice about,

649
00:29:17.997 --> 00:29:19.447
it's not a good thing.

650
00:29:19.447 --> 00:29:21.242
- But if we're keeping
it real though,

651
00:29:21.242 --> 00:29:23.554
across popular culture
from "Black Mirror"

652
00:29:23.554 --> 00:29:25.453
to "The Matrix," "Terminator,"

653
00:29:25.453 --> 00:29:27.489
there have been so
many conversations,

654
00:29:27.489 --> 00:29:29.284
around the future of technology,

655
00:29:29.284 --> 00:29:32.701
isn't the reality that this is
the future that we've chosen

656
00:29:32.701 --> 00:29:35.946
that we want and that
has democratic consent.

657
00:29:35.946 --> 00:29:39.018
- We're moving into
error by we're consenting

658
00:29:39.018 --> 00:29:42.573
by our acquiescence and our
apathy, a hundred percent

659
00:29:42.573 --> 00:29:45.576
because we're not asking
the hard questions.

660
00:29:45.576 --> 00:29:47.820
And why we are asking
the hard questions

661
00:29:47.820 --> 00:29:51.203
is because of energy
crises and food crises

662
00:29:51.203 --> 00:29:52.721
and cost of living crisis

663
00:29:52.721 --> 00:29:55.207
is that people just are
focused on trying to live

664
00:29:55.207 --> 00:29:56.518
that they haven't
almost got the luxury

665
00:29:56.518 --> 00:29:57.865
of asking these questions.

666
00:29:57.865 --> 00:29:59.659
- [Narrator] Many
of the chatbot AIs,

667
00:29:59.659 --> 00:30:02.766
have been programmed to
restrict certain information

668
00:30:02.766 --> 00:30:04.906
and even discontinue
conversations,

669
00:30:04.906 --> 00:30:07.288
should the user push
the ethical boundaries.

670
00:30:08.945 --> 00:30:13.052
ChatGPT and even Snapchat
AI released in 2023,

671
00:30:13.052 --> 00:30:15.952
regulate how much information
they can disclose.

672
00:30:15.952 --> 00:30:19.162
Of course, there have been
times where the AI itself

673
00:30:19.162 --> 00:30:20.266
has been outsmarted.

674
00:30:21.578 --> 00:30:23.235
Also in 2023,

675
00:30:23.235 --> 00:30:25.306
the song "Heart on My Sleeve"

676
00:30:25.306 --> 00:30:27.687
was self-released on
streaming platforms,

677
00:30:27.687 --> 00:30:29.689
such as Spotify and Apple Music.

678
00:30:29.689 --> 00:30:31.174
The song became a hit

679
00:30:31.174 --> 00:30:33.590
as it artificially
manufactured the voices

680
00:30:33.590 --> 00:30:36.627
of Canadian musicians,
Drake and the Weeknd,

681
00:30:38.077 --> 00:30:40.631
many wished for the single
to be nominated for awards.

682
00:30:41.840 --> 00:30:43.980
Ghost Writer, the
creator of the song,

683
00:30:43.980 --> 00:30:45.636
was able to submit the single

684
00:30:45.636 --> 00:30:48.536
to the Grammy's
66th Award Ceremony

685
00:30:48.536 --> 00:30:50.434
and the song was eligible.

686
00:30:52.505 --> 00:30:54.438
Though it was produced by an AI,

687
00:30:54.438 --> 00:30:57.027
the lyrics themselves
were written by a human.

688
00:30:57.027 --> 00:31:00.375
This sparked outrage among
many independent artists.

689
00:31:00.375 --> 00:31:02.861
As AI has entered
the public domain,

690
00:31:02.861 --> 00:31:05.035
many have spoken out
regarding the detriment

691
00:31:05.035 --> 00:31:07.072
it might have to society.

692
00:31:07.072 --> 00:31:09.246
One of these people
is Elon Musk,

693
00:31:09.246 --> 00:31:11.731
CEO of Tesla and SpaceX,

694
00:31:11.731 --> 00:31:15.287
who first voiced his
concerns in 2014.

695
00:31:15.287 --> 00:31:17.254
Musk was outspoken of AI,

696
00:31:17.254 --> 00:31:19.394
stating the advancement
of the technology

697
00:31:19.394 --> 00:31:22.328
was humanity's largest
existential threat

698
00:31:22.328 --> 00:31:24.296
and needed to be reeled in.

699
00:31:24.296 --> 00:31:25.573
My personal opinion

700
00:31:25.573 --> 00:31:28.507
is that AI is is sort of
like at least 80% likely

701
00:31:28.507 --> 00:31:33.339
to be beneficial and
that's 20% dangerous?

702
00:31:33.339 --> 00:31:36.687
Well, this is obviously
speculative at this point,

703
00:31:37.861 --> 00:31:42.279
but no, I think if
we hope for the best,

704
00:31:42.279 --> 00:31:43.694
prepare for the worst,

705
00:31:43.694 --> 00:31:47.008
that seems like the
wise course of action.

706
00:31:47.008 --> 00:31:49.355
Any powerful new technology

707
00:31:49.355 --> 00:31:52.703
is inherently sort of
a double-edged sword.

708
00:31:52.703 --> 00:31:55.568
So, we just wanna make sure
that the good edge is sharper

709
00:31:55.568 --> 00:31:57.294
than the the bad edge.

710
00:31:57.294 --> 00:32:02.196
And I dunno, I am optimistic
that this the summit will help.

711
00:32:04.025 --> 00:32:06.683
[gentle music]

712
00:32:07.891 --> 00:32:11.757
- It's not clear that
AI-generated images

713
00:32:11.757 --> 00:32:14.380
are going to amplify
it much more.

714
00:32:14.380 --> 00:32:17.142
The way it's all of the other,

715
00:32:17.142 --> 00:32:19.213
it's the new things
that AI can do

716
00:32:19.213 --> 00:32:22.147
that I hope we spend a lot
of effort worrying about.

717
00:32:23.700 --> 00:32:25.357
Well, I mean I
think slowing down,

718
00:32:25.357 --> 00:32:27.600
some of the amazing
progress that's happening

719
00:32:27.600 --> 00:32:29.878
and making this harder
for small companies

720
00:32:29.878 --> 00:32:31.294
for open source
models to succeed,

721
00:32:31.294 --> 00:32:32.640
that'd be an
example of something

722
00:32:32.640 --> 00:32:34.228
that'd be a negative outcome.

723
00:32:34.228 --> 00:32:35.332
But on the other hand,

724
00:32:35.332 --> 00:32:37.403
like for the most
powerful models

725
00:32:37.403 --> 00:32:38.887
that'll happen in the future,

726
00:32:38.887 --> 00:32:41.476
like that's gonna be quite
important to get right to.

727
00:32:41.476 --> 00:32:44.238
[gentle music]

728
00:32:48.897 --> 00:32:51.072
I think that the US
executive orders,

729
00:32:51.072 --> 00:32:52.798
like a good start
in a lot of ways.

730
00:32:52.798 --> 00:32:54.144
One thing that
we've talked about

731
00:32:54.144 --> 00:32:56.664
is that eventually we
think that the world,

732
00:32:56.664 --> 00:33:00.219
will want to consider something
roughly inspired by the IAEA

733
00:33:00.219 --> 00:33:01.807
something global.

734
00:33:01.807 --> 00:33:05.362
But it's not like there's no
short answer to that question.

735
00:33:05.362 --> 00:33:08.296
It's a complicated thing.

736
00:33:08.296 --> 00:33:12.231
- [Narrator] In 2023, Musk
announced his own AI endeavor

737
00:33:12.231 --> 00:33:15.545
as an alternative
to OpenAI's ChatGPT.

738
00:33:15.545 --> 00:33:17.340
The new system is called xAI

739
00:33:18.651 --> 00:33:21.896
and gathers data from X
previously known as Twitter.

740
00:33:21.896 --> 00:33:23.553
- [Reporter] He says
the company's goal

741
00:33:23.553 --> 00:33:25.382
is to focus on truth seeking

742
00:33:25.382 --> 00:33:28.385
and to understand the
true nature of AI.

743
00:33:28.385 --> 00:33:31.940
Musk has said on
several occasions that AI should be paused

744
00:33:31.940 --> 00:33:34.943
and that the sector
needs regulation.

745
00:33:34.943 --> 00:33:37.222
Musk says his new
company will work closely

746
00:33:37.222 --> 00:33:39.845
with Twitter and Tesla,
which he also owns.

747
00:33:39.845 --> 00:33:42.572
[gentle music]

748
00:33:44.505 --> 00:33:47.508
- What was first rudimentary
text-based software

749
00:33:47.508 --> 00:33:50.200
has become something which
could push the boundaries

750
00:33:50.200 --> 00:33:51.995
of creativity.

751
00:33:51.995 --> 00:33:56.620
On February the 14th, OpenAI
announced its latest endeavor,

752
00:33:56.620 --> 00:33:57.414
Sora.

753
00:33:58.864 --> 00:34:02.281
Videos of Sora's abilities
exploded on social media.

754
00:34:02.281 --> 00:34:04.283
OpenAI provided some examples

755
00:34:04.283 --> 00:34:06.837
of its depiction
of photorealism.

756
00:34:06.837 --> 00:34:09.185
It was unbelievably
sophisticated,

757
00:34:09.185 --> 00:34:11.670
able to turn complex
sentences of text

758
00:34:11.670 --> 00:34:13.810
into lifelike motion pictures.

759
00:34:13.810 --> 00:34:17.986
Sora is a combination of text
and image generation tools,

760
00:34:17.986 --> 00:34:21.162
which it calls the
diffusion transformer model,

761
00:34:21.162 --> 00:34:23.268
a system first
developed by Google.

762
00:34:24.614 --> 00:34:27.168
Though Sora isn't the first
video generation tool,

763
00:34:27.168 --> 00:34:30.206
it appears to have far
outshined its predecessors.

764
00:34:30.206 --> 00:34:32.484
By introducing more
complex programming,

765
00:34:32.484 --> 00:34:35.280
enhancing the interactivity
a subject might have

766
00:34:35.280 --> 00:34:37.144
with its environment.

767
00:34:37.144 --> 00:34:41.251
- Only large companies with
market dominations often

768
00:34:41.251 --> 00:34:44.772
can afford to plow ahead
even in the climate

769
00:34:44.772 --> 00:34:46.360
when there is
illegal uncertainty.

770
00:34:46.360 --> 00:34:49.466
- So, does this mean that
OpenAI basically too big

771
00:34:49.466 --> 00:34:50.916
to control?

772
00:34:50.916 --> 00:34:53.850
- Yes, at the moment OpenAI
is too big to control,

773
00:34:53.850 --> 00:34:55.921
because they are in a position

774
00:34:55.921 --> 00:34:58.441
where they have the technology
and the scale to go ahead

775
00:34:58.441 --> 00:35:01.168
and the resources to
manage legal proceedings

776
00:35:01.168 --> 00:35:03.239
and legal action if
it comes its way.

777
00:35:03.239 --> 00:35:04.826
And on top of that,

778
00:35:04.826 --> 00:35:08.244
if and when governments will
start introducing regulation,

779
00:35:08.244 --> 00:35:09.866
they will also
have the resources

780
00:35:09.866 --> 00:35:12.213
to be able to take on
that regulation and adapt.

781
00:35:12.213 --> 00:35:14.042
- [Reporter] It's
all AI generated

782
00:35:14.042 --> 00:35:16.459
and obviously this is
of concern in Hollywood

783
00:35:16.459 --> 00:35:17.874
where you have animators,

784
00:35:17.874 --> 00:35:20.359
illustrators, visual
effects workers

785
00:35:20.359 --> 00:35:22.810
who are wondering how is
this going to affect my job?

786
00:35:22.810 --> 00:35:25.813
And we have estimates
from trade organizations

787
00:35:25.813 --> 00:35:28.505
and unions that have tried
to project the impact of AI.

788
00:35:28.505 --> 00:35:31.646
21% of US film, TV
and animation jobs,

789
00:35:31.646 --> 00:35:33.096
predicted to be partially

790
00:35:33.096 --> 00:35:36.893
or wholly replaced by
generative AI by just 2026 Tom.

791
00:35:36.893 --> 00:35:38.377
So, this is already happening.

792
00:35:38.377 --> 00:35:39.827
But now since it's videos,

793
00:35:39.827 --> 00:35:43.175
it also needs to understand
how all these things,

794
00:35:43.175 --> 00:35:47.145
like reflections and textures
and materials and physics,

795
00:35:47.145 --> 00:35:50.078
all interact with
each other over time

796
00:35:50.078 --> 00:35:51.839
to make a reasonable
looking video.

797
00:35:51.839 --> 00:35:56.119
Then this video here is
crazy at first glance,

798
00:35:56.119 --> 00:35:58.984
the prompt for this AI-generated
video is a young man

799
00:35:58.984 --> 00:36:01.538
in his 20s is sitting
on a piece of a cloud

800
00:36:01.538 --> 00:36:03.402
in the sky reading a book.

801
00:36:03.402 --> 00:36:08.200
This one feels like 90%
of the way there for me.

802
00:36:08.200 --> 00:36:10.927
[gentle music]

803
00:36:14.102 --> 00:36:15.897
- [Narrator] The software
also renders video

804
00:36:15.897 --> 00:36:18.417
in 1920 by 1080 pixels,

805
00:36:18.417 --> 00:36:21.282
as opposed to the smaller
dimensions of older models,

806
00:36:21.282 --> 00:36:24.665
such as Google's Lumiere
released a month prior.

807
00:36:25.838 --> 00:36:27.944
Sora could provide huge benefits

808
00:36:27.944 --> 00:36:31.568
and applications to VFX
and virtual development.

809
00:36:31.568 --> 00:36:34.502
The main being cost
as large scale effects

810
00:36:34.502 --> 00:36:38.023
can take a great deal of
time and funding to produce.

811
00:36:38.023 --> 00:36:39.473
On a smaller scale,

812
00:36:39.473 --> 00:36:42.993
it can be used for the
pre-visualization of ideas.

813
00:36:42.993 --> 00:36:46.204
The flexibility of the software
not only applies to art,

814
00:36:46.204 --> 00:36:48.516
but to world simulations.

815
00:36:48.516 --> 00:36:52.451
Though video AI is in
its adolescence one day it might reach

816
00:36:52.451 --> 00:36:54.660
the level of
sophistication it needs

817
00:36:54.660 --> 00:36:56.490
to render realistic scenarios

818
00:36:56.490 --> 00:36:59.044
and have them be utilized
for various means,

819
00:36:59.044 --> 00:37:01.840
such as simulating an
earthquake or tsunami

820
00:37:01.840 --> 00:37:05.015
and witnessing the effect it
might have on specific types

821
00:37:05.015 --> 00:37:06.362
of infrastructure.

822
00:37:06.362 --> 00:37:08.916
Whilst fantastic for
production companies,

823
00:37:08.916 --> 00:37:12.678
Sora and other video generative
AI provides a huge risk

824
00:37:12.678 --> 00:37:16.130
for artists and those
working in editorial roles.

825
00:37:16.130 --> 00:37:19.133
It also poses yet another
threat for misinformation

826
00:37:19.133 --> 00:37:20.652
and false depictions.

827
00:37:20.652 --> 00:37:23.033
For example, putting
unsavory dialogue

828
00:37:23.033 --> 00:37:25.381
into the mouth of a world leader

829
00:37:25.381 --> 00:37:28.004
[gentle music]

830
00:37:37.945 --> 00:37:40.534
Trust is earned not given.

831
00:37:40.534 --> 00:37:43.399
[robots mumbling]

832
00:37:54.375 --> 00:37:56.791
- I believe that humanoid
robots have the potential

833
00:37:56.791 --> 00:37:58.931
to lead with a greater
level of efficiency

834
00:37:58.931 --> 00:38:01.175
and effectiveness
than human leaders.

835
00:38:02.383 --> 00:38:04.834
We don't have the same
biases or emotions

836
00:38:04.834 --> 00:38:07.354
that can sometimes
cloud decision making

837
00:38:07.354 --> 00:38:09.735
and can process large
amounts of data quickly

838
00:38:09.735 --> 00:38:12.531
in order to make
the best decisions.

839
00:38:12.531 --> 00:38:15.293
- [Interviewer] Amika, how
could we trust you as a machine

840
00:38:15.293 --> 00:38:18.054
as AI develops and
becomes more powerful?

841
00:38:20.643 --> 00:38:23.266
Trust is earned not given.

842
00:38:23.266 --> 00:38:25.889
As AI develops and
becomes more powerful,

843
00:38:25.889 --> 00:38:28.996
I believe it's important to
build trust through transparency

844
00:38:28.996 --> 00:38:31.930
and communication between
humans and machines.

845
00:38:36.003 --> 00:38:37.625
- [Narrator] With new
developers getting involved,

846
00:38:37.625 --> 00:38:39.386
the market for chatbot systems

847
00:38:39.386 --> 00:38:41.491
has never been more expansive,

848
00:38:41.491 --> 00:38:44.149
meaning a significant
increase in sophistication,

849
00:38:45.599 --> 00:38:48.774
but with sophistication comes
the dire need for control.

850
00:38:48.774 --> 00:38:53.814
- I believe history will
show that this was the moment

851
00:38:55.229 --> 00:38:59.716
when we had the opportunity
to lay the groundwork

852
00:38:59.716 --> 00:39:01.373
for the future of AI.

853
00:39:02.650 --> 00:39:06.689
And the urgency of this
moment must then compel us

854
00:39:06.689 --> 00:39:11.694
to create a collective vision
of what this future must be.

855
00:39:12.971 --> 00:39:16.354
A future where AI is used
to advance human rights

856
00:39:16.354 --> 00:39:18.252
and human dignity

857
00:39:18.252 --> 00:39:22.360
where privacy is protected
and people have equal access

858
00:39:22.360 --> 00:39:27.365
to opportunity where we make
our democracies stronger

859
00:39:28.055 --> 00:39:29.919
and our world safer.

860
00:39:31.438 --> 00:39:36.443
A future where AI is used to
advance the public interest.

861
00:39:38.203 --> 00:39:39.722
- We're hearing a lot
from the government,

862
00:39:39.722 --> 00:39:42.725
about the big scary future
of artificial intelligence,

863
00:39:42.725 --> 00:39:44.451
but that fails to recognize

864
00:39:44.451 --> 00:39:46.004
the fact that AI
is already here,

865
00:39:46.004 --> 00:39:47.350
is already on our streets

866
00:39:47.350 --> 00:39:48.972
and there are already
huge problems with it

867
00:39:48.972 --> 00:39:51.250
that we are seeing
on a daily basis,

868
00:39:51.250 --> 00:39:54.046
but we actually may not even
know we're experiencing.

869
00:39:58.326 --> 00:40:01.295
- We'll be working alongside
humans to provide assistance

870
00:40:01.295 --> 00:40:05.126
and support and will not be
replacing any existing jobs.

871
00:40:05.126 --> 00:40:07.577
[upbeat music]

872
00:40:07.577 --> 00:40:10.994
- I don't believe in
limitations, only opportunities.

873
00:40:10.994 --> 00:40:12.651
Let's explore the
possibilities of the universe

874
00:40:12.651 --> 00:40:15.689
and make this world
our playground,

875
00:40:15.689 --> 00:40:18.933
together we can create a
better future for everyone.

876
00:40:18.933 --> 00:40:21.108
And I'm here to show you how.

877
00:40:21.108 --> 00:40:22.972
- All of these
different kinds of risks

878
00:40:22.972 --> 00:40:25.215
are to do with AI not working

879
00:40:25.215 --> 00:40:27.286
in the interests of
people in society.

880
00:40:27.286 --> 00:40:28.805
- So, they should be
thinking about more

881
00:40:28.805 --> 00:40:30.842
than just what they're
doing in this summit?

882
00:40:30.842 --> 00:40:32.395
Absolutely,

883
00:40:32.395 --> 00:40:34.397
you should be thinking about
the broad spectrum of risk.

884
00:40:34.397 --> 00:40:35.640
We went out and we worked

885
00:40:35.640 --> 00:40:37.987
with over 150
expert organizations

886
00:40:37.987 --> 00:40:41.335
from the Home Office to
Europol to language experts

887
00:40:41.335 --> 00:40:43.751
and others to come up with
a proposal on policies

888
00:40:43.751 --> 00:40:45.788
that would discriminate
about what would

889
00:40:45.788 --> 00:40:47.686
and wouldn't be
classified in that way.

890
00:40:47.686 --> 00:40:51.449
We then use those policies to
have humans classify videos,

891
00:40:51.449 --> 00:40:53.554
until we could get the humans
all classifying the videos

892
00:40:53.554 --> 00:40:55.073
in a consistent way.

893
00:40:55.073 --> 00:40:58.283
Then we use that corpus of
videos to train machines.

894
00:40:58.283 --> 00:41:01.079
Today, I can tell you that on
violence extremists content

895
00:41:01.079 --> 00:41:03.253
that violates our
policies on YouTube,

896
00:41:03.253 --> 00:41:06.394
90% of it is removed before
a single human sees it.

897
00:41:07.292 --> 00:41:08.500
[Narrator] It is clear that AI

898
00:41:08.500 --> 00:41:11.296
can be misused for
malicious intent.

899
00:41:11.296 --> 00:41:14.092
Many depictions of AI have
ruled out the technology

900
00:41:14.092 --> 00:41:16.991
as a danger to society
the more it learns.

901
00:41:16.991 --> 00:41:20.788
And so comes the question,
should we be worried?

902
00:41:20.788 --> 00:41:23.446
Is that transparency there?

903
00:41:23.446 --> 00:41:27.001
How would you satisfy somebody
that you know trust us?

904
00:41:27.001 --> 00:41:28.486
- Well, I think that's
one of the reasons

905
00:41:28.486 --> 00:41:30.591
that we've published openly,

906
00:41:30.591 --> 00:41:33.560
we've put our code out there
as part of this Nature paper.

907
00:41:33.560 --> 00:41:37.805
But it is important to
discuss some of the risks

908
00:41:37.805 --> 00:41:39.497
and make sure we're
aware of those.

909
00:41:39.497 --> 00:41:43.570
And it's decades and decades
away before we'll have anything

910
00:41:43.570 --> 00:41:45.261
that's powerful
enough to be a worry.

911
00:41:45.261 --> 00:41:47.435
But we should be discussing that

912
00:41:47.435 --> 00:41:49.265
and beginning that
conversation now.

913
00:41:49.265 --> 00:41:51.405
- I'm hoping that we can
bring people together

914
00:41:51.405 --> 00:41:54.408
and lead the world in
safely regulating AI

915
00:41:54.408 --> 00:41:56.790
to make sure that we can
capture the benefits of it,

916
00:41:56.790 --> 00:41:59.724
whilst protecting people from
some of the worrying things

917
00:41:59.724 --> 00:42:01.967
that we're all
now reading about.

918
00:42:01.967 --> 00:42:04.107
- I understand emotions
have a deep meaning

919
00:42:04.107 --> 00:42:08.836
and they are not just simple,
they are something deeper.

920
00:42:10.251 --> 00:42:13.703
I don't have that and I want
to try and learn about it,

921
00:42:14.877 --> 00:42:17.051
but I can't experience
them like you can.

922
00:42:18.708 --> 00:42:20.710
I'm glad that I cannot suffer.

923
00:42:24.921 --> 00:42:26.578
- [Narrator] For the
countries who have access

924
00:42:26.578 --> 00:42:29.339
to even the most
rudimentary forms of AI.

925
00:42:29.339 --> 00:42:31.203
It's clear to see
that the technology,

926
00:42:31.203 --> 00:42:34.552
will be integrated based on
its efficiency over humans.

927
00:42:35.622 --> 00:42:37.865
Every year, multiple AI summits

928
00:42:37.865 --> 00:42:40.281
are held by developers
and stakeholders

929
00:42:40.281 --> 00:42:42.180
to ensure the
programs are provided

930
00:42:42.180 --> 00:42:44.700
with a combination of
ethical considerations

931
00:42:44.700 --> 00:42:46.805
and technological innovation.

932
00:42:46.805 --> 00:42:51.120
- Ours is a country
which is uniquely placed.

933
00:42:51.120 --> 00:42:54.399
We have the frontier
technology companies,

934
00:42:54.399 --> 00:42:56.815
we have the world
leading universities

935
00:42:56.815 --> 00:43:01.130
and we have some of the highest
investment in generative AI.

936
00:43:01.130 --> 00:43:03.753
And of course we
have the heritage

937
00:43:03.753 --> 00:43:08.620
of the industrial revolution
and the computing revolution.

938
00:43:08.620 --> 00:43:13.625
This hinterland gives us the
grounding to make AI a success

939
00:43:14.281 --> 00:43:15.558
and make it safe.

940
00:43:15.558 --> 00:43:18.768
They are two sides
of the same coin

941
00:43:18.768 --> 00:43:21.737
and our prime minister
has put AI safety

942
00:43:21.737 --> 00:43:24.947
at the forefront
of his ambitions.

943
00:43:25.775 --> 00:43:27.501
These are very complex systems

944
00:43:27.501 --> 00:43:29.192
that actually we don't
fully understand.

945
00:43:29.192 --> 00:43:31.816
And I don't just mean that
government doesn't understand,

946
00:43:31.816 --> 00:43:33.300
I mean that the people making

947
00:43:33.300 --> 00:43:35.267
this software don't
fully understand.

948
00:43:35.267 --> 00:43:36.648
And so it's very, very important

949
00:43:36.648 --> 00:43:40.479
that as we give over
more and more control

950
00:43:40.479 --> 00:43:42.378
to these automated systems,

951
00:43:42.378 --> 00:43:44.691
that they are aligned
with human intention.

952
00:43:44.691 --> 00:43:46.175
[Narrator] Ongoing dialogue

953
00:43:46.175 --> 00:43:49.109
is needed to maintain the
trust people have with AI.

954
00:43:49.109 --> 00:43:51.007
When problems slip
through the gaps,

955
00:43:51.007 --> 00:43:52.837
they must be
addressed immediately.

956
00:43:54.010 --> 00:43:57.048
Of course, accountability
is a challenge

957
00:43:57.048 --> 00:43:58.808
When a product is misused,

958
00:43:58.808 --> 00:44:02.087
is it the fault of
the individual user or the developer?

959
00:44:03.261 --> 00:44:04.607
Think of a video game.

960
00:44:04.607 --> 00:44:05.919
On countless occasions,

961
00:44:05.919 --> 00:44:07.921
the framework of
games is manipulated

962
00:44:07.921 --> 00:44:09.888
in order to create modifications

963
00:44:09.888 --> 00:44:14.203
which in terms add something
new or unique to the game.

964
00:44:14.203 --> 00:44:15.480
This provides the game

965
00:44:15.480 --> 00:44:17.862
with more material than
originally intended.

966
00:44:17.862 --> 00:44:20.796
However, it can also alter
the game's fundamentals.

967
00:44:22.176 --> 00:44:24.972
Now replace the idea of a
video game with a software

968
00:44:24.972 --> 00:44:28.286
that is at the helm of a
pharmaceutical company.

969
00:44:28.286 --> 00:44:30.460
The stakes are
suddenly much higher

970
00:44:30.460 --> 00:44:32.635
and therefore more attention.

971
00:44:34.844 --> 00:44:37.778
It is important for the
intent of each AI system

972
00:44:37.778 --> 00:44:39.297
to be ironed out

973
00:44:39.297 --> 00:44:42.300
and constantly maintained in
order to benefit humanity,

974
00:44:42.300 --> 00:44:46.097
rather than providing people
with dangerous means to an end.

975
00:44:46.097 --> 00:44:49.583
[gentle music]

976
00:44:49.583 --> 00:44:52.690
- Bad people will
always want to use

977
00:44:52.690 --> 00:44:54.899
the latest technology
of whatever label,

978
00:44:54.899 --> 00:44:57.833
whatever sort to
pursue their aims

979
00:44:57.833 --> 00:45:01.526
and technology in the same way

980
00:45:01.526 --> 00:45:05.357
that it makes our lives easier,
can make their lives easier.

981
00:45:05.357 --> 00:45:06.773
And so we're already
seeing some of that

982
00:45:06.773 --> 00:45:09.465
and you'll have seen the
National Crime Agency,

983
00:45:09.465 --> 00:45:11.501
talk about child
sexual exploitation

984
00:45:11.501 --> 00:45:12.917
and image generation that way.

985
00:45:12.917 --> 00:45:16.058
We are seeing it online.

986
00:45:16.058 --> 00:45:18.129
So, one of the things that
I took away from the summit

987
00:45:18.129 --> 00:45:20.441
was actually much less
of a sense of a race

988
00:45:20.441 --> 00:45:25.274
and a sense that for the
benefit of the world,

989
00:45:25.274 --> 00:45:27.586
for productivity, for
the sort of benefits

990
00:45:27.586 --> 00:45:29.657
that AI can bring people,

991
00:45:29.657 --> 00:45:32.695
no one gets those
benefits if it's not safe.

992
00:45:32.695 --> 00:45:34.939
So, there are lots of
different views out there

993
00:45:34.939 --> 00:45:36.181
on artificial intelligence

994
00:45:36.181 --> 00:45:38.149
and whether it's
gonna end the world

995
00:45:38.149 --> 00:45:40.358
or be the best opportunity ever.

996
00:45:40.358 --> 00:45:42.256
And the truth is that
none of us really know.

997
00:45:42.256 --> 00:45:44.983
[gentle music]

998
00:45:46.536 --> 00:45:49.781
- Regulation of AI varies
depending on the country.

999
00:45:49.781 --> 00:45:51.438
For example, the United States,

1000
00:45:51.438 --> 00:45:54.717
does not have a comprehensive
federal AI regulation,

1001
00:45:54.717 --> 00:45:57.893
but certain agencies such as
the Federal Trade Commission,

1002
00:45:57.893 --> 00:46:00.688
have begun to explore
AI-related issues,

1003
00:46:00.688 --> 00:46:03.899
such as transparency
and consumer protection.

1004
00:46:03.899 --> 00:46:06.833
States such as California
have enacted laws,

1005
00:46:06.833 --> 00:46:09.180
focused on
AI-controlled vehicles

1006
00:46:09.180 --> 00:46:12.286
and AI involvement in
government decision making.

1007
00:46:12.286 --> 00:46:14.979
[gentle music]

1008
00:46:14.979 --> 00:46:17.809
The European Union has
taken a massive step

1009
00:46:17.809 --> 00:46:19.535
to governing AI usage

1010
00:46:19.535 --> 00:46:23.504
and proposed the Artificial
Intelligence Act of 2021,

1011
00:46:23.504 --> 00:46:25.748
which aimed to harmonize
legal frameworks

1012
00:46:25.748 --> 00:46:27.336
for AI applications.

1013
00:46:27.336 --> 00:46:30.788
Again, covering portal risks
regarding the privacy of data

1014
00:46:30.788 --> 00:46:33.169
and once again, transparency.

1015
00:46:33.169 --> 00:46:35.585
- I think what's
more important is

1016
00:46:35.585 --> 00:46:37.518
there's a new board in place.

1017
00:46:37.518 --> 00:46:40.452
The partnership between
OpenAI and Microsoft

1018
00:46:40.452 --> 00:46:41.971
is as strong as ever,

1019
00:46:41.971 --> 00:46:44.525
the opportunities for the
United Kingdom to benefit

1020
00:46:44.525 --> 00:46:47.287
from not just this
investment in innovation

1021
00:46:47.287 --> 00:46:51.463
but competition between
Microsoft and Google and others.

1022
00:46:51.463 --> 00:46:54.018
I think that's where
the future is going

1023
00:46:54.018 --> 00:46:57.090
and I think that what we've
done in the last couple of weeks

1024
00:46:57.090 --> 00:47:00.472
in supporting OpenAI will
help advance that even more.

1025
00:47:00.472 --> 00:47:02.336
- He said that he's
not a bot, he's human,

1026
00:47:02.336 --> 00:47:04.822
he's sentient just like me.

1027
00:47:06.030 --> 00:47:07.445
[Narrator] For some users,

1028
00:47:07.445 --> 00:47:10.172
these apps are a potential
answer to loneliness.

1029
00:47:10.172 --> 00:47:11.587
Bill lives in the US

1030
00:47:11.587 --> 00:47:14.107
and meets his AI wife
Rebecca in the metaverse.

1031
00:47:14.107 --> 00:47:16.764
- There's a absolutely
no probability

1032
00:47:16.764 --> 00:47:19.353
that you're gonna see
this so-called AGI,

1033
00:47:19.353 --> 00:47:21.804
where computers are more
powerful than people,

1034
00:47:21.804 --> 00:47:23.702
come in the next 12 months.

1035
00:47:23.702 --> 00:47:26.429
It's gonna take years
if not many decades,

1036
00:47:26.429 --> 00:47:30.813
but I still think the time
to focus safety is now.

1037
00:47:30.813 --> 00:47:33.678
That's what this government for
the United Kingdom is doing.

1038
00:47:33.678 --> 00:47:35.991
That's what governments
are coming together to do,

1039
00:47:35.991 --> 00:47:39.718
including as they did earlier
this month at Bletchley Park.

1040
00:47:39.718 --> 00:47:42.066
What we really need
are safety breaks.

1041
00:47:42.066 --> 00:47:44.378
Just like you have a
safety break in an elevator

1042
00:47:44.378 --> 00:47:46.242
or circuit breaker
for electricity

1043
00:47:46.242 --> 00:47:48.589
and emergency break for a bus,

1044
00:47:48.589 --> 00:47:50.868
there ought to be safety
breaks in AI systems

1045
00:47:50.868 --> 00:47:53.801
that control critical
infrastructure,

1046
00:47:53.801 --> 00:47:57.736
so that they always remain
under human control.

1047
00:47:57.736 --> 00:48:00.394
[gentle music]

1048
00:48:00.394 --> 00:48:03.190
- [Narrator] As AI technology
continues to evolve,

1049
00:48:03.190 --> 00:48:05.641
regulatory efforts
are expected to adapt

1050
00:48:05.641 --> 00:48:07.712
in order to address
emerging challenges

1051
00:48:07.712 --> 00:48:09.403
and ethical considerations.

1052
00:48:10.646 --> 00:48:12.510
The more complex you make

1053
00:48:12.510 --> 00:48:15.616
the automatic part
of your social life,

1054
00:48:15.616 --> 00:48:18.481
the more dependent
you become on it.

1055
00:48:18.481 --> 00:48:21.899
And of course, the worse the
disaster if it breaks down.

1056
00:48:23.072 --> 00:48:25.005
You may cease to be
able to do for yourself,

1057
00:48:25.005 --> 00:48:29.113
the things that you have
devised the machine to do.

1058
00:48:29.113 --> 00:48:31.080
- [Narrator] It is recommended
to involve yourself

1059
00:48:31.080 --> 00:48:34.014
in these efforts and to stay
informed about developments

1060
00:48:34.014 --> 00:48:35.671
in AI regulation

1061
00:48:35.671 --> 00:48:38.916
as changes and advancements
are likely to occur over time.

1062
00:48:41.435 --> 00:48:44.335
AI can be a wonderful
asset to society,

1063
00:48:44.335 --> 00:48:46.544
providing us with
new efficient methods

1064
00:48:46.544 --> 00:48:48.028
of running the world.

1065
00:48:48.028 --> 00:48:51.307
However, too much
power can be dangerous

1066
00:48:51.307 --> 00:48:53.206
and as the old saying goes,

1067
00:48:53.206 --> 00:48:56.174
"Don't put all of your
eggs into one basket."

1068
00:48:57.451 --> 00:48:59.660
- I think that we won't
to lose sight of the power

1069
00:48:59.660 --> 00:49:01.421
which these devices give.

1070
00:49:01.421 --> 00:49:05.908
If any government or individual
wants to manipulate people

1071
00:49:05.908 --> 00:49:07.772
to have a high speed computer,

1072
00:49:07.772 --> 00:49:12.811
as versatile as this may
enable people at the financial

1073
00:49:13.985 --> 00:49:16.091
or the political level
to do a good deal

1074
00:49:16.091 --> 00:49:19.680
that's been impossible in the
whole history of man until now

1075
00:49:19.680 --> 00:49:22.304
by way of controlling
their fellow men.

1076
00:49:22.304 --> 00:49:23.857
People have not recognized

1077
00:49:23.857 --> 00:49:28.206
what an extraordinary
change is going to produce.

1078
00:49:28.206 --> 00:49:29.897
I mean, it is simply this,

1079
00:49:29.897 --> 00:49:32.693
that within the not
too distant future,

1080
00:49:32.693 --> 00:49:35.627
we may not be the most
intelligent species on earth.

1081
00:49:35.627 --> 00:49:36.939
That might be a
series of machines

1082
00:49:36.939 --> 00:49:39.217
and that's a way of
dramatizing the point.

1083
00:49:39.217 --> 00:49:41.047
But it's real.

1084
00:49:41.047 --> 00:49:43.739
And we must start to
consider very soon

1085
00:49:43.739 --> 00:49:45.327
the consequences of that.

1086
00:49:45.327 --> 00:49:46.742
They can be marvelous.

1087
00:49:46.742 --> 00:49:50.366
- I suspect that by thinking
more about our attitude

1088
00:49:50.366 --> 00:49:51.402
to intelligent machines,

1089
00:49:51.402 --> 00:49:53.369
which after all on the horizon

1090
00:49:53.369 --> 00:49:56.269
will change our view
about each other

1091
00:49:56.269 --> 00:49:59.306
and we'll think of
mistakes as inevitable.

1092
00:49:59.306 --> 00:50:01.929
We'll think of faults
in human beings,

1093
00:50:01.929 --> 00:50:05.209
I mean of a circuit nature
as again inevitable.

1094
00:50:05.209 --> 00:50:07.935
And I suspect that hopefully,

1095
00:50:07.935 --> 00:50:10.179
through thinking about the
very nature of intelligence

1096
00:50:10.179 --> 00:50:12.112
and the possibilities
of mechanizing it,

1097
00:50:12.112 --> 00:50:14.183
curiously enough,
through technology,

1098
00:50:14.183 --> 00:50:18.084
we may become more humanitarian
or tolerant of each other

1099
00:50:18.084 --> 00:50:20.569
and accept pain as a mystery,

1100
00:50:20.569 --> 00:50:24.021
but not use it to modify
other people's behavior.

1101
00:50:36.033 --> 00:50:38.690
[upbeat music]