1
00:00:04,600 --> 00:00:08,360
Speaker 1: Sleepwalkers is a production of our Heart Radio and unusual productions.

2
00:00:12,360 --> 00:00:16,079
Speaker 1: So I'm here for a surprise poetry reading. It's about

3
00:00:16,120 --> 00:00:21,120
Speaker 1: to start. The silence is hardly final. Somewhere in the street,

4
00:00:21,400 --> 00:00:24,600
Speaker 1: I can see the trees begin to rise and fall

5
00:00:24,720 --> 00:00:27,319
Speaker 1: for the light of the dark thing above me. The

6
00:00:27,400 --> 00:00:29,600
Speaker 1: dream is like a shiny black hair, and the sun

7
00:00:29,760 --> 00:00:32,800
Speaker 1: is like a dream. I stand up and watch the

8
00:00:32,800 --> 00:00:35,800
Speaker 1: sun shine on a single day, and the sun has

9
00:00:35,840 --> 00:00:39,320
Speaker 1: a chance to accomplish from the springs of my own delight.

10
00:00:40,760 --> 00:00:45,960
Speaker 1: Kind of haunting, abstract, yes, but beautiful too. And crucially,

11
00:00:47,400 --> 00:00:50,360
Speaker 1: when I read this, I felt, as you just did.

12
00:00:50,400 --> 00:00:53,120
Speaker 1: I hope that it is beautiful. I found it evocative

13
00:00:53,240 --> 00:00:55,280
Speaker 1: of experience that I've had it in the past. I

14
00:00:55,280 --> 00:00:58,279
Speaker 1: found it nostalgic um and I'm like, oh my god,

15
00:00:58,320 --> 00:01:01,720
Speaker 1: I'm having real human being emotions. That was filmmaker Oscar Sharp,

16
00:01:02,000 --> 00:01:05,639
Speaker 1: and that poem wasn't written by him or by anyone else.

17
00:01:06,200 --> 00:01:10,119
Speaker 1: It was written by a computer, a machine poet. We're

18
00:01:10,120 --> 00:01:13,160
Speaker 1: more and more worried about robots coming to take our jobs.

19
00:01:13,400 --> 00:01:16,160
Speaker 1: And though perhaps few would regret less hips to poets

20
00:01:16,160 --> 00:01:20,280
Speaker 1: shambly around Brooklyn, somehow machines in the creative world are

21
00:01:20,480 --> 00:01:24,800
Speaker 1: especially uncanny, even frightening, because poetry and music and humor

22
00:01:24,840 --> 00:01:27,560
Speaker 1: are supposed to be the things that define our humanity,

23
00:01:27,600 --> 00:01:30,720
Speaker 1: aren't they. In this episode, we look at how AI

24
00:01:30,920 --> 00:01:33,520
Speaker 1: is being used in the creative arts, and in doing so,

25
00:01:33,959 --> 00:01:36,920
Speaker 1: we understand a lot more about how this often intimidating

26
00:01:36,959 --> 00:01:55,520
Speaker 1: technology actually works. I'm Moza Lush and welcome to Sleepwalkers. Hi,

27
00:01:56,040 --> 00:02:01,120
Speaker 1: Hi Karen So did Oscar's poems spring your delight? I

28
00:02:01,160 --> 00:02:04,040
Speaker 1: would have preferred it in your British act? Then I think, well,

29
00:02:04,040 --> 00:02:06,040
Speaker 1: I can't blame you for that. It did remind me

30
00:02:06,080 --> 00:02:08,360
Speaker 1: of Um. My friend is in a band called the

31
00:02:08,520 --> 00:02:12,040
Speaker 1: x Ambassadors and they partnered with this producer called Alex

32
00:02:12,160 --> 00:02:15,040
Speaker 1: the Kid who was asked by IBM to make a

33
00:02:15,080 --> 00:02:17,799
Speaker 1: song using Watson. And the song is not bad. The

34
00:02:17,840 --> 00:02:21,200
Speaker 1: song is not bad. Um, the sounds good. But basically,

35
00:02:21,840 --> 00:02:24,519
Speaker 1: the way in which they used Watson was that they

36
00:02:24,680 --> 00:02:29,640
Speaker 1: crunched the twenty six thousand songs from the top one charts.

37
00:02:30,400 --> 00:02:32,359
Speaker 1: It's from over what time period, like over a few

38
00:02:32,400 --> 00:02:35,960
Speaker 1: years presume, I don't know, But the point is that

39
00:02:36,000 --> 00:02:39,640
Speaker 1: they used them to discover patterns in the songs. What

40
00:02:39,800 --> 00:02:42,120
Speaker 1: makes the top hundred song. Basically that's right, and then

41
00:02:42,160 --> 00:02:44,560
Speaker 1: that's reproduce it right, which is interesting because I think

42
00:02:44,560 --> 00:02:46,640
Speaker 1: anybody could kind of tell you what makes the top

43
00:02:46,680 --> 00:02:49,040
Speaker 1: on hundred song right. They're called earworms. But when you

44
00:02:49,040 --> 00:02:52,200
Speaker 1: think about a data set of twenty six thousand, like,

45
00:02:52,320 --> 00:02:54,200
Speaker 1: no human being can listen to that many songs and

46
00:02:54,400 --> 00:02:57,320
Speaker 1: do any productive after they hear it. So in this episode,

47
00:02:57,360 --> 00:03:00,359
Speaker 1: we're going to look at AI and art and different

48
00:03:00,400 --> 00:03:03,800
Speaker 1: kinds of fields to really understand how computers crunched data

49
00:03:03,919 --> 00:03:06,840
Speaker 1: to crack open this creative code. But first I want

50
00:03:06,880 --> 00:03:09,400
Speaker 1: to go back to the algorithm who wrote the poem

51
00:03:09,440 --> 00:03:11,920
Speaker 1: we heard at the beginning of the episode, because that

52
00:03:12,000 --> 00:03:15,239
Speaker 1: algorithm also wrote the film that Steven Spielberg of the

53
00:03:15,280 --> 00:03:18,799
Speaker 1: machine learning world, and the algorithm is called Benjamin. So

54
00:03:18,880 --> 00:03:22,080
Speaker 1: we're going to meet Benjamin and a few people not

55
00:03:22,240 --> 00:03:26,480
Speaker 1: named Benjamin. I was a speech trator for John Kerry,

56
00:03:27,040 --> 00:03:30,000
Speaker 1: Tim Gaitner, and Barack Obama, um, not in that order,

57
00:03:30,080 --> 00:03:32,640
Speaker 1: And I'm essentially a ghost writer and a photographer who

58
00:03:32,760 --> 00:03:37,040
Speaker 1: learned to code one giant Frankenstein monster. That's Ross Goodwin

59
00:03:37,480 --> 00:03:41,240
Speaker 1: and that Frankenstein monster it goes by Benjamin and it's

60
00:03:41,280 --> 00:03:44,400
Speaker 1: the work of Ross Goodwin and Oscar Sharp, who you'll

61
00:03:44,400 --> 00:03:47,680
Speaker 1: remember from that poetry reading, but neither of them actually

62
00:03:47,800 --> 00:03:51,800
Speaker 1: named Benjamin. Well, they named itself, or rather, there was

63
00:03:51,840 --> 00:03:53,520
Speaker 1: a piece of paper that came out of it that

64
00:03:53,600 --> 00:03:55,840
Speaker 1: said my name is Benjamin on it. I read that

65
00:03:56,040 --> 00:03:58,360
Speaker 1: in response to a question that was put to it,

66
00:03:58,760 --> 00:04:02,480
Speaker 1: and a room full of people went, oh, a program

67
00:04:02,520 --> 00:04:06,080
Speaker 1: that names itself is rather uncanny. And Oscar had been

68
00:04:06,120 --> 00:04:08,400
Speaker 1: chasing that uncanny nous ever since he was at n

69
00:04:08,560 --> 00:04:11,280
Speaker 1: y U for graduate school. Whenever I met anyone who

70
00:04:11,280 --> 00:04:14,160
Speaker 1: good program, I would grab them by the pels and

71
00:04:14,280 --> 00:04:16,440
Speaker 1: yell into the face. Can you can you build something

72
00:04:16,480 --> 00:04:19,520
Speaker 1: that can write like people talk in some way? And

73
00:04:19,600 --> 00:04:22,720
Speaker 1: one day in class Oscar notices there's this sneakerhead and

74
00:04:22,720 --> 00:04:25,000
Speaker 1: he's sitting on his laptop and his laptop is writing

75
00:04:25,160 --> 00:04:30,080
Speaker 1: without him touching it. And I'm like, oh, so we go,

76
00:04:30,320 --> 00:04:33,360
Speaker 1: So we go for coffee, and that gut coffee was.

77
00:04:33,400 --> 00:04:35,560
Speaker 1: It was a lengthy coffee. We're still having that coffee.

78
00:04:35,560 --> 00:04:39,719
Speaker 1: We're still having such a cold coffee. By now, you

79
00:04:39,800 --> 00:04:41,960
Speaker 1: might not believe it if it happened in a film,

80
00:04:42,080 --> 00:04:44,800
Speaker 1: but Oscar had stumbled on exactly the person he was

81
00:04:44,839 --> 00:04:47,680
Speaker 1: looking for. Oscar came to me and he said, I

82
00:04:47,720 --> 00:04:50,239
Speaker 1: want to make a movie from a computer generated screenplay.

83
00:04:50,360 --> 00:04:52,120
Speaker 1: And I said, you know, of course that sounds amazing.

84
00:04:52,200 --> 00:04:54,280
Speaker 1: Let's do it. But let's figure out how we're going

85
00:04:54,279 --> 00:04:57,480
Speaker 1: to generate the screenplay, because that's a nuanced process with

86
00:04:57,520 --> 00:04:59,679
Speaker 1: lots of stabs, and we need to consider like every

87
00:04:59,680 --> 00:05:03,760
Speaker 1: part it. So Oscar volunteered himself to teach me all

88
00:05:03,880 --> 00:05:07,800
Speaker 1: the things about storytelling and narrative and filmmaking. He turned

89
00:05:07,800 --> 00:05:11,360
Speaker 1: me onto like Vladimir prop Joseph campbell Um, all these

90
00:05:11,400 --> 00:05:16,000
Speaker 1: theories of storytelling, and so they begin to experiment. I

91
00:05:16,040 --> 00:05:19,400
Speaker 1: tried a bunch of prototypes that used like various structures

92
00:05:19,440 --> 00:05:23,159
Speaker 1: that had been postulated by these theorists over time, and

93
00:05:23,360 --> 00:05:27,680
Speaker 1: the output was not interesting. Despite following the rules laid

94
00:05:27,680 --> 00:05:31,680
Speaker 1: out by narrative theorists, Ross couldn't get anything good just

95
00:05:31,760 --> 00:05:35,599
Speaker 1: telling his programs what a story should contain. So a

96
00:05:35,680 --> 00:05:39,000
Speaker 1: year passes, Oscar moves to l A and when I

97
00:05:39,040 --> 00:05:42,320
Speaker 1: get this email from Ross and it's the results in

98
00:05:42,360 --> 00:05:43,960
Speaker 1: one of those experiments that he wants me to read,

99
00:05:44,200 --> 00:05:47,240
Speaker 1: and read it he did. Rossity mailed the poem from

100
00:05:47,240 --> 00:05:50,480
Speaker 1: the beginning of the episode, the room is blown away

101
00:05:50,480 --> 00:05:53,240
Speaker 1: from the door and the stones are beginning to shine.

102
00:05:53,760 --> 00:05:55,839
Speaker 1: I immediately was like, oh my god, I don't know

103
00:05:55,839 --> 00:05:57,400
Speaker 1: how he's doing this. But he said, I don't know

104
00:05:57,400 --> 00:06:00,120
Speaker 1: what technology you're using right now, but can we it

105
00:06:00,200 --> 00:06:03,680
Speaker 1: for screenplay? And so they did, and not just a screenplay,

106
00:06:03,720 --> 00:06:06,680
Speaker 1: they actually produced a short film called Sunspring and they

107
00:06:06,720 --> 00:06:09,839
Speaker 1: even got Thomas middle Ditch, the lead on Hbos Silaken Valley,

108
00:06:10,000 --> 00:06:15,599
Speaker 1: to star in it. Principle is completely constructed. Of the

109
00:06:15,600 --> 00:06:18,320
Speaker 1: same time, it's all about you. To be true, you

110
00:06:18,360 --> 00:06:20,279
Speaker 1: didn't even watch the movie with the rest of the base.

111
00:06:20,360 --> 00:06:22,880
Speaker 1: I don't know, I don't care. I know it's a

112
00:06:22,920 --> 00:06:26,279
Speaker 1: consequence whatever you need to know about the presence of

113
00:06:26,320 --> 00:06:29,120
Speaker 1: the story. I'm a little bit of a boy on

114
00:06:29,160 --> 00:06:32,680
Speaker 1: the floor. So what do you think, Carol? It kind

115
00:06:32,680 --> 00:06:34,440
Speaker 1: of reminds me of when my parents used to take

116
00:06:34,480 --> 00:06:37,520
Speaker 1: me to like a bad production of Macbeth or as

117
00:06:37,560 --> 00:06:41,880
Speaker 1: you like it. Traumatic. You're there and you're seven or eight,

118
00:06:42,200 --> 00:06:44,880
Speaker 1: and you want to understand what's going on, and so

119
00:06:45,160 --> 00:06:47,640
Speaker 1: you kind of pay as close attention as you possibly

120
00:06:47,680 --> 00:06:50,520
Speaker 1: can to what the actors are doing because you have

121
00:06:50,720 --> 00:06:53,840
Speaker 1: no idea what the dialogue means. Yeah, I mean that

122
00:06:53,920 --> 00:06:57,279
Speaker 1: to me is quite impressive because a machine can create

123
00:06:57,360 --> 00:07:00,120
Speaker 1: something which has enough of the elements in common the

124
00:07:00,240 --> 00:07:02,400
Speaker 1: film that we can talk about a real film. You

125
00:07:02,440 --> 00:07:05,080
Speaker 1: can't say it's not a film absolutely. Of course. What's

126
00:07:05,080 --> 00:07:07,680
Speaker 1: different is it didn't take Benjamin very long at all

127
00:07:07,800 --> 00:07:10,800
Speaker 1: to make it. Once you press the button fraction of

128
00:07:10,800 --> 00:07:12,760
Speaker 1: a second there was a couple of seconds perpase, maybe

129
00:07:12,800 --> 00:07:14,480
Speaker 1: maybe a couple of seconds total, actually a fraction of

130
00:07:14,480 --> 00:07:17,880
Speaker 1: a second per page. That's right. After months of agonizing

131
00:07:17,920 --> 00:07:21,720
Speaker 1: over centuries of storytelling theory, the final output only took

132
00:07:21,800 --> 00:07:25,640
Speaker 1: a couple of seconds. So what was Ross's breakthrough? To

133
00:07:25,760 --> 00:07:28,080
Speaker 1: understand we turned to one of the most famous AI

134
00:07:28,120 --> 00:07:32,760
Speaker 1: scientists in the world, Sebastian Throne. Recently, something magical had

135
00:07:32,760 --> 00:07:37,800
Speaker 1: happened recently. The feat has discovered was called machine learning.

136
00:07:38,080 --> 00:07:41,680
Speaker 1: With AI, computers can now find their own rules. They

137
00:07:41,720 --> 00:07:45,440
Speaker 1: are called neural networks. They're comprised of hundreds of millions

138
00:07:45,440 --> 00:07:48,280
Speaker 1: of little vase sample processing units, and those units are

139
00:07:48,360 --> 00:07:51,800
Speaker 1: modeled after what a neurons do in our physical brains.

140
00:07:52,040 --> 00:07:54,640
Speaker 1: You just give them examples, very much like the way

141
00:07:54,680 --> 00:07:58,600
Speaker 1: we we waste children. We don't give our children rules

142
00:07:58,640 --> 00:08:01,560
Speaker 1: for every contingency in life. In the first eight years

143
00:08:01,560 --> 00:08:04,720
Speaker 1: of education. We let them learn, They experience the world,

144
00:08:05,120 --> 00:08:07,880
Speaker 1: and they loan behold. They make their own rules. And

145
00:08:07,880 --> 00:08:09,560
Speaker 1: we are now in the world where computers can do

146
00:08:09,680 --> 00:08:12,360
Speaker 1: the same thing. And this means machine learning can be

147
00:08:12,440 --> 00:08:16,160
Speaker 1: used in all kinds of different fields. Sebastian himself applied

148
00:08:16,160 --> 00:08:19,240
Speaker 1: the technology at Google, where he led the initial development

149
00:08:19,280 --> 00:08:21,840
Speaker 1: of their self driving car. When you want to read

150
00:08:21,880 --> 00:08:24,720
Speaker 1: a book, a book on like what the car should

151
00:08:24,760 --> 00:08:27,640
Speaker 1: do in every situation, that rule book is really complicated

152
00:08:27,680 --> 00:08:29,680
Speaker 1: and it can promise you no matter how many years

153
00:08:29,760 --> 00:08:33,040
Speaker 1: you spent writing it, it's not gonna work. But when

154
00:08:33,080 --> 00:08:36,360
Speaker 1: you give the machine the ability to learn its own rules,

155
00:08:36,640 --> 00:08:40,040
Speaker 1: it is actually able to surpass how people can drive.

156
00:08:41,200 --> 00:08:45,079
Speaker 1: We'll hear more from Sebastian later, but machine learning mL

157
00:08:45,520 --> 00:08:48,040
Speaker 1: is the engine that drives almost all of the excitement

158
00:08:48,080 --> 00:08:52,280
Speaker 1: about AI today, from identifying targets on the battlefield to

159
00:08:52,600 --> 00:08:56,640
Speaker 1: understanding genetic diseases. And it's also what allowed Ross and

160
00:08:56,640 --> 00:08:59,959
Speaker 1: Oscar to create a usable movie script. Rather than laying

161
00:09:00,000 --> 00:09:04,199
Speaker 1: down storytelling rules, they simply showed Benjamin hundreds of examples

162
00:09:04,520 --> 00:09:12,160
Speaker 1: and the algorithm found patterns and learned for itself more sleepwalkers.

163
00:09:12,360 --> 00:09:23,160
Speaker 1: After the break, you're like, oh, did we essentially we

164
00:09:23,240 --> 00:09:27,000
Speaker 1: teach this algorithm anything else about screenplay other than just

165
00:09:27,040 --> 00:09:30,280
Speaker 1: putting in a bunch of screenplays, right, And that's the

166
00:09:30,320 --> 00:09:33,680
Speaker 1: way that machine learning works. What is happening in a

167
00:09:33,760 --> 00:09:36,439
Speaker 1: deep learning algorithm of this kind is it's building an

168
00:09:36,480 --> 00:09:41,199
Speaker 1: extraordinarily complicated mathematical formula by reading all of this stuff

169
00:09:41,240 --> 00:09:43,439
Speaker 1: over and over again, like the auto complete on your phone.

170
00:09:43,440 --> 00:09:46,160
Speaker 1: The neural that is actually sampling from a probability distribution

171
00:09:46,200 --> 00:09:49,640
Speaker 1: of which letters, bass or punctuation become next. So the

172
00:09:49,679 --> 00:09:53,240
Speaker 1: script for sun Spring was essentially the most mathematically probable

173
00:09:53,280 --> 00:09:56,800
Speaker 1: Sci Fi script except Ross and Oscar did have one

174
00:09:56,880 --> 00:10:01,040
Speaker 1: important lever of creative control. The other of parameters that

175
00:10:01,080 --> 00:10:04,480
Speaker 1: you're probably wondering about, there's one called the temperature is

176
00:10:04,480 --> 00:10:08,240
Speaker 1: the riskiness of those next letter predictions. What Ross is

177
00:10:08,280 --> 00:10:12,200
Speaker 1: describing is almost like a dial for creativity. Turn it

178
00:10:12,240 --> 00:10:14,680
Speaker 1: up to a really high temperature, and the neural net

179
00:10:14,720 --> 00:10:17,880
Speaker 1: is going to be extra creative and start making up words,

180
00:10:17,960 --> 00:10:22,600
Speaker 1: babbling at a very high temperature. It's essentially drunk. Low temperature,

181
00:10:22,600 --> 00:10:25,200
Speaker 1: it's going to be very repetitive and possible even begin

182
00:10:25,240 --> 00:10:28,760
Speaker 1: to plagiarize its source material. So it'll be very repetitive.

183
00:10:28,760 --> 00:10:30,480
Speaker 1: It'll be like the streets and the streets, and the

184
00:10:30,520 --> 00:10:34,400
Speaker 1: streets and the streets. It's essentially went working for network television. Yeah, exactly.

185
00:10:34,480 --> 00:10:36,280
Speaker 1: So we wanted it to be sort of in the middle.

186
00:10:37,000 --> 00:10:39,280
Speaker 1: In the middle is where we found the best output

187
00:10:39,440 --> 00:10:44,280
Speaker 1: and the most I think usable output, and Sunspring was born.

188
00:10:45,960 --> 00:10:49,320
Speaker 1: So Benjamin Ross and Oscar right together now they write

189
00:10:49,320 --> 00:10:53,080
Speaker 1: poetry and movies and sometimes what Benjamin spits out is good.

190
00:10:53,600 --> 00:10:55,360
Speaker 1: Often they have to sift through it to find the

191
00:10:55,400 --> 00:10:59,440
Speaker 1: best stuff. But he's prolific and he never ever suffers

192
00:10:59,440 --> 00:11:04,520
Speaker 1: from writers. Look, so Kara was telling us earlier about

193
00:11:04,559 --> 00:11:07,320
Speaker 1: Alex the kidd and using AI to make music, and

194
00:11:07,320 --> 00:11:09,199
Speaker 1: that's something I want to understand a bit more about.

195
00:11:09,559 --> 00:11:13,640
Speaker 1: So Julian went on a little bit of an expedition. Yes,

196
00:11:13,720 --> 00:11:15,920
Speaker 1: I did. I've been seeing a lot of articles lately

197
00:11:15,960 --> 00:11:18,040
Speaker 1: about AI and the arts, and I've been pretty curious

198
00:11:18,040 --> 00:11:20,760
Speaker 1: about music specifically. We might take it for granted, but

199
00:11:20,920 --> 00:11:24,040
Speaker 1: music is this primal emotional thing that's been with us forever.

200
00:11:24,080 --> 00:11:27,240
Speaker 1: It might even predate language. But now Warner Music Group

201
00:11:27,320 --> 00:11:30,360
Speaker 1: made history in April two nineteen, is the first major

202
00:11:30,440 --> 00:11:33,800
Speaker 1: label to sign an AI to a record deal. Yeah,

203
00:11:33,840 --> 00:11:37,720
Speaker 1: they signed this bot called Endel, which makes ambient noises

204
00:11:37,760 --> 00:11:39,840
Speaker 1: based on where you are and what the weather is

205
00:11:39,880 --> 00:11:41,760
Speaker 1: and what time of day it is. When I think

206
00:11:41,800 --> 00:11:44,240
Speaker 1: of this kind of music, I think of those Spotify

207
00:11:44,320 --> 00:11:48,880
Speaker 1: playlists like Peaceful Piano and Blissed Out Dinner Party, which

208
00:11:48,920 --> 00:11:52,440
Speaker 1: would become extremely popular. It's not the same thing as Beyonce, No,

209
00:11:52,640 --> 00:11:56,280
Speaker 1: definitely not. But Warner Music Group signed Endl to generate

210
00:11:56,320 --> 00:11:59,600
Speaker 1: twenty albums of ambient music. And now that we live

211
00:11:59,600 --> 00:12:03,120
Speaker 1: in a world where aies can get record deals, what

212
00:12:03,240 --> 00:12:05,559
Speaker 1: does this mean for artists? What does this mean for

213
00:12:05,600 --> 00:12:08,920
Speaker 1: even just music as we know it? Well, in my

214
00:12:09,000 --> 00:12:12,000
Speaker 1: quest to find out, I visited this company called Amper.

215
00:12:12,640 --> 00:12:15,840
Speaker 1: My name is Drew Silverstein. I am the co founder

216
00:12:15,920 --> 00:12:18,720
Speaker 1: and CEO of Amper Music. Amper is an AI music

217
00:12:18,760 --> 00:12:22,120
Speaker 1: company that Drew says will enable anyone to create music.

218
00:12:22,360 --> 00:12:24,320
Speaker 1: In fact, the only things you need to know are

219
00:12:24,600 --> 00:12:27,160
Speaker 1: the genre of music you want to create, the mood

220
00:12:27,440 --> 00:12:29,400
Speaker 1: you'd like to convey, and the length of your piece

221
00:12:29,400 --> 00:12:31,760
Speaker 1: of music. That's all you know. You can create a

222
00:12:31,760 --> 00:12:34,160
Speaker 1: brand new, unique piece of music in a matter of seconds.

223
00:12:34,280 --> 00:12:39,679
Speaker 1: So the big question is should musicians worry about computers

224
00:12:39,720 --> 00:12:43,400
Speaker 1: taking their job? Well, let's try it and see. So

225
00:12:43,760 --> 00:12:48,200
Speaker 1: what do you want to do? Cinematic, documentary, folk cinematic cinematic,

226
00:12:48,640 --> 00:12:53,960
Speaker 1: minimal percussion or quirky percussion. It's rendering a song right now.

227
00:12:54,000 --> 00:13:00,520
Speaker 1: And here we go. We've got something. I'm were deep

228
00:13:00,559 --> 00:13:04,439
Speaker 1: in the forest of Nicaragua. There's a breed of jaguar.

229
00:13:05,360 --> 00:13:07,960
Speaker 1: You might have heard of it. It's called the take

230
00:13:08,000 --> 00:13:12,480
Speaker 1: a Killer panther. Look, here's the thing. I don't know

231
00:13:12,760 --> 00:13:17,439
Speaker 1: what the difference between that and music is. I really don't. Yeah,

232
00:13:17,440 --> 00:13:20,760
Speaker 1: so you're woud Yeah, all right, So there's that. And

233
00:13:21,000 --> 00:13:24,520
Speaker 1: this isn't the only AI music app out there. Another

234
00:13:24,600 --> 00:13:28,440
Speaker 1: major player is called Magenta, and big surprise, they're at Google.

235
00:13:28,720 --> 00:13:31,160
Speaker 1: Magenta are using AI to create a ton of new tools.

236
00:13:31,240 --> 00:13:34,360
Speaker 1: From a piano genie that makes it impossible to play

237
00:13:34,400 --> 00:13:37,439
Speaker 1: bad notes to something that can generate drum loops, or

238
00:13:37,520 --> 00:13:40,560
Speaker 1: something that can even play piano duets with you. You

239
00:13:40,559 --> 00:13:44,560
Speaker 1: can even translate raw audio to a piano score. Raw audio,

240
00:13:44,640 --> 00:13:47,480
Speaker 1: like if I play just something raw on the piano,

241
00:13:47,880 --> 00:13:53,760
Speaker 1: raw audio like oh, literal raw audio, literal raw audio.

242
00:13:54,520 --> 00:13:57,040
Speaker 1: And Magenta has also trained a neural network just like

243
00:13:57,360 --> 00:14:00,240
Speaker 1: Ross and Oscar, only instead of sci fi scripts, they

244
00:14:00,320 --> 00:14:04,559
Speaker 1: trained on over four hundred performances by skilled pianists. They

245
00:14:04,600 --> 00:14:07,080
Speaker 1: fed it into the neural network and let me play

246
00:14:07,120 --> 00:14:09,480
Speaker 1: one of the piano experts. First. This is a real

247
00:14:09,640 --> 00:14:17,440
Speaker 1: piano player, all right, so nice? Right? Yeah, okay, ready

248
00:14:17,440 --> 00:14:28,680
Speaker 1: for the AI. What that's a computer that was all

249
00:14:28,680 --> 00:14:30,920
Speaker 1: a computer. I didn't ever play a human one. That's

250
00:14:30,920 --> 00:14:33,800
Speaker 1: a computer that was trained by a human playing piano.

251
00:14:34,480 --> 00:14:37,080
Speaker 1: And then how do you make a computer come up

252
00:14:37,120 --> 00:14:39,560
Speaker 1: with that? Right? So even though it's not a screenplay,

253
00:14:39,600 --> 00:14:41,520
Speaker 1: it's still data that you can feed a neural network

254
00:14:41,560 --> 00:14:44,440
Speaker 1: with to find patterns. And in this case, Magenta used

255
00:14:44,440 --> 00:14:47,680
Speaker 1: a data set from the Yamahai Piano competition. So human

256
00:14:47,720 --> 00:14:50,480
Speaker 1: pianists played on these digital keyboards which recorded the nuances

257
00:14:50,480 --> 00:14:52,640
Speaker 1: of their performance, like how long they hit notes, and

258
00:14:52,680 --> 00:14:54,960
Speaker 1: it recorded all that information into a digital score that

259
00:14:55,000 --> 00:14:57,600
Speaker 1: a computer could interpret. And we've actually had that technology

260
00:14:57,640 --> 00:14:59,840
Speaker 1: for a while now, it's called MIDI. But training and

261
00:15:00,000 --> 00:15:02,080
Speaker 1: on network on the data is new. See. The thing

262
00:15:02,120 --> 00:15:05,520
Speaker 1: that I come back to is that a computer doesn't

263
00:15:05,600 --> 00:15:09,040
Speaker 1: know it's playing music, so much of watching a musical

264
00:15:09,080 --> 00:15:13,480
Speaker 1: performance is knowing that this is coming from someone who

265
00:15:13,560 --> 00:15:18,280
Speaker 1: is emoting. Right, Yeah, there's actually there's an emotional communication happening, right,

266
00:15:18,360 --> 00:15:21,960
Speaker 1: that's right. I do think though the future is not

267
00:15:22,200 --> 00:15:26,240
Speaker 1: rejecting this. It's better to imagine what would Stravinsky have

268
00:15:26,440 --> 00:15:29,040
Speaker 1: done with this kind of technology, because Stravinsky is still

269
00:15:29,080 --> 00:15:42,600
Speaker 1: a musical genius. Right, Yeah, Definitely it's cool to listen

270
00:15:42,640 --> 00:15:45,840
Speaker 1: to those musical examples of machine learning because you can

271
00:15:45,880 --> 00:15:50,720
Speaker 1: really hear how the algorithm is reinterpreting existing material. Of course,

272
00:15:51,200 --> 00:15:54,040
Speaker 1: listening to the output is one thing. Tasting it is

273
00:15:54,120 --> 00:15:58,440
Speaker 1: quite another. The problem was that somebody had told me

274
00:15:58,480 --> 00:16:00,520
Speaker 1: that they had made the recipe for Stan, that it

275
00:16:00,640 --> 00:16:03,360
Speaker 1: was good, and what it was as a recipe called

276
00:16:03,440 --> 00:16:08,160
Speaker 1: chocolate baked and serves. That's Janelle Shane. She's a research

277
00:16:08,200 --> 00:16:11,240
Speaker 1: scientist and the author of a blog called AI Weirdness.

278
00:16:11,800 --> 00:16:14,080
Speaker 1: She's talking about a recipe written by Ai that she

279
00:16:14,120 --> 00:16:19,680
Speaker 1: actually cooked and eight. It starts out as a perfectly ordinary,

280
00:16:19,760 --> 00:16:24,000
Speaker 1: flowerless chocolate brownie all the way until the very last ingredient,

281
00:16:24,400 --> 00:16:26,760
Speaker 1: which is a cup of horse badish. I knew I

282
00:16:26,800 --> 00:16:28,480
Speaker 1: was in trouble when I opened the oven door and

283
00:16:28,520 --> 00:16:33,760
Speaker 1: my eyes just started watering. It was yeah, it was terrible.

284
00:16:34,080 --> 00:16:36,960
Speaker 1: On her blog, Jenelle experiments with putting AI to a

285
00:16:37,080 --> 00:16:40,800
Speaker 1: range of tasks, from writing new pickup lines to naming

286
00:16:40,840 --> 00:16:44,720
Speaker 1: Halloween costumes, and often her experiments with machine learning are

287
00:16:44,720 --> 00:16:49,640
Speaker 1: pretty revealing about us. It plays into this thought experiment,

288
00:16:49,760 --> 00:16:52,560
Speaker 1: what would an alien think of our world? It takes

289
00:16:52,560 --> 00:16:57,400
Speaker 1: something that's very ordinary and mixes it up into this

290
00:16:57,520 --> 00:17:01,560
Speaker 1: thing that sounds like the original, but the meaning has

291
00:17:01,600 --> 00:17:05,240
Speaker 1: been completely changed. Chopped whipping cream may be an ingredient

292
00:17:05,320 --> 00:17:08,840
Speaker 1: and fold water, enrolled it into cubes, or spread the

293
00:17:08,840 --> 00:17:12,080
Speaker 1: butter in the refrigerator. That's another direction that came up with.

294
00:17:12,640 --> 00:17:15,560
Speaker 1: Remember Ross and Oscar playing with the creativity setting for

295
00:17:15,600 --> 00:17:19,439
Speaker 1: their scripts. Janelle plays with herbot's temperature too, so I

296
00:17:19,480 --> 00:17:22,160
Speaker 1: can turn it up and the neural net may choose

297
00:17:22,160 --> 00:17:24,880
Speaker 1: its second best or third best guess as to what

298
00:17:25,080 --> 00:17:28,200
Speaker 1: letter comes next. And if I turn the creativity all

299
00:17:28,200 --> 00:17:31,440
Speaker 1: the way down, then everything maybe something like the the

300
00:17:32,560 --> 00:17:36,200
Speaker 1: the or recipes may be just you know, one teaspoon

301
00:17:36,240 --> 00:17:38,960
Speaker 1: of vanilla over and over and over again, because that's

302
00:17:39,040 --> 00:17:44,640
Speaker 1: just a very likely ingredient. It's really interesting with the

303
00:17:44,680 --> 00:17:48,879
Speaker 1: recipes to turn down the creativity and see what it

304
00:17:48,920 --> 00:17:52,320
Speaker 1: comes up with as the most quintessential recipes. At the

305
00:17:52,359 --> 00:17:55,320
Speaker 1: lowest setting, you may not get hole Strandish brownies, but

306
00:17:55,400 --> 00:17:57,320
Speaker 1: you do get a clear picture of what we eat

307
00:17:57,640 --> 00:18:00,399
Speaker 1: and who we are. I look at what kinds of

308
00:18:00,760 --> 00:18:03,200
Speaker 1: recipe titles that comes up with. There are things like

309
00:18:03,760 --> 00:18:07,919
Speaker 1: chocolate chicken chicken cake, and another one that's chocolate chocolate

310
00:18:08,000 --> 00:18:10,520
Speaker 1: chocolate chocolate cake. And there was a lot of cheese

311
00:18:10,560 --> 00:18:14,080
Speaker 1: in these recipes too, so it's kind of revealing about

312
00:18:14,119 --> 00:18:18,880
Speaker 1: what sorts of things we cook with. Then we like chocolate,

313
00:18:18,920 --> 00:18:21,960
Speaker 1: cheese and chicken apparently. But then I did the same

314
00:18:22,000 --> 00:18:26,920
Speaker 1: experiment with recipes from Bone Appetite, and then the most

315
00:18:26,920 --> 00:18:31,280
Speaker 1: common ingredients that kept using were cilantro and pomegranate juice.

316
00:18:32,160 --> 00:18:35,000
Speaker 1: So these algorithms essentially hold up a mirror to the

317
00:18:35,080 --> 00:18:37,840
Speaker 1: data sets that we give them. They do, yeah, they

318
00:18:37,880 --> 00:18:41,800
Speaker 1: reflect the data sets back to us in really weird ways,

319
00:18:42,160 --> 00:18:45,280
Speaker 1: and they can absolutely pick up whatever bias there is

320
00:18:45,320 --> 00:18:48,240
Speaker 1: in a input data set. And I think what we're

321
00:18:48,280 --> 00:18:52,359
Speaker 1: discovering is just how prevalent that bias is and how

322
00:18:52,400 --> 00:18:55,840
Speaker 1: easy it is for neural networks to latch onto that

323
00:18:55,880 --> 00:18:59,720
Speaker 1: bias and copy it as a handy tool toward copying

324
00:18:59,720 --> 00:19:02,800
Speaker 1: whatever where the humans are doing. They say that the

325
00:19:02,800 --> 00:19:05,679
Speaker 1: way to a person's heart is through their stomach. But

326
00:19:05,760 --> 00:19:09,160
Speaker 1: Janelle didn't stop at chocolate, bakes and surfs. She's also

327
00:19:09,200 --> 00:19:12,680
Speaker 1: turned AI onto some more direct roots. I really liked

328
00:19:12,720 --> 00:19:15,000
Speaker 1: the pickup lines. And there are all these puns and

329
00:19:15,040 --> 00:19:17,879
Speaker 1: all this wordplay that it didn't have any way to

330
00:19:17,880 --> 00:19:20,159
Speaker 1: grab hold of and figure out how to use. But

331
00:19:20,600 --> 00:19:25,880
Speaker 1: I think what it produced this sort of charming surrealism

332
00:19:26,000 --> 00:19:30,480
Speaker 1: and kind of garble nonsensical. I think it's an improvement

333
00:19:30,640 --> 00:19:34,080
Speaker 1: on every single one of the originals. My very favorite

334
00:19:34,080 --> 00:19:37,240
Speaker 1: one is you look like a thing and I love you.

335
00:19:37,240 --> 00:19:39,400
Speaker 1: You are so beautiful that you make me feel better

336
00:19:39,480 --> 00:19:42,679
Speaker 1: to see you. Or you must be a tringle because

337
00:19:42,680 --> 00:19:46,800
Speaker 1: you're the only thing here. Are you a camera? Because

338
00:19:46,840 --> 00:19:50,480
Speaker 1: I want to see the most beautiful than you. Yeah,

339
00:19:49,640 --> 00:19:54,400
Speaker 1: I'll definitely lie with you. No one's have a used

340
00:19:54,480 --> 00:19:56,960
Speaker 1: real pickup line on me? Use one on you know?

341
00:19:57,640 --> 00:20:01,280
Speaker 1: Do I look like someone who would receive a pickup Well,

342
00:20:01,320 --> 00:20:04,880
Speaker 1: here's one of them. I don't know you. That's good

343
00:20:05,119 --> 00:20:07,760
Speaker 1: a lot of girls are into that. Are you a candle?

344
00:20:07,880 --> 00:20:12,639
Speaker 1: Because you're so hard of the looks with you. So

345
00:20:12,640 --> 00:20:15,680
Speaker 1: in effect, what the algorithm is doing is highlighting patterns

346
00:20:15,680 --> 00:20:18,360
Speaker 1: in the data. I mean, there's sound structural like pickup

347
00:20:18,400 --> 00:20:20,960
Speaker 1: lines about the words themselves don't make any sense. The

348
00:20:21,000 --> 00:20:24,440
Speaker 1: machines are reflecting their creators and spitting back something which

349
00:20:24,880 --> 00:20:27,560
Speaker 1: resembles pick up lines and makes us think a little

350
00:20:27,560 --> 00:20:31,320
Speaker 1: bit more carefully about what a pickup line is. And

351
00:20:31,359 --> 00:20:34,280
Speaker 1: while training Benjamin Ross and Oscar found the same thing

352
00:20:34,840 --> 00:20:38,680
Speaker 1: as the algorithm learned patterns revealed bias present in our cinema.

353
00:20:40,600 --> 00:20:43,199
Speaker 1: When you train an algorithm like Benjamin on millions and

354
00:20:43,240 --> 00:20:45,879
Speaker 1: millions in this case of synopsis from the Internet of

355
00:20:45,920 --> 00:20:49,119
Speaker 1: the movies, the synopsis that come out have certain patterns

356
00:20:49,160 --> 00:20:51,520
Speaker 1: in them. For example, they mentioned men full times more

357
00:20:51,520 --> 00:20:54,680
Speaker 1: often than they mentioned women. But you you learn other

358
00:20:54,760 --> 00:20:57,120
Speaker 1: things about it than that. You learned that the most

359
00:20:57,119 --> 00:20:59,080
Speaker 1: common phrase in the in the output is a young

360
00:20:59,119 --> 00:21:02,040
Speaker 1: man in a small town. So what does a filmmaker

361
00:21:02,080 --> 00:21:04,600
Speaker 1: like Oscar learn from this? I used to call this

362
00:21:04,640 --> 00:21:06,960
Speaker 1: project the average movie projects. And the reason I called

363
00:21:06,960 --> 00:21:09,359
Speaker 1: it that is the theory was for me, if you

364
00:21:09,359 --> 00:21:11,520
Speaker 1: could make the right kind of algorithm that the movie

365
00:21:11,560 --> 00:21:14,200
Speaker 1: that you would make that would be the theoretically perfect

366
00:21:14,280 --> 00:21:17,240
Speaker 1: movie would also be the most boring movie ever made,

367
00:21:17,480 --> 00:21:19,120
Speaker 1: and that it would. It would it would be by

368
00:21:19,119 --> 00:21:21,399
Speaker 1: definition all of the things that were the most clich

369
00:21:21,640 --> 00:21:23,399
Speaker 1: because that's what cliche means, is the thing that that

370
00:21:23,480 --> 00:21:26,400
Speaker 1: you can rely on to work. And why do that?

371
00:21:26,560 --> 00:21:28,679
Speaker 1: Because the thing I'm most interested in is doing the

372
00:21:28,720 --> 00:21:30,360
Speaker 1: thing that we haven't done yet. I want to move

373
00:21:30,400 --> 00:21:33,440
Speaker 1: the form forward. Seeing all of these biases and assumptions

374
00:21:33,480 --> 00:21:36,639
Speaker 1: that are baked into our movies and our snacks doesn't

375
00:21:36,680 --> 00:21:39,560
Speaker 1: mean we're doomed to repeat them. In fact, the awareness

376
00:21:39,560 --> 00:21:42,800
Speaker 1: can be liberating. That's what's helped me, I think, is

377
00:21:42,840 --> 00:21:46,600
Speaker 1: seeing Benjamin's capacity to show me more directly what it

378
00:21:46,680 --> 00:21:49,160
Speaker 1: is that are our patents, are our habits, and then

379
00:21:49,359 --> 00:21:51,600
Speaker 1: I can ask more easily how to move forward from that.

380
00:21:53,720 --> 00:22:05,000
Speaker 1: We'll get there after the break. So we've heard about

381
00:22:05,080 --> 00:22:08,240
Speaker 1: Janelle Shane using AI to reveal bias, and Ross and

382
00:22:08,280 --> 00:22:10,879
Speaker 1: Oscar using it to help them think more creatively about

383
00:22:10,920 --> 00:22:13,919
Speaker 1: filmmaking as well as how it can be applied to music.

384
00:22:14,520 --> 00:22:17,200
Speaker 1: And that's the great promise of AI. We may worry

385
00:22:17,200 --> 00:22:20,159
Speaker 1: about replacing jobs, but it can augment our lives in

386
00:22:20,200 --> 00:22:23,520
Speaker 1: so many ways. At least that's how Sebastian Throne sees it.

387
00:22:24,480 --> 00:22:27,359
Speaker 1: I would say the term AI is a bit deceptive

388
00:22:27,359 --> 00:22:29,760
Speaker 1: because it sets up computers to be on equal power

389
00:22:29,760 --> 00:22:33,000
Speaker 1: with people. I see it to be stronger where we

390
00:22:33,040 --> 00:22:36,480
Speaker 1: are weak, and weaker where you're strong. It's not a

391
00:22:36,680 --> 00:22:40,680
Speaker 1: technology that will replace us, as it's not really empowers

392
00:22:40,680 --> 00:22:44,080
Speaker 1: but what might that empowerment look like beyond bias detection

393
00:22:44,240 --> 00:22:48,800
Speaker 1: and piano playing well. In seventeen, Sebastian published a paper

394
00:22:48,840 --> 00:22:52,840
Speaker 1: in Nature on using AI to diagnose skin cancer using

395
00:22:52,840 --> 00:22:56,600
Speaker 1: just an iPhone. So in medicine, you can think of

396
00:22:56,640 --> 00:23:00,960
Speaker 1: your iPhone that can find skin cancer as turning regular

397
00:23:01,400 --> 00:23:04,879
Speaker 1: physicians or anybody in the world into an expert on

398
00:23:04,960 --> 00:23:07,320
Speaker 1: day one, because now they have the superpower to be

399
00:23:07,359 --> 00:23:10,359
Speaker 1: able to distinguish something that previously would have taken tens

400
00:23:10,359 --> 00:23:13,199
Speaker 1: of years to learn. The same is true for the

401
00:23:13,240 --> 00:23:16,239
Speaker 1: self driving car. Now children can drive and and and

402
00:23:16,320 --> 00:23:18,240
Speaker 1: blind people can drive, head blind people drive around and

403
00:23:18,240 --> 00:23:21,800
Speaker 1: self driving cars. So for me, the real opportunities to

404
00:23:22,160 --> 00:23:24,800
Speaker 1: use the eye to extract the knowledge from some human

405
00:23:24,840 --> 00:23:27,760
Speaker 1: experts that are well trained and transpose this knowledge to

406
00:23:27,840 --> 00:23:30,480
Speaker 1: other brains that people not so well trained. This is

407
00:23:30,480 --> 00:23:33,560
Speaker 1: what I personally find so fascinating about AI and a

408
00:23:33,600 --> 00:23:37,400
Speaker 1: big reason I wanted to do Sleepwalkers. The same technology

409
00:23:37,560 --> 00:23:42,719
Speaker 1: underlies rossan Oscar's films, Janelle's recipes and self driving cars

410
00:23:42,720 --> 00:23:46,959
Speaker 1: and cancer diagnostics and so much more. The training for

411
00:23:47,359 --> 00:23:50,800
Speaker 1: skin cancer detection or cancer detection radeology and the training

412
00:23:50,800 --> 00:23:55,040
Speaker 1: for the self dime car amazingly similar. In both cases,

413
00:23:55,080 --> 00:23:57,840
Speaker 1: what you do is you compile the data set typically

414
00:23:57,920 --> 00:24:01,120
Speaker 1: hundreds of thousands up to hundreds of millions of images.

415
00:24:01,600 --> 00:24:04,760
Speaker 1: In skin cancer, we use biopsies. We had a database

416
00:24:04,800 --> 00:24:07,399
Speaker 1: of a hundred twenty nine thousand images that a lab

417
00:24:07,480 --> 00:24:11,159
Speaker 1: had biopsied and provided. In the self driving car, you

418
00:24:11,560 --> 00:24:13,880
Speaker 1: could be as easy as having a human driver provide

419
00:24:13,920 --> 00:24:16,480
Speaker 1: inputs with their student veel and the and the break

420
00:24:16,800 --> 00:24:19,119
Speaker 1: as to what the right thing is to do, and

421
00:24:19,200 --> 00:24:22,600
Speaker 1: then the network mimics human behavior, it mimics the diagnostics

422
00:24:22,680 --> 00:24:25,560
Speaker 1: of a physician, or it mimics the style of a driver.

423
00:24:25,800 --> 00:24:29,800
Speaker 1: The underlying albums are amazingly similar. What makes this moment

424
00:24:29,800 --> 00:24:32,119
Speaker 1: all the more interesting is that AI is in the

425
00:24:32,200 --> 00:24:37,160
Speaker 1: process of being consumerized like Sebastion said, your iPhone can

426
00:24:37,200 --> 00:24:41,199
Speaker 1: diagnose skin cancer. Self driving cars are already on the roads,

427
00:24:41,240 --> 00:24:44,639
Speaker 1: and as these tools become more and more accessible, society

428
00:24:44,800 --> 00:24:48,840
Speaker 1: will start to change. These technologies become closer and closer

429
00:24:48,880 --> 00:24:51,320
Speaker 1: to us. The fact that you carry yourself phone is

430
00:24:51,320 --> 00:24:53,960
Speaker 1: a big deal. You might not see this way, but

431
00:24:54,119 --> 00:24:57,679
Speaker 1: what it does it puts the computer seamlessly into your life.

432
00:24:58,040 --> 00:25:01,280
Speaker 1: You're texting app, your SMS is so close to you

433
00:25:01,520 --> 00:25:04,200
Speaker 1: that you can now talk to people thousands of miles

434
00:25:04,200 --> 00:25:06,600
Speaker 1: of a on a button press or on a microphone.

435
00:25:06,960 --> 00:25:11,400
Speaker 1: That makes you effectively super human without the actual physical implant.

436
00:25:11,680 --> 00:25:14,400
Speaker 1: But when it comes to AI, for now, leading edge

437
00:25:14,400 --> 00:25:17,200
Speaker 1: algorithms are off limits to those of us who can't

438
00:25:17,240 --> 00:25:20,040
Speaker 1: code or who don't have the means to learn. Oscar

439
00:25:20,080 --> 00:25:24,000
Speaker 1: wouldn't have been able to make Sunspring without Rosses technical expertise,

440
00:25:24,680 --> 00:25:28,119
Speaker 1: but that's all starting to change, Julian, you spoke with

441
00:25:28,200 --> 00:25:31,480
Speaker 1: somebody making AI more accessible. Yeah, while we were looking

442
00:25:31,480 --> 00:25:34,040
Speaker 1: into AI and music, we came across Runway m L

443
00:25:34,359 --> 00:25:37,680
Speaker 1: their lab based in Bushwick, and they feel strongly about

444
00:25:37,720 --> 00:25:41,399
Speaker 1: letting more people into work with AI Creatively. I spoke

445
00:25:41,400 --> 00:25:43,840
Speaker 1: with Christo bal Valence Weela, the co founder of Runway,

446
00:25:43,880 --> 00:25:46,560
Speaker 1: which is basically like the Adobe Creative Suite for AI.

447
00:25:46,760 --> 00:25:49,320
Speaker 1: So think of a program that looks like Photoshop. They're

448
00:25:49,320 --> 00:25:51,920
Speaker 1: adding tons of different AI models to the Runway app,

449
00:25:51,960 --> 00:25:54,199
Speaker 1: where instead of having to know how to code, you

450
00:25:54,200 --> 00:25:57,199
Speaker 1: can just manipulate some sliders and dials and still have

451
00:25:57,240 --> 00:25:59,800
Speaker 1: a I generate something. If we really think of this

452
00:26:00,080 --> 00:26:02,439
Speaker 1: the game changing technology that will impact us for like

453
00:26:02,560 --> 00:26:05,160
Speaker 1: years to come, we need to have more people from

454
00:26:05,160 --> 00:26:08,960
Speaker 1: different backgrounds and disciplines jumping to that discussion and proposing

455
00:26:09,000 --> 00:26:11,640
Speaker 1: ways of looking at algorithms that those researchers and those

456
00:26:11,640 --> 00:26:13,960
Speaker 1: scientists are not thinking of. This is gonna impact us,

457
00:26:14,040 --> 00:26:15,679
Speaker 1: and it's going to change the way we see not

458
00:26:15,840 --> 00:26:18,879
Speaker 1: just the world, but ourselves. Not just ourselves, but how

459
00:26:18,960 --> 00:26:21,440
Speaker 1: we think about our creativity. And the list of things

460
00:26:21,520 --> 00:26:24,440
Speaker 1: that Runway can help you do is frankly crazy. Will

461
00:26:24,480 --> 00:26:28,280
Speaker 1: post some on our Instagram at Sleepwalkers podcast, but just

462
00:26:28,359 --> 00:26:31,480
Speaker 1: one example, you can take video of anyone adds and

463
00:26:31,520 --> 00:26:34,199
Speaker 1: have their body copy the poses that you make in

464
00:26:34,240 --> 00:26:37,639
Speaker 1: your webcam. So Krystoval actually tweeted one of him controlling

465
00:26:37,680 --> 00:26:40,879
Speaker 1: Stephen Colbert's body on The Late Show just with his

466
00:26:40,920 --> 00:26:44,400
Speaker 1: webcam moving his arms. Stephen moves his arms. It's nuts, right,

467
00:26:44,560 --> 00:26:47,120
Speaker 1: And imagine that kind of technology in the hands of artists.

468
00:26:47,440 --> 00:26:50,440
Speaker 1: Start thinking about them as not like something that's gonna

469
00:26:50,480 --> 00:26:53,760
Speaker 1: destroy our creativity or gonna replace writers an artists or whatever.

470
00:26:53,880 --> 00:26:55,440
Speaker 1: This is gonna be a typewriter, This is gonna be

471
00:26:55,480 --> 00:26:58,239
Speaker 1: a paintbrush, and people will start building and using it

472
00:26:58,359 --> 00:27:02,600
Speaker 1: to understand their own creativity in a new way. Of course,

473
00:27:02,680 --> 00:27:05,440
Speaker 1: as always, it's up to us to make sure we

474
00:27:05,560 --> 00:27:09,879
Speaker 1: use these new tools for good. If I build a shovel, okay,

475
00:27:09,920 --> 00:27:12,280
Speaker 1: and you decide to go to the beach and digging sand,

476
00:27:12,880 --> 00:27:15,200
Speaker 1: you're biased. You're digging in sand and guess what You're

477
00:27:15,200 --> 00:27:17,440
Speaker 1: shoveled only to an up sand. The same as true

478
00:27:17,480 --> 00:27:19,280
Speaker 1: for AI. If you give it a certain type of

479
00:27:19,359 --> 00:27:21,800
Speaker 1: data set, I can promise you whatever I get out

480
00:27:21,960 --> 00:27:24,600
Speaker 1: reflects the data you're put in. It's up to us,

481
00:27:24,720 --> 00:27:27,960
Speaker 1: the people, to make responsible decisions. And as we want

482
00:27:28,000 --> 00:27:31,800
Speaker 1: to create equal opportunity and evadicate certain biases society that

483
00:27:31,920 --> 00:27:34,480
Speaker 1: exists today, is up to us to do it, and

484
00:27:34,520 --> 00:27:37,320
Speaker 1: I promise you, if you work hard on this, technologies

485
00:27:37,560 --> 00:27:41,280
Speaker 1: will reflect that. But even Sebastian, one of Silicon Valleys

486
00:27:41,320 --> 00:27:47,119
Speaker 1: Great Optimists, recognizes the risks all technologies can harm people.

487
00:27:47,520 --> 00:27:50,480
Speaker 1: In fact, technologies can be abused to harm people, Like

488
00:27:50,920 --> 00:27:53,360
Speaker 1: my kitchen knife, which serves me a great purpose every

489
00:27:53,359 --> 00:27:56,280
Speaker 1: time I've guests over in shopping. My produce can also

490
00:27:56,359 --> 00:28:01,240
Speaker 1: be abused to harm people. In the next episode of Sleepwalkers,

491
00:28:01,320 --> 00:28:04,760
Speaker 1: we dive deep into the ability of algorithms to cause harm.

492
00:28:04,800 --> 00:28:07,800
Speaker 1: We traveled from China and the social credit system to

493
00:28:08,119 --> 00:28:10,800
Speaker 1: a parole board in New York and we speak with

494
00:28:10,840 --> 00:28:15,120
Speaker 1: people building technology they believe will make us safer. I'm

495
00:28:15,160 --> 00:28:31,440
Speaker 1: as Vlachen, see you next time. Sleepwalkers is a production

496
00:28:31,480 --> 00:28:34,920
Speaker 1: of our heart Radio and unusual productions. There's so much

497
00:28:34,960 --> 00:28:37,080
Speaker 1: we don't have time for in our episodes, but that

498
00:28:37,119 --> 00:28:39,680
Speaker 1: we'd love to share with you. So for the latest

499
00:28:39,720 --> 00:28:43,200
Speaker 1: AI news, live interviews, and behind the scenes footage, find

500
00:28:43,280 --> 00:28:47,440
Speaker 1: us on Instagram, at Sleepwalker's podcast or at sleepwalkers podcast

501
00:28:47,480 --> 00:28:50,360
Speaker 1: dot com. Special thanks on this episode to paw Suris,

502
00:28:50,440 --> 00:28:53,960
Speaker 1: who introduced us to Oscar Ross and Benjamin and to

503
00:28:54,200 --> 00:28:58,160
Speaker 1: artificial intelligence, which composed over half of music in this episode.

504
00:28:58,520 --> 00:29:02,479
Speaker 1: Could you tell which was which Sleepwalkers is hosted by

505
00:29:02,520 --> 00:29:05,640
Speaker 1: me Ozveloshin and co hosted by me Kara Price. Were

506
00:29:05,640 --> 00:29:08,600
Speaker 1: produced by Julian Weller with help from Jacopo Penzo and

507
00:29:08,680 --> 00:29:12,280
Speaker 1: Taylor Chacog. Mixing by Tristan McNeil and Julian Weller. Our

508
00:29:12,320 --> 00:29:15,800
Speaker 1: story editor is Matthew Riddle. Recording assistance. This episode from

509
00:29:15,920 --> 00:29:20,160
Speaker 1: Joanne de Luna. Sleepwalkers is executive produced by me Ozveloshin

510
00:29:20,280 --> 00:29:23,600
Speaker 1: and Mangesh hatt Together. For more podcasts from my heart Radio,

511
00:29:23,680 --> 00:29:26,600
Speaker 1: visit the i heart Radio app, Apple Podcasts, or wherever

512
00:29:26,640 --> 00:29:27,960
Speaker 1: you listen to your favorite shows,