WEBVTT

0
00:00:00.000 --> 00:01:47.620
 Undertexter av Nicolai Winther Dramatisk musikk so Thank you. Hi everyone and welcome back.

1
00:01:47.620 --> 00:01:53.159
 My name is Victoria Szabo and I'll be chairing this second session in the art track.

2
00:01:53.159 --> 00:01:54.439
 Our theme is about authorship.

3
00:01:54.439 --> 00:01:58.560
 It's a topic that's already come up a lot throughout the day and we look forward to

4
00:01:58.560 --> 00:02:02.280
 hearing from all of the speakers on this subject.

5
00:02:02.280 --> 00:02:06.480
 Our first paper is called Sentience, a Performative Game on

6
00:02:06.480 --> 00:02:11.939
 Tarot-Stealed Language Models and the Ambiguous First Person. The speaker today is Marcus

7
00:02:11.939 --> 00:02:17.120
 Round. Marcus Round is a freelance creative technologist and a computational artist whose

8
00:02:17.120 --> 00:02:22.759
 practice explores language and agency in the age of AI. He defamiliarizes language games

9
00:02:22.759 --> 00:02:30.800
 into absurd, immersive installations that invite participants to confront the mechanics of linguistic choice and responsibility when mediated by AI.

10
00:02:31.300 --> 00:02:35.599
 His work has been exhibited at the V&A in London and TIAT in San Francisco.

11
00:02:35.900 --> 00:02:37.539
 Please join me in welcoming Marcus Round.

12
00:02:43.599 --> 00:02:47.000
 Hello. Hi. Hi, everyone. I'm Marcus Round..

13
00:02:50.000 --> 00:02:52.000
 Hello. Hi. Hi, everyone. I'm Marcus Round..

14
00:03:00.000 --> 00:03:09.139
 Okay. So, hi, everyone. I'm Marcus Round. I recently graduated with a masters in computational art from Goldsmiths University in London and

15
00:03:10.080 --> 00:03:14.960
 Currently looking for a PhD. So if you know of one that's suitable then please do you talk to me?

16
00:03:17.099 --> 00:03:19.099
 But this project is a

17
00:03:21.419 --> 00:03:23.840
 Immersive performance game is what I called it

18
00:03:27.000 --> 00:03:32.000
 I started around the question of when ChatGPT says I, what does it mean? And that's less a question about ChatGPT than it is about the word I.

19
00:03:32.000 --> 00:03:38.000
 And what does it mean? What does it mean when you say I? What does it mean when I say I?

20
00:03:38.000 --> 00:03:45.939
 And I think that's a really interesting question. a question about ChatGPT than it is about the word I. And what does it mean?

21
00:03:46.020 --> 00:03:47.520
 What does it mean when you say I?

22
00:03:47.860 --> 00:03:49.699
 What does it mean when I say I?

23
00:03:52.379 --> 00:03:56.340
 And I use large language models as a tool to kind of investigate that.

24
00:03:56.419 --> 00:03:57.139
 That is what they are.

25
00:03:57.199 --> 00:04:01.379
 They're a computational kind of tool for analysing language.

26
00:04:01.379 --> 00:04:04.219
 And there's a whole suite of tools that have emerged

27
00:04:04.219 --> 00:04:05.340
 in the field of what's

28
00:04:05.340 --> 00:04:07.800
 called mechanistic interpretability, which

29
00:04:07.800 --> 00:04:12.500
 is when these AI labs have sort of created these tools

30
00:04:12.500 --> 00:04:14.840
 through a blind kind of machine learning process.

31
00:04:14.840 --> 00:04:17.759
 And the sort of later step is to then figure out

32
00:04:17.759 --> 00:04:21.220
 what exactly is going on in its quote unquote mind.

33
00:04:21.220 --> 00:04:24.600
 And you can do all sorts of analysis.

34
00:04:24.600 --> 00:04:26.180
 So you can give it a piece of

35
00:04:26.180 --> 00:04:34.279
 text and sort of measure which patterns of neurons within the algorithm fire

36
00:04:34.279 --> 00:04:40.560
 most strongly when it encounters certain words as an example this is a graph of

37
00:04:40.560 --> 00:04:47.240
 what happens one of the many patterns that fires when it encounters the word I.

38
00:04:47.240 --> 00:04:49.639
 This is from a tool called neuronpedia.org.

39
00:04:49.639 --> 00:04:50.639
 Definitely look that up.

40
00:04:50.639 --> 00:04:51.639
 I always give them a shout out.

41
00:04:51.639 --> 00:04:53.980
 It's a great site.

42
00:04:53.980 --> 00:04:55.980
 So this is Sentience.

43
00:04:55.980 --> 00:04:57.980
 I'm going to see if I can play the video.

44
00:04:57.980 --> 00:04:58.980
 Here we go.

45
00:04:58.980 --> 00:05:00.779
 Can I play the video?

46
00:05:00.779 --> 00:05:03.459
 Here we go.

47
00:05:03.459 --> 00:05:07.680
 So it's a performance game for an audience of two.

48
00:05:07.680 --> 00:05:09.079
 I'm there as the performer,

49
00:05:09.079 --> 00:05:10.579
 and we have two participants come in.

50
00:05:10.579 --> 00:05:11.939
 It lasts about 20 minutes.

51
00:05:11.939 --> 00:05:14.100
 They sit down, I do this kind of performance.

52
00:05:14.100 --> 00:05:16.040
 They make some decisions, play this game,

53
00:05:16.040 --> 00:05:18.759
 experience this thing, and then they stand up and leave.

54
00:05:18.759 --> 00:05:19.779
 The next two people come in,

55
00:05:19.779 --> 00:05:21.720
 and I do that over and over and over again

56
00:05:21.720 --> 00:05:23.860
 for an entire weekend.

57
00:05:25.120 --> 00:05:29.480
 And you can see I sort of look like this ambiguous poker

58
00:05:29.480 --> 00:05:32.240
 dealer, bureaucrat, tarot reader figure.

59
00:05:34.720 --> 00:05:40.980
 What's going on is that I hand them cards.

60
00:05:40.980 --> 00:05:44.019
 And it looks a bit like they're playing poker,

61
00:05:44.019 --> 00:05:45.019
 but they're playing cards one at a time onto the table. And each of these it looks a bit like they're playing poker, but they're playing

62
00:05:45.019 --> 00:05:50.139
 cards one at a time onto the table. And each of these cards has a little snippet of language that

63
00:05:50.139 --> 00:05:56.360
 I authored. I wrote this, like I wrote each of these phrases and they're all very personal

64
00:05:56.360 --> 00:06:02.839
 things about like my future, what I'm afraid of, what I didn't do, what I want to do, what I hope

65
00:06:02.839 --> 00:06:05.000
 and my family and all this kind of thing.

66
00:06:05.000 --> 00:06:12.000
 And they take turns, between the two of them, to place a card onto the table and construct this sentence.

67
00:06:12.000 --> 00:06:17.000
 So that it becomes this very personal...

68
00:06:17.000 --> 00:06:22.000
 So if you look at this example, I'm not ready yet, but I trust myself, and I want to try because I want more, and I am brave.

69
00:06:22.000 --> 00:06:25.639
 That's a sentence that could be constructed by the two participants.

70
00:06:29.660 --> 00:06:30.540
 It becomes a question then of, like, who is this I that's in there?

71
00:06:32.639 --> 00:06:35.379
 Because it's been constructed between two people using cards that I've authored.

72
00:06:36.459 --> 00:06:39.980
 And sometimes these sentences that are played to the table

73
00:06:39.980 --> 00:06:44.339
 seem quite profound or quite meaningful,

74
00:06:44.439 --> 00:06:46.100
 and sometimes they can be quite meaningful.

75
00:06:46.100 --> 00:06:51.139
 I've had people come up to me afterwards and say it brought up memories of their brother

76
00:06:51.139 --> 00:06:55.360
 that had passed on or something like this because it happened to mention that.

77
00:06:55.360 --> 00:06:59.160
 But it is at the end of the day just sort of like a random chance with a little bit

78
00:06:59.160 --> 00:07:06.439
 of human agency sprinkled in, constrained by language and by things like that.

79
00:07:06.439 --> 00:07:10.000
 So I also have these poker chips you might have seen.

80
00:07:10.000 --> 00:07:13.420
 I call them tokens because that's like a pun about AI.

81
00:07:13.420 --> 00:07:16.779
 But I just have this massive deck of cards.

82
00:07:16.779 --> 00:07:18.899
 Well, it's 52 cards, exactly.

83
00:07:18.899 --> 00:07:22.220
 And you can see that they're different when

84
00:07:22.220 --> 00:07:24.600
 upright versus upside down.

85
00:07:24.600 --> 00:07:26.379
 And that is a reference to tarot.

86
00:07:26.439 --> 00:07:28.180
 There's another part of this thing.

87
00:07:28.459 --> 00:07:29.500
 They write this sentence,

88
00:07:29.699 --> 00:07:32.360
 and then they send it off to AI for autocompletion.

89
00:07:32.740 --> 00:07:37.000
 It's this sort of satirical performance thing

90
00:07:37.000 --> 00:07:38.040
 where I get them in and I say,

91
00:07:38.120 --> 00:07:39.579
 you're going to write a personal affirmation

92
00:07:39.579 --> 00:07:41.079
 about your deepest desires,

93
00:07:41.259 --> 00:07:42.720
 but you have to do it with a partner,

94
00:07:42.899 --> 00:07:44.980
 and I'm going to get an AI to autocomplete it for you.

95
00:07:45.540 --> 00:07:48.620
 And before the autocompletion is written,

96
00:07:49.399 --> 00:07:52.480
 they have to draw another card, which is one of these cards.

97
00:07:53.000 --> 00:07:57.339
 And they might draw the chariot, which you can see here, upside down.

98
00:07:57.860 --> 00:08:00.819
 And then you've got a bunch of these tarot cards.

99
00:08:00.879 --> 00:08:02.500
 They don't have imagery. They just have text.

100
00:08:03.199 --> 00:08:06.319
 And you can see that they're grouped around different themes.

101
00:08:06.560 --> 00:08:10.860
 Like if you look at the words that are highlighted in this chariot upside down,

102
00:08:11.220 --> 00:08:15.519
 you've got dispute, defeat, litigation, sort out, dispute, fighting, defeat.

103
00:08:15.680 --> 00:08:16.720
 Like there's a theme there.

104
00:08:16.860 --> 00:08:18.959
 You might not exactly be able to pin down what it is,

105
00:08:19.279 --> 00:08:24.399
 but there is that you can sort of gradually get this vague kind of theme.

106
00:08:24.399 --> 00:08:25.259
 is that you can sort of gradually get this vague kind of theme.

107
00:08:30.220 --> 00:08:30.379
 And that is related to what the chariot reversed is ostensibly about.

108
00:08:31.879 --> 00:08:32.919
 It's sort of divinatory meaning.

109
00:08:36.879 --> 00:08:40.299
 The way that I did that was by feeding the...

110
00:08:43.639 --> 00:08:47.480
 A.E. Waite wrote this sort of canonical book about the divinatory meaning of each of the major arcana,

111
00:08:47.980 --> 00:08:56.879
 and I fed each of those into this tool and extracted the features that fire in the AI's mind when it reads that text.

112
00:08:57.279 --> 00:09:02.679
 So there'll be a feature to do with self-sacrifice or a feature to do with divination and prophecies of the future.

113
00:09:02.700 --> 00:09:03.879
 or a feature to do with divination and prophecies of the future.

114
00:09:10.240 --> 00:09:10.980
 And I then, by extracting that, like that's the hanged man when it's upright.

115
00:09:14.299 --> 00:09:14.659
 And so when the participants draw that particular card,

116
00:09:17.419 --> 00:09:20.720
 I then take those features and I dial that up in the model. So the prompt doesn't change.

117
00:09:20.799 --> 00:09:22.259
 It's actually a very, very simple prompt.

118
00:09:22.379 --> 00:09:24.299
 It's not about prompt engineering or anything like that.

119
00:09:24.860 --> 00:09:27.320
 It's a very simple prompt, simple text completion.

120
00:09:27.779 --> 00:09:31.600
 They write the first half of a sentence, and I just get the AI to complete that text.

121
00:09:33.159 --> 00:09:35.259
 That's the basic function of a large language model.

122
00:09:36.419 --> 00:09:39.700
 But sort of almost without the model knowing, if you want to think about it like that,

123
00:09:39.779 --> 00:09:46.899
 I've actually manipulated its weights by turning up the themes and references

124
00:09:46.899 --> 00:09:52.059
 related to self-sacrifice so it becomes more relevant or more probable to

125
00:09:52.059 --> 00:09:59.320
 mention things about that as it completes their sentence and that will

126
00:09:59.320 --> 00:10:03.220
 change depending on which tarot card they drew. We do that three times so the

127
00:10:03.220 --> 00:10:08.100
 two participants go back and forth to write one first half of a sentence.

128
00:10:08.100 --> 00:10:10.279
 The AI completes the second half,

129
00:10:10.279 --> 00:10:12.019
 then they write the first half, second half,

130
00:10:12.019 --> 00:10:12.919
 first half, second half,

131
00:10:12.919 --> 00:10:15.360
 until you've got three full sentences,

132
00:10:15.360 --> 00:10:17.279
 and that is their sort of like affirmation

133
00:10:17.279 --> 00:10:18.379
 that they've authored.

134
00:10:18.379 --> 00:10:20.580
 It becomes, I haven't sort of gotten into

135
00:10:20.580 --> 00:10:23.620
 the whole dramaturgy and theater of it,

136
00:10:23.620 --> 00:10:24.740
 because that's sort of not what

137
00:10:24.740 --> 00:10:32.559
 this particular paper is about, but it's quite a satirical, bureaucratic process that's going

138
00:10:32.559 --> 00:10:35.200
 on.

139
00:10:35.200 --> 00:10:41.399
 That's my operator interface, which the participants never see, but just to give you the idea of

140
00:10:41.399 --> 00:10:46.340
 what's going on, this is like, you can see the sort of sentences

141
00:10:46.340 --> 00:10:47.580
 that they've written at the top.

142
00:10:47.580 --> 00:10:49.460
 I think these are probably random examples,

143
00:10:49.460 --> 00:10:51.799
 so don't read too much into this particular example.

144
00:10:51.799 --> 00:10:57.700
 But you can just see that I'm there typing in what

145
00:10:57.700 --> 00:11:00.419
 they've written, and then I pick the card that they've drawn,

146
00:11:00.419 --> 00:11:03.419
 and it gives you the features on the top right that

147
00:11:03.419 --> 00:11:04.460
 are to do with that card.

148
00:11:04.460 --> 00:11:12.899
 And then I send that off to the Neuronpedia API and I get a response back.

149
00:11:12.899 --> 00:11:17.379
 That is a little, that's sort of presented to the audience outside of this performance

150
00:11:17.379 --> 00:11:18.379
 space that I'm in.

151
00:11:18.379 --> 00:11:21.620
 There's a little space to sort of show the last one.

152
00:11:21.620 --> 00:11:25.519
 And you can then go and see which features activated most strongly in

153
00:11:25.519 --> 00:11:31.700
 the actual affirmation that they authored. So a little bit of interactive exhibition

154
00:11:31.700 --> 00:11:45.779
 space to it. The, there's a set, so tarot is kind of, it comes up a lot in computational art and could be somewhat cliched maybe,

155
00:11:46.019 --> 00:11:49.679
 but I think it's very relevant to this piece.

156
00:11:50.259 --> 00:11:53.879
 I think, I mean, the fact that these tarot cards just have text on them

157
00:11:53.879 --> 00:12:01.980
 and not imagery is sort of, it's really interesting to think about these models

158
00:12:01.980 --> 00:12:05.919
 whose only perception of the world is through text.

159
00:12:05.919 --> 00:12:08.080
 They have no embodied kind of perception of the world.

160
00:12:08.080 --> 00:12:12.080
 They only experience the world through text.

161
00:12:12.080 --> 00:12:17.320
 And just sort of the way it steers the conversation between a reader and a readee through these

162
00:12:17.320 --> 00:12:31.320
 kind of loosely grouped themes, I thought made it quite relevant to this Tool this is called sparse auto encoder is the name of this thing that lets you sort of draw out the particular

163
00:12:32.559 --> 00:12:35.179
 themes in text and then dial them up in the model

164
00:12:36.379 --> 00:12:39.879
 I'll just finish by mentioning one other part which is

165
00:12:42.919 --> 00:12:45.820
 Before I get the you remember they have three

166
00:12:45.820 --> 00:12:48.240
 auto completions that the AI will give them,

167
00:12:48.240 --> 00:12:50.600
 but before they can receive the auto completion,

168
00:12:50.600 --> 00:12:52.620
 I say, I'm sorry, I'm in character.

169
00:12:52.620 --> 00:12:54.879
 And I say, I'm sorry, but we have a very strict security

170
00:12:54.879 --> 00:12:56.659
 policy in place here.

171
00:12:56.659 --> 00:12:58.960
 And before you can get your auto completion,

172
00:12:58.960 --> 00:13:01.840
 I just need you to prove to me that you're human.

173
00:13:01.840 --> 00:13:05.899
 And which is kind of started as like a funny kind of thing that they have to prove to me that you're human. Which started as a funny kind of thing

174
00:13:05.899 --> 00:13:07.919
 that they have to prove to me that they're human

175
00:13:07.919 --> 00:13:10.000
 before the AI will auto-generate their

176
00:13:10.000 --> 00:13:11.080
 personal statement.

177
00:13:12.340 --> 00:13:14.179
 But it became part

178
00:13:14.179 --> 00:13:14.539
 of the

179
00:13:14.539 --> 00:13:18.220
 dramaturgy of this immersive piece

180
00:13:18.220 --> 00:13:19.860
 where, first of all, I hand them

181
00:13:19.860 --> 00:13:22.080
 this capture, which instead of traffic lights

182
00:13:22.080 --> 00:13:24.159
 it has street signs with words on them

183
00:13:24.159 --> 00:13:28.600
 and I ask you to pick the words that remind you of your mother which is

184
00:13:29.559 --> 00:13:32.080
 like a reference to Blade Runner and be

185
00:13:33.740 --> 00:13:39.279
 sort of again the difference between machine vision and language and like different people's like

186
00:13:39.980 --> 00:13:41.840
 projecting their own

187
00:13:41.840 --> 00:13:43.840
 associations onto language and

188
00:13:44.120 --> 00:13:47.460
 Then there's other ones like I give them a book,

189
00:13:47.460 --> 00:13:48.799
 which is just a table of numbers,

190
00:13:48.799 --> 00:13:49.980
 and I get them to read a number to me

191
00:13:49.980 --> 00:13:53.100
 as if it's like a sort of confirmation code.

192
00:13:53.100 --> 00:13:55.379
 But then the final one, the sort of climax of the piece

193
00:13:55.379 --> 00:13:59.100
 before they can get to the final generation,

194
00:14:00.100 --> 00:14:02.059
 I hand them a book, and I've highlighted

195
00:14:02.059 --> 00:14:04.519
 a single piece of text, I am afraid of dying,

196
00:14:04.519 --> 00:14:06.059
 and they have to look me in the eye.

197
00:14:06.220 --> 00:14:08.679
 I say, you know, you've got to, I'm sorry, but the system says

198
00:14:08.679 --> 00:14:09.879
 you've got to prove to me that you're human.

199
00:14:10.659 --> 00:14:12.899
 So I give them this book, which is Everything is Illuminated

200
00:14:12.899 --> 00:14:18.120
 by Jonathan Safran Foer, and this is like a fictional character

201
00:14:18.120 --> 00:14:19.620
 saying this, written by an author.

202
00:14:19.840 --> 00:14:22.620
 I hand it to them and they have to look me in the eye

203
00:14:22.620 --> 00:14:24.279
 and read the words out loud.

204
00:14:24.620 --> 00:14:25.500
 So I've had like hundreds of people look me in the eye and read the words out loud. So I've had like

205
00:14:25.500 --> 00:14:32.419
 hundreds of people look me in the eye and tell me I'm afraid of dying. And it's just very interesting

206
00:14:32.419 --> 00:14:36.779
 that eye that's there every time they say I and I'm looking and they're looking me in the eye.

207
00:14:36.980 --> 00:14:42.519
 I think that that's there's something very interesting in that that is not captured by

208
00:14:42.519 --> 00:14:45.000
 this piece of analysis here.

209
00:14:46.279 --> 00:14:48.480
 But that will end my presentation, thank you.

210
00:14:54.919 --> 00:14:57.100
 I think we have time for maybe a question,

211
00:14:57.100 --> 00:14:58.399
 if anyone has a question.

212
00:15:00.799 --> 00:15:02.000
 Yes, in the back please.

213
00:15:07.980 --> 00:15:17.000
 Oh, please. Does it work? Yes.

214
00:15:17.379 --> 00:15:23.580
 Okay, so I was wondering if you personally believe in this tarot thing.

215
00:15:23.580 --> 00:15:26.559
 Do you feel like there is meaning behind that,

216
00:15:26.659 --> 00:15:30.620
 or were you just interested in this entertainment

217
00:15:30.620 --> 00:15:33.220
 kind of part of the project?

218
00:15:34.700 --> 00:15:40.620
 I would say that I think...

219
00:15:40.620 --> 00:15:42.740
 I'm not a big tarot person,

220
00:15:43.360 --> 00:15:45.659
 but I've been to a couple of tarot readings,

221
00:15:46.139 --> 00:15:49.100
 and on at least one occasion, the reader has sort of said,

222
00:15:49.659 --> 00:15:51.460
 like, you know, there's nothing mystical going on here.

223
00:15:51.899 --> 00:15:55.519
 The point of this is not that God is speaking from on high.

224
00:15:55.639 --> 00:15:57.240
 It's that we are drawing random cards,

225
00:15:57.899 --> 00:15:59.639
 and that you and I, as the reader and the reedy,

226
00:15:59.779 --> 00:16:04.299
 are sort of drawing our own meaning out from this process.

227
00:16:06.000 --> 00:16:10.480
 And I think that's quite interesting, and I think that's what's going into this piece.

228
00:16:10.480 --> 00:16:17.679
 I think this piece is not about tarot as much as it's about language models and language,

229
00:16:17.679 --> 00:16:20.720
 the human use of language, and tarot is just a tool to kind of get there.

230
00:16:21.039 --> 00:16:23.200
 to kind of get there.

231
00:16:28.200 --> 00:16:30.480
 But it's definitely about just the human projection

232
00:16:35.480 --> 00:16:38.000
 of meaning onto a kind of random or semi-random process

233
00:16:42.460 --> 00:16:44.440
 and how that relates to AI when you're speaking to ChatGPT. That's the kind of semi-random process

234
00:16:44.440 --> 00:16:47.539
 that you're conversing with and you're still projecting a lot of meaning onto that

235
00:16:47.539 --> 00:16:52.120
 conversation. So I think that's what was relevant there.

236
00:16:52.120 --> 00:16:59.820
 Thank you. Quick related question. Do you think that if you'd used a different

237
00:16:59.820 --> 00:17:06.619
 divination system that you would yield significantly different results?

238
00:17:16.359 --> 00:17:16.460
 That's an interesting question. I think tarot was a good one because it is so familiar to people.

239
00:17:20.180 --> 00:17:20.259
 I think that's why it's used so much is that people, in the piece, I never mentioned tarot.

240
00:17:25.859 --> 00:17:28.579
 I never used the word tarot, but people can kind of see what's going on. I think that it's because it's about a conversation as well.

241
00:17:28.579 --> 00:17:35.039
 Like tarot is very much you have a reader and a readie, and it's that kind of one-to-one

242
00:17:35.039 --> 00:17:42.299
 conversation, but done through what ends up being like a very bureaucratic sort of office

243
00:17:42.299 --> 00:17:45.839
 desk persona slash casino gambler thing.

244
00:17:47.599 --> 00:17:50.000
 Yeah, the cards was an important part of it for sure

245
00:17:50.000 --> 00:17:51.579
 because I related it then to poker

246
00:17:51.579 --> 00:17:53.900
 and the sort of random shuffling of cards,

247
00:17:54.119 --> 00:17:54.700
 things like that.

248
00:17:55.720 --> 00:17:55.839
 Yeah.

249
00:17:57.839 --> 00:17:59.039
 All right, great. Thank you.

250
00:17:59.319 --> 00:17:59.660
 Thank you.

251
00:18:15.400 --> 00:18:29.380
 All right, our next talk is called Vertigo, Vertigo, Reconstructing a Cinematic Ideal Through Its Predictive AI Double. And our speaker is Adam Cole.

252
00:18:30.019 --> 00:18:33.880
 Adam Cole deals with the complexities of intimacy and identity in the age of AI

253
00:18:33.880 --> 00:18:37.559
 through works that explore desire in the shadow of artificial representation.

254
00:18:39.279 --> 00:18:39.759
 Okay.

255
00:18:41.240 --> 00:18:43.200
 I printed my notes, which felt very retro,

256
00:18:43.299 --> 00:18:45.119
 but I did not think I'd also be holding a microphone.

257
00:18:45.539 --> 00:18:46.799
 So we'll make it work.

258
00:18:47.579 --> 00:18:48.000
 All right.

259
00:18:48.759 --> 00:18:49.460
 Here we go.

260
00:18:50.160 --> 00:19:00.819
 I'm very excited to present my project, Vertigo Vertigo, which is a scene-for-scene AI reconstruction of Hitchcock's original film using only 2.78% of the film's original frames.

261
00:19:01.480 --> 00:19:03.380
 As was just said, my name is Adam Cole.

262
00:19:03.579 --> 00:19:09.420
 I'm a researcher and lecturer at the Creative Computing Institute in London. And my work does explore desire in the shadow

263
00:19:09.420 --> 00:19:15.279
 of artificial representation. So a little interactive experience. Raise your hand if

264
00:19:15.279 --> 00:19:21.819
 you've seen Vertigo before. Okay. Nice. Well, spoiler alert, I'm going to give you a little

265
00:19:21.819 --> 00:19:26.200
 bit of the plot. It's been out for about 70 years, so you've had your time.

266
00:19:27.180 --> 00:19:32.319
 In this film, a retired detective becomes obsessed with a woman he's hired to follow.

267
00:19:32.319 --> 00:19:39.759
 And after her apparent death, he attempts to remold this new acquaintance into the exact image of this lost love.

268
00:19:40.299 --> 00:19:43.660
 But it turns out this image was entirely artificial to begin with.

269
00:19:44.240 --> 00:19:48.579
 In one of the final scenes, he demands this woman perform his fantasy for him.

270
00:19:48.660 --> 00:19:51.539
 He says, I need you to be Madeline for a while.

271
00:19:52.359 --> 00:19:58.400
 This fixation on constructing a desired ideal finds a modern parallel in our relationship to prompt-based AI tools, right?

272
00:19:58.599 --> 00:20:01.839
 Where we might literally type into a text box, I need you to be.

273
00:20:02.119 --> 00:20:04.839
 into a text box I need you to beat.

274
00:20:08.500 --> 00:20:11.160
 So Vertigo is itself a film about an obsessive reconstruction of an artificial ideal.

275
00:20:11.160 --> 00:20:15.519
 Vertigo Vertigo extends this logic to the material

276
00:20:15.519 --> 00:20:19.099
 of the film itself, treating the canonical text as a probe

277
00:20:19.099 --> 00:20:22.059
 to understand the normative conventions of classical cinema

278
00:20:22.059 --> 00:20:23.579
 encoded in the model.

279
00:20:27.039 --> 00:20:28.859
 So why did we choose Vertigo, right?

280
00:20:28.900 --> 00:20:31.480
 We could have chosen many other films for this experiment.

281
00:20:32.119 --> 00:20:33.319
 There's three key reasons.

282
00:20:33.960 --> 00:20:36.660
 The first has to do with this idea of the cinematic ideal.

283
00:20:37.059 --> 00:20:40.599
 Vertigo is considered by many to be the absolute best movie ever made.

284
00:20:40.740 --> 00:20:41.880
 This is not necessarily my opinion.

285
00:20:41.880 --> 00:20:51.759
 It was just ranked number one and number two in the side and sound polls. It also is very self-aware within the film about this idea of desire and this

286
00:20:51.759 --> 00:20:56.960
 attempt to achieve this artificial ideal that is totally untethered to reality. The second reason

287
00:20:56.960 --> 00:21:02.200
 it was chosen is its relevance to this idea of normative gender representation and the fact that

288
00:21:02.200 --> 00:21:08.700
 it was famously attacked by Laura Mulvey in her initial articulation of the male gaze. And the third reason it was chosen is

289
00:21:08.700 --> 00:21:12.380
 the title sequence developed by the artist John Whitney which is one of the

290
00:21:12.380 --> 00:21:16.640
 absolute first uses of computer-generated imagery in a feature

291
00:21:16.640 --> 00:21:22.140
 film. So the triangulation of these three kind of ideas, the cinematic ideal,

292
00:21:22.140 --> 00:21:25.759
 normative representation, the computational image, forms

293
00:21:25.759 --> 00:21:28.319
 the pillars that make the foundation of this project.

294
00:21:29.720 --> 00:21:31.119
 So what did we actually do?

295
00:21:31.980 --> 00:21:37.519
 We extract keyframes from the original film at every hard cut or every five to six seconds

296
00:21:37.519 --> 00:21:38.539
 for very long takes.

297
00:21:38.940 --> 00:21:44.220
 We then use a large open source model, 14 billion parameter WAN image to video model,

298
00:21:44.380 --> 00:21:48.880
 to do first, last frame interpolation between those extracted keyframes. The reason we use

299
00:21:48.880 --> 00:21:54.460
 this kind of process is that it isolates the model's predictive tendencies. It

300
00:21:54.460 --> 00:21:58.940
 doesn't require semantic injection, there's no scene specific prompt that we

301
00:21:58.940 --> 00:22:02.799
 use from a close-up compared to a long shot, we kind of just are interested in

302
00:22:02.799 --> 00:22:05.019
 what happens in the spaces in between.

303
00:22:06.740 --> 00:22:08.980
 So we reconstructed the entire film,

304
00:22:08.980 --> 00:22:11.220
 and then we had to create some way to evaluate

305
00:22:11.220 --> 00:22:12.099
 what did we make here?

306
00:22:12.099 --> 00:22:13.940
 And there's three ways that we did that.

307
00:22:13.940 --> 00:22:16.500
 The first is a computational analysis

308
00:22:16.500 --> 00:22:19.180
 of the structural difference, the perceptual difference,

309
00:22:19.180 --> 00:22:21.400
 temporal difference between the original

310
00:22:21.400 --> 00:22:22.880
 and the reconstruction.

311
00:22:22.880 --> 00:22:25.599
 We also evaluate this new film kind of qualitatively

312
00:22:25.599 --> 00:22:28.259
 through a close reading to understand where the model fails,

313
00:22:28.460 --> 00:22:30.319
 where it succeeds, and what happens in between.

314
00:22:30.980 --> 00:22:33.279
 And our final form of analysis was enlisting

315
00:22:33.279 --> 00:22:36.240
 sort of expert media theorists to get their opinion

316
00:22:36.240 --> 00:22:38.440
 on what this reconstruction really means.

317
00:22:40.940 --> 00:22:44.180
 So quantitatively, what did we find?

318
00:22:44.680 --> 00:22:52.779
 We kind of created five tiers based on both pixel-based and perceptual measures.

319
00:22:52.779 --> 00:22:55.240
 And we saw the following results.

320
00:22:55.240 --> 00:22:58.319
 About 16% of the frames are nearly identical.

321
00:22:58.319 --> 00:23:03.240
 Here you kind of see the difference is brighter and the darker parts are identical.

322
00:23:03.240 --> 00:23:05.660
 And about 3.6% are severe failure.

323
00:23:05.660 --> 00:23:10.160
 But what's interesting is that the majority of frames are somewhere in the middle, right? It's this

324
00:23:10.880 --> 00:23:12.880
 plausible middle that is

325
00:23:13.440 --> 00:23:17.259
 perceptibly like Vertigo, but not exactly the same as the original.

326
00:23:20.440 --> 00:23:25.099
 But this aggregate number doesn't really tell us what it feels like to watch it.

327
00:23:25.099 --> 00:23:28.680
 And so we wanted to understand where the model fails and where the model succeeds.

328
00:23:29.259 --> 00:23:34.440
 So one of the places the model fails is that these AI reconstructions, they don't talk.

329
00:23:34.500 --> 00:23:39.440
 And they also don't necessarily emote in the same way that their original character does.

330
00:23:40.420 --> 00:23:46.519
 We also find these big spikes of divergences in scenes that deviate from conventional narrative film.

331
00:23:46.759 --> 00:23:52.940
 So the opening sequence created by John Whitney, this sort of psychedelic dream sequence that the main character has,

332
00:23:53.279 --> 00:23:56.240
 these have big spikes in their divergence from the prediction.

333
00:23:57.259 --> 00:24:01.200
 What was surprising is the scenes that also succeeded very well.

334
00:24:01.539 --> 00:24:05.680
 And these include scenes of incredible complexity, like this great tracking

335
00:24:05.680 --> 00:24:11.519
 zoom-in shot of tons of characters in this really ornate kitchen with waiters crossing the screen.

336
00:24:11.980 --> 00:24:18.460
 And the model was actually surprisingly capable of matching the original really closely. So what we

337
00:24:18.460 --> 00:24:28.500
 find is that there's no correlation between interpolation distance and divergence depth.

338
00:24:28.720 --> 00:24:31.119
 So it doesn't matter how many frames you're interpolating necessarily.

339
00:24:31.680 --> 00:24:34.519
 We find that it's really content that governs the fidelity.

340
00:24:34.859 --> 00:24:37.019
 Certain images are easier to predict than others.

341
00:24:39.740 --> 00:24:45.599
 Aesthetically, the most fascinating scenes are not the ones that are catastrophic failures

342
00:24:45.599 --> 00:24:48.220
 and it's not the ones that are perfect successes.

343
00:24:48.220 --> 00:24:52.279
 It's these moments that are near perfect.

344
00:24:52.279 --> 00:24:57.339
 This plausible middle that feels like it could have been a possible take that simply didn't

345
00:24:57.339 --> 00:25:00.940
 make it into the film.

346
00:25:00.940 --> 00:25:05.339
 In these moments, it is really hard to tell which is the original, which is the reconstruction,

347
00:25:05.339 --> 00:25:09.839
 and it produces this legitimate doubt in the viewer between the two, this felt discorrelation

348
00:25:09.839 --> 00:25:13.220
 that creates the aesthetic core of the work.

349
00:25:13.220 --> 00:25:15.779
 So I said we'll come back to our three pillars.

350
00:25:15.779 --> 00:25:18.660
 Let's talk a little bit about this cinematic ideal.

351
00:25:18.660 --> 00:25:22.500
 After consulting several media theorists, they were both, they were all kind of in agreement

352
00:25:22.500 --> 00:25:30.099
 that the intervention was legible, and they were all somewhat surprised about how neat the reconstruction was, how closely it tied

353
00:25:30.099 --> 00:25:35.460
 to the original. There's two readings that we have as to what might be happening here, and they're

354
00:25:35.460 --> 00:25:41.619
 not necessarily mutually exclusive. The first is that perhaps Vertigo is well represented in the

355
00:25:41.619 --> 00:25:46.859
 training set, right? Perhaps it is compressed in the the model itself and this is a form of lossy retrieval for

356
00:25:46.859 --> 00:25:48.640
 the original film.

357
00:25:48.640 --> 00:25:54.339
 Another potential reading is that maybe Hollywood is statistically predictable, right?

358
00:25:54.339 --> 00:25:59.579
 Maybe this film is not so original and that it's easy to predict it based on a consistent

359
00:25:59.579 --> 00:26:06.140
 grammar that the model is trained on from a larger corpus of large video samples.

360
00:26:06.920 --> 00:26:08.779
 Regardless of which one of these is true,

361
00:26:09.640 --> 00:26:12.380
 it's important to recognize that if such a classic canonical film

362
00:26:12.380 --> 00:26:15.720
 can be approximated from such a small sample set,

363
00:26:16.220 --> 00:26:20.480
 then these questions that have arisen already from the 2023 SAG-AFTRA strike

364
00:26:20.480 --> 00:26:23.579
 about likeness, about ownership, about consent,

365
00:26:23.819 --> 00:26:29.000
 are dramatically acute and certainly getting more relevant by the day.

366
00:26:29.000 --> 00:26:35.000
 Let's talk about our second pillar, which has to do with normative representations.

367
00:26:35.000 --> 00:26:39.000
 While Vertigo is often considered this paragon of classical cinema,

368
00:26:39.000 --> 00:26:45.039
 it's also one of the defining examples that Laura Mulvey uses to articulate her ideas about the male gaze,

369
00:26:45.180 --> 00:26:50.099
 which is a really compelling formulation to reconsider in the age of machine vision.

370
00:26:50.740 --> 00:26:53.700
 This is, of course, because machines do not actually have a gaze, right?

371
00:26:53.740 --> 00:26:57.619
 There's no meaningful subject from which an AI model is looking.

372
00:26:58.079 --> 00:27:02.960
 But what it does have is a statistical disposition towards certain images.

373
00:27:02.960 --> 00:27:06.359
 And those images that it is most likely to produce are the

374
00:27:06.359 --> 00:27:11.319
 ones that have been repeated by cinema time and time again. Whether it's the close-up of the blonde

375
00:27:11.319 --> 00:27:16.579
 woman in soft focus, the heroic male lead, the damsel in distress, of course all these cliches

376
00:27:16.579 --> 00:27:21.579
 play out in Vertigo quite beautifully. This reconstruction's pattern of successes and failures

377
00:27:21.579 --> 00:27:29.859
 therefore suggests some inheritance, right? The point is not that the model was intentionally taught to look in a particular way. It's that certain ways

378
00:27:29.859 --> 00:27:34.740
 of looking are simply more statistically likely to emerge. Ways of looking that align with this

379
00:27:34.740 --> 00:27:39.720
 tradition of the male gaze. This is interesting to think about because the rhetoric that surrounds

380
00:27:39.720 --> 00:27:48.619
 these models is this promise that we can produce any image that we desire. If we consider this inheritance, we have to consider the possibility that we will require

381
00:27:48.619 --> 00:27:55.000
 some form of active resistance to flow against this normative current.

382
00:27:55.000 --> 00:27:56.859
 All right.

383
00:27:56.859 --> 00:28:01.619
 And finally, let's talk a little bit about the computational image.

384
00:28:01.619 --> 00:28:05.339
 The paper claims about the present are expressed most effectively through the experience

385
00:28:05.339 --> 00:28:11.000
 of actually watching the work. And here it's in its sort of difference mode where brighter pixels

386
00:28:11.000 --> 00:28:17.599
 represent bigger divergences between the original and the reconstruction. The original and its

387
00:28:17.599 --> 00:28:21.740
 double rhythmically diverge and then harmonize at every keyframe without the viewer necessarily

388
00:28:21.740 --> 00:28:29.259
 being able to say in the moment which is which. And here's where someone's memory of the film and the prediction of the film become difficult to

389
00:28:29.259 --> 00:28:34.079
 separate. But it's important for me to stress that I don't think we are entering some new world of

390
00:28:34.079 --> 00:28:39.839
 unreality, right? We have been inundated with fictional images from when this movie came out

391
00:28:39.839 --> 00:28:46.400
 in 1958 to 50 years before that when cinema got started, really all the way back to the arrival of mechanically reproducible images.

392
00:28:46.900 --> 00:28:52.960
 What is new is this rise of uniquely predictive environments where content is algorithmically generated,

393
00:28:53.240 --> 00:28:55.859
 personalized, and served at speeds beyond perception.

394
00:28:56.559 --> 00:29:03.680
 One of the media theorists I spoke to said that cinema has always had this idea of compression, right? Cell animation, MPEG compression.

395
00:29:04.259 --> 00:29:07.400
 Vertigo Vertigo fits that lineage not as a paradigm shift

396
00:29:07.400 --> 00:29:10.400
 but as an acceleration, a confluence of speed, desire,

397
00:29:10.400 --> 00:29:12.960
 and doubt that will reshape contemporary experience.

398
00:29:14.140 --> 00:29:18.940
 Okay, in conclusion, Vertigo Vertigo shows that

399
00:29:18.940 --> 00:29:21.440
 classical cinema's visual and structural norms

400
00:29:21.440 --> 00:29:23.480
 are, in our opinion, remarkably encoded

401
00:29:23.480 --> 00:29:30.700
 within contemporary generative systems. Quantitatively, we found that about 73% of frames were recognizably tethered to the

402
00:29:30.700 --> 00:29:36.799
 original film, with failures really kind of clustered around scenes that depart from cinematic

403
00:29:36.799 --> 00:29:42.619
 convention. Qualitatively, we explored this plausible middle between a copy and a hallucination.

404
00:29:43.440 --> 00:29:45.279
 We could say that the experience of watching it

405
00:29:45.279 --> 00:29:50.220
 produces its own 21st century vertigo, a felt discorrelation between the canonical original

406
00:29:50.220 --> 00:29:54.960
 and its algorithmically predicted shadow, an experience that I think we're all getting quite

407
00:29:54.960 --> 00:30:00.740
 familiar with as we're inundated with these AI doubles more and more. Rather than marking a

408
00:30:00.740 --> 00:30:09.960
 rupture, the project argues that generative media extends and accelerates the historical logic of cinema itself, its voyeuristic gaze and its desire of perfecting

409
00:30:09.960 --> 00:30:11.900
 an artificial ideal.

410
00:30:11.900 --> 00:30:16.019
 The problem is not whether these images are real or artificial, but how these inherited

411
00:30:16.019 --> 00:30:21.299
 structures are being produced, scaled and naturalized within predictive media environments

412
00:30:21.299 --> 00:30:24.880
 at a rate beyond the grasp of our perception.

413
00:30:24.880 --> 00:30:27.359
 Thank you very much.

414
00:30:30.640 --> 00:30:34.960
 We have time for a couple of questions.

415
00:30:37.660 --> 00:30:43.319
 Just one question or so maybe like a comment on the cemented decision you

416
00:30:43.319 --> 00:30:47.779
 make while your research because you use the word like total failure when

417
00:30:48.440 --> 00:30:50.440
 addressing frames that are

418
00:30:50.859 --> 00:30:58.039
 highly, let's say, differentiated. And I was just wondering what's your view on that, because it might be that this

419
00:30:58.480 --> 00:31:00.480
 differentiation created by the

420
00:31:00.519 --> 00:31:04.319
 degenerative system are both on the one hand like the creativity,

421
00:31:04.319 --> 00:31:06.059
 let's say, of those systems,

422
00:31:06.240 --> 00:31:08.859
 and also as you from the practitioner

423
00:31:08.859 --> 00:31:10.000
 who built those systems.

424
00:31:10.480 --> 00:31:13.099
 Like, don't you feel that the more it diverges,

425
00:31:13.500 --> 00:31:17.319
 it should be giving a more positive notation to that?

426
00:31:17.619 --> 00:31:19.339
 Yeah, that's a really good question.

427
00:31:19.640 --> 00:31:24.599
 I would say that our intention was to measure

428
00:31:24.599 --> 00:31:26.660
 the difference between the original and the reconstruction.

429
00:31:26.660 --> 00:31:32.279
 So we set up a system where failure really just meant more divergent and identical meant

430
00:31:32.279 --> 00:31:35.319
 more identical, more similar.

431
00:31:35.319 --> 00:31:38.519
 That's not a value judgment about which one is better or which one is preferred.

432
00:31:38.519 --> 00:31:46.460
 And you're exactly right that it's predictability, it's tetheredness to convention is actually a negative if you're trying to

433
00:31:46.460 --> 00:31:48.599
 make something that is not just repeating the past.

434
00:31:48.599 --> 00:31:52.779
 So I wouldn't say my goal is to make Vertigo perfectly.

435
00:31:52.779 --> 00:31:56.140
 My dream is not to get a model that is 100% perfect.

436
00:31:56.140 --> 00:32:01.460
 We found this to be an effective tool to understand to what degree are these conventions really

437
00:32:01.460 --> 00:32:07.319
 embedded in the model and what can we understand about how concrete they are?

438
00:32:14.720 --> 00:32:16.420
 Thanks for the presentation.

439
00:32:16.420 --> 00:32:18.240
 It's very interesting.

440
00:32:18.240 --> 00:32:18.859
 I'm curious.

441
00:32:18.859 --> 00:32:21.779
 I'm not really interested in which model specifically it is.

442
00:32:21.779 --> 00:32:24.460
 But the fact that it's not specified

443
00:32:24.460 --> 00:32:29.319
 presents it as if it's kind of like the models do that.

444
00:32:29.319 --> 00:32:31.319
 And so I was wondering if you experimented

445
00:32:31.319 --> 00:32:34.039
 with other models and had other outcomes,

446
00:32:34.039 --> 00:32:37.400
 or if to some extent you could train or retrain

447
00:32:37.400 --> 00:32:40.960
 some models to requalify some observations you had

448
00:32:40.960 --> 00:32:43.559
 as to, let's say, the protagonist's reaction

449
00:32:43.559 --> 00:32:45.579
 in some certain scenes?

450
00:32:50.559 --> 00:32:53.960
 Yeah, so for this experiment, all the numbers are based on one single model, and that model is WAN 2.2, which came out probably about a year ago.

451
00:32:54.480 --> 00:32:58.799
 We didn't do a formal test with other models, but I did try earlier models to see, like,

452
00:32:58.859 --> 00:32:59.960
 has something changed?

453
00:33:00.039 --> 00:33:03.859
 And there is a remark, it was remarkable how well the reconstruction worked.

454
00:33:03.859 --> 00:33:06.359
 I mean, I didn't, the way this project started

455
00:33:06.359 --> 00:33:07.740
 is I just tried one scene, obviously.

456
00:33:07.740 --> 00:33:09.220
 I didn't do the full film to begin,

457
00:33:09.220 --> 00:33:12.660
 and I really was quite shocked about how well it worked,

458
00:33:12.660 --> 00:33:14.140
 and I was like, has this always been the case?

459
00:33:14.140 --> 00:33:14.980
 And it's not true.

460
00:33:14.980 --> 00:33:17.000
 So we don't know.

461
00:33:17.000 --> 00:33:18.819
 It could just be the models got bigger and better.

462
00:33:18.819 --> 00:33:20.740
 It could be that these open source models

463
00:33:20.740 --> 00:33:23.579
 are training on more copyrighted material

464
00:33:23.579 --> 00:33:27.500
 that aligns with classical cinema from the mid-century.

465
00:33:27.500 --> 00:33:30.759
 And it would 100% be interesting to try it with other models

466
00:33:30.759 --> 00:33:33.559
 to see if they behave very differently,

467
00:33:33.559 --> 00:33:35.619
 and especially with commercial models,

468
00:33:35.619 --> 00:33:40.079
 where bigger companies are very opaque

469
00:33:40.079 --> 00:33:42.519
 about where their training data comes from.

470
00:33:42.519 --> 00:33:44.519
 I mean, even open source models are very opaque

471
00:33:44.519 --> 00:33:47.380
 about where their training model data comes from. I mean, even open source models are very opaque about where their training model data comes from.

472
00:33:47.380 --> 00:33:50.220
 So it could be an interesting study to see

473
00:33:50.220 --> 00:33:52.759
 if they behave differently or not.

474
00:34:02.880 --> 00:34:04.259
 Hi, thanks for that.

475
00:34:04.259 --> 00:34:06.880
 So I have two questions for you.

476
00:34:06.880 --> 00:34:16.320
 One is, so you showed us the differentiated images overlaid, but you made this decision

477
00:34:16.320 --> 00:34:21.219
 not to show us the two scenes by themselves to compare.

478
00:34:21.219 --> 00:34:26.920
 So I'm curious how the generated scene holds up in terms of how natural it looks,

479
00:34:27.019 --> 00:34:32.860
 the movements, the facial expressions, etc. And then also, you know, are you considering

480
00:34:32.860 --> 00:34:39.840
 trying this out on a less canonical film to see how that works outside?

481
00:34:40.340 --> 00:34:46.860
 Yeah, so for the first one, the way that we kind of shared it with external researchers to get their feedback was

482
00:34:47.260 --> 00:34:51.559
 one of them side-by-side, one of them as the overlay, and one of them as the difference because I think you're right

483
00:34:51.559 --> 00:34:53.559
 they all give a different experience of

484
00:34:54.280 --> 00:35:00.599
 what you can take away. I would say the side-by-side being more analytical and the overlay kind of being more aesthetic.

485
00:35:00.599 --> 00:35:04.099
 So in our research, we absolutely looked at them side-by-side as well.

486
00:35:04.599 --> 00:35:05.980
 I think

487
00:35:05.980 --> 00:35:11.340
 trying it on another film would be really, really rewarding. I think one of the researchers

488
00:35:11.340 --> 00:35:16.760
 recommended trying it with Cassavetes films. What was the other one he mentioned? Basically

489
00:35:16.760 --> 00:35:23.019
 trying it with more idiosyncratic filmmakers to see if there is a divergence or if it is sort of

490
00:35:23.019 --> 00:35:25.039
 just, you know, if you have two keyframes,

491
00:35:25.260 --> 00:35:27.420
 there's a predictable trend from one to the next.

492
00:35:27.579 --> 00:35:30.420
 So that definitely would be an interesting follow-up project.

493
00:35:31.480 --> 00:35:31.780
 Thanks.

494
00:35:36.300 --> 00:35:39.420
 Yeah, just speaking of, like, trying another film,

495
00:35:39.420 --> 00:35:43.840
 this idea struck me that maybe, like, in that model,

496
00:35:44.340 --> 00:35:47.239
 well, definitely, like, vertigo is definitely one of

497
00:35:47.239 --> 00:35:52.480
 like training materials for that specific model maybe when you are trying to recreate vertigo

498
00:35:52.480 --> 00:35:59.940
 with that model it's like retracing the past in its memory so i would be really curious to see

499
00:35:59.940 --> 00:36:06.320
 how it will behave or perform uh to recreate a film that's totally original.

500
00:36:06.619 --> 00:36:08.659
 It's definitely not in the training set.

501
00:36:09.119 --> 00:36:11.199
 That would be something I really want to see.

502
00:36:11.820 --> 00:36:13.500
 Yeah, I think that's absolutely true.

503
00:36:13.579 --> 00:36:16.039
 I think there's something interesting that a lot of these models

504
00:36:16.039 --> 00:36:18.519
 are advertising themselves as increasingly cinematic.

505
00:36:18.800 --> 00:36:20.780
 That's sort of when a new model comes out,

506
00:36:20.780 --> 00:36:24.059
 that's the metric that they are interested in.

507
00:36:24.360 --> 00:36:25.340
 But I think that the convention that they are interested in but I think that

508
00:36:25.340 --> 00:36:29.760
 the convention that they are implying by cinematic is actually quite limited so I

509
00:36:29.760 --> 00:36:33.440
 think being able to prove that sort of in a creative way would be a worthwhile

510
00:36:33.440 --> 00:36:36.159
 endeavor

511
00:36:55.980 --> 00:36:56.980
 Our next speaker unfortunately is not able to be physically here. It's Tomas Lorenzo.

512
00:36:56.980 --> 00:37:02.139
 He's an associate professor at the University of Colorado Boulder, and the talk is called

513
00:37:02.139 --> 00:37:08.059
 Speculative Pasts, Generative AI and the Political Imagination of History.

514
00:37:21.760 --> 00:37:29.760
 Hello, thank you very much for coming to this talk that is titled Speculative Pasts, Generative

515
00:37:29.760 --> 00:37:32.860
 AI and the Political Imagination of History.

516
00:37:32.860 --> 00:37:36.920
 And very sorry I cannot be there in person.

517
00:37:36.920 --> 00:37:39.760
 My name is Tomas Laurenzo.

518
00:37:39.760 --> 00:37:43.940
 I am an associate professor at the Department of Critical Media Practices at the University

519
00:37:43.940 --> 00:37:45.519
 of Colorado Boulder.

520
00:37:51.280 --> 00:37:52.159
 And what I'm presenting here is Speculative Past, which is an ongoing research and artistic program

521
00:37:57.440 --> 00:38:05.559
 that uses generative artificial intelligence to construct images and videos of histories that did not, could not, or cannot fully appear in the archive. The program approaches the past,

522
00:38:05.559 --> 00:38:07.639
 not as a stable object to be recovered,

523
00:38:07.639 --> 00:38:10.320
 but as a contested field of effect,

524
00:38:10.320 --> 00:38:13.719
 ideology, omission, and technical mediation.

525
00:38:14.940 --> 00:38:18.460
 AI-generated media can operate as a critical apparatus

526
00:38:18.460 --> 00:38:20.940
 for the production of speculative memory,

527
00:38:21.780 --> 00:38:23.460
 not the falsification of history,

528
00:38:23.460 --> 00:38:26.559
 nor the simulation of documentary evidence, but

529
00:38:26.559 --> 00:38:31.320
 the construction of plausible images that make visible the instability through which

530
00:38:31.320 --> 00:38:35.960
 memory, city and political imagination are produced.

531
00:38:35.960 --> 00:38:42.500
 General systems within this program are seen as instruments of estrangement, whose errors,

532
00:38:42.500 --> 00:38:46.480
 excesses, recognizable biases and capacity for historical pastiche

533
00:38:47.039 --> 00:38:52.639
 are not simply problems to be corrected but material conditions through which the politics

534
00:38:52.639 --> 00:38:59.840
 of representation can be examined. Speculative systems therefore asked how synthetic images

535
00:38:59.840 --> 00:39:05.000
 may help articulate memories marked by, among other things,

536
00:39:05.000 --> 00:39:07.780
 dictatorship, interrupted modernization,

537
00:39:07.780 --> 00:39:10.380
 artistic inheritance, and the exhaustion

538
00:39:10.380 --> 00:39:13.780
 of official narratives, while remaining ethically distinct

539
00:39:13.780 --> 00:39:16.820
 from claims of historical verification.

540
00:39:16.820 --> 00:39:19.760
 So I'm gonna now present four artworks

541
00:39:19.760 --> 00:39:23.980
 created within the program, and the first one is from 2024,

542
00:39:23.980 --> 00:39:25.460
 and it's titled Abandoned Fictions.

543
00:39:26.699 --> 00:39:32.780
 An aim of Uruguayan on abandoned fictions reflects on Uruguay's mid-20th century industrialisation,

544
00:39:33.440 --> 00:39:38.880
 and specifically on the import substitution model preserved under Luis Valleveros, president

545
00:39:38.880 --> 00:39:46.659
 in the mid-50s, which produced real social progress before its industries declined under structural

546
00:39:46.659 --> 00:39:48.880
 and global market pressure.

547
00:39:48.880 --> 00:39:56.539
 So it approaches that trajectory through speculation rather than documentary representation.

548
00:39:56.539 --> 00:40:02.519
 Large format images generated with stable diffusion depict fictional abandoned factories

549
00:40:02.519 --> 00:40:06.159
 drawn from the architectural language of mid-century Uruguayan

550
00:40:06.159 --> 00:40:14.880
 industry. And the locations clearly do not exist. The model assembles composites from learned

551
00:40:14.880 --> 00:40:20.960
 patterns, algorithmic interpretations of an unrealized future, a modernity that once appeared

552
00:40:20.960 --> 00:40:26.840
 attainable and remained incomplete. Abandoned Fictions earned, was awarded the first prize

553
00:40:26.840 --> 00:40:30.559
 at the inaugural, the first Colorado AI Art Competition

554
00:40:30.559 --> 00:40:31.840
 in 2024.

555
00:40:31.840 --> 00:40:36.079
 And then I continued exploring this collection

556
00:40:36.079 --> 00:40:40.800
 or this perspective of fictional factories

557
00:40:40.800 --> 00:40:47.039
 on a title that Abandoned Features in 2025, which extended our work into a serial

558
00:40:47.039 --> 00:40:53.519
 meditation on the same industrial project with large format images being generated. So

559
00:40:54.480 --> 00:41:01.199
 these are examples of the series of images.

560
00:41:09.280 --> 00:41:19.360
 series of images. The second artwork I want to present is titled Montevideo, Montevideo 1983. Montevideo is the capital city of Uruguay. It was commissioned by the Sote Museum that

561
00:41:19.360 --> 00:41:26.760
 belongs to the city. And this work reconstructs the city of my childhood during the final years of

562
00:41:26.760 --> 00:41:33.159
 the Uruguayan civic military dictatorship that went from 1973 to 1985. It works not

563
00:41:33.159 --> 00:41:39.400
 from documentation but from fragments, childhood perception, inherited fears, street atmospheres,

564
00:41:39.400 --> 00:41:47.199
 family stories, political silence. For a child, such a context is not political analysis. It is tone, prohibition,

565
00:41:47.199 --> 00:41:54.480
 adult silence, faces on television, the sense that some things cannot be said. And in 1983,

566
00:41:54.480 --> 00:42:01.199
 I was six years old and everything was terrible. So the art world approaches the past as a field

567
00:42:01.199 --> 00:42:09.000
 of political and effective reconstruction. And generative AI matters here, because the world concerns the impossibility of direct access.

568
00:42:09.000 --> 00:42:12.000
 I cannot photograph my childhood.

569
00:42:12.000 --> 00:42:18.000
 I can only reconstruct an image of the distance between it and the present.

570
00:42:18.000 --> 00:42:23.000
 Childhood memory is not a stable recording, but a constantly reconstructed narrative.

571
00:42:23.000 --> 00:42:26.519
 And generative systems offer a deliberately imperfect analogy

572
00:42:26.519 --> 00:42:28.320
 for that reconstruction.

573
00:42:28.320 --> 00:42:31.400
 And it's imperfect because they do not only infer.

574
00:42:31.400 --> 00:42:34.360
 These models have been shown to memorize stream images

575
00:42:34.360 --> 00:42:36.400
 and emit them at generational time.

576
00:42:36.400 --> 00:42:39.559
 So what returns is neither reliably invented,

577
00:42:39.559 --> 00:42:42.239
 nor reliably remembered.

578
00:42:42.239 --> 00:42:45.619
 And the difference cannot be established from outside.

579
00:42:45.619 --> 00:42:49.699
 So I'm gonna show a portion of this video.

580
00:42:56.780 --> 00:42:59.940
 And this is of course not how the city looked in 1983.

581
00:42:59.940 --> 00:43:02.119
 This is an effective reconstruction.

582
00:43:02.119 --> 00:43:30.000
 It's an artwork. so

583
00:43:54.000 --> 00:44:02.000
 I'm going to fast forward a little bit, for the sake of time. An important part of the artwork is the cabezudos.

584
00:44:02.000 --> 00:44:08.000
 The cabezudos are like large-headed fierce characters from the local carnival.

585
00:44:08.000 --> 00:44:14.000
 They stopped being part of the carnival in 2021, but they were part of my childhood's carnival.

586
00:44:14.000 --> 00:44:20.000
 And myself and everybody my age, we were all absolutely terrified by these figures.

587
00:44:20.000 --> 00:44:28.340
 And I decided to insert that memory into this reconstruction and it's not a

588
00:44:28.340 --> 00:44:35.619
 representation of how they did actually look like but it's an effective again

589
00:44:35.619 --> 00:45:01.159
 recollection on here they are The video ends with a reconstruction of a building that looks a little bit like one

590
00:45:01.159 --> 00:45:05.320
 very famous building in the city and also a little bit like the

591
00:45:05.320 --> 00:45:16.440
 Senate building and it makes it explicit that this is not a photographic

592
00:45:16.440 --> 00:45:28.400
 reconstruction by adding this generated image. The second artwork I want to show is titled Hommage Numerique, that's

593
00:45:28.400 --> 00:45:35.139
 digital homage in French, and it's a work that engages the legacy of early

594
00:45:35.139 --> 00:45:41.059
 20th century avant-garde cinemen, Rotman, Egeling, Len Lai and very

595
00:45:41.059 --> 00:45:47.179
 especially Hans Richter. It revisits that tradition through contemporary generative AI.

596
00:45:47.179 --> 00:45:48.860
 And the film does not claim to escape

597
00:45:48.860 --> 00:45:52.320
 the recognizable aesthetic of current models,

598
00:45:52.320 --> 00:45:53.960
 but it works against the assumption

599
00:45:53.960 --> 00:45:57.679
 that this aesthetic is the medium's natural or only form.

600
00:45:57.679 --> 00:46:01.860
 And it treats it as a historically specific platform style,

601
00:46:01.860 --> 00:46:47.119
 something with a history and therefore an outside. I'm again cutting short.

602
00:46:47.119 --> 00:46:56.420
 The videos are all available online.

603
00:46:56.420 --> 00:47:04.940
 The third artwork is titled La Masa, the mouse, but it's also a boardplay with NASA, the American

604
00:47:04.940 --> 00:47:05.000
 space agency. is also a board play with NASA, the American Space Agency.

605
00:47:06.500 --> 00:47:10.300
 And La Masa is a surrealist short in black and white,

606
00:47:10.300 --> 00:47:12.699
 blending science fiction, historical imagination,

607
00:47:12.699 --> 00:47:14.440
 and invented ritual.

608
00:47:14.440 --> 00:47:18.039
 It recounts the only successful interplanetary voyage

609
00:47:18.039 --> 00:47:21.820
 of a fictional Mexican space agency in the 1960s,

610
00:47:21.820 --> 00:47:23.320
 said to have combined

611
00:47:23.320 --> 00:47:26.119
 mid-century technological experimentation

612
00:47:26.119 --> 00:47:29.019
 with a ceremonial apparatus the film invents

613
00:47:29.019 --> 00:47:30.320
 rather than documents.

614
00:47:31.159 --> 00:47:36.159
 And it's narrated by Sol's surviving astronaut.

615
00:47:36.159 --> 00:47:38.400
 It follows its recruitment and training

616
00:47:38.400 --> 00:47:40.019
 until the boundary between technical

617
00:47:40.019 --> 00:47:41.500
 and ceremonial knowledge dissolves.

618
00:47:41.500 --> 00:47:46.000
 And it's clearly an invented mythology it does

619
00:47:46.000 --> 00:47:50.880
 not claim any kind of knowledge about the traditions of Mesoamerica and

620
00:47:50.880 --> 00:47:56.739
 it's Ayam Uruguayan and this was clearly a foreign construction which was also an

621
00:47:56.739 --> 00:48:08.239
 exploration of how these imageries can represent perspectives that are created from the outside but again the

622
00:48:08.239 --> 00:48:26.360
 the narration makes it clear that this is kind of like a dreamt story Cumplí 18 años en 1960.

623
00:48:27.860 --> 00:48:30.079
 Toda mi vida había estado soñando con el espacio.

624
00:48:30.820 --> 00:48:35.539
 Las galaxias, el ejército, el presidente de la UPABOM,

625
00:48:35.539 --> 00:48:37.659
 el ejército de la UPABOM, el ejército de la UPABOM,

626
00:48:37.659 --> 00:48:40.219
 el ejército de la UPABOM, el ejército de la UPABOM,

627
00:48:40.219 --> 00:48:44.380
 el ejército de la UPABOM, el ejército de la UPABOM,

628
00:48:44.380 --> 00:49:23.119
 el ejército de la UPABOM, el ejército de la UPABOM, Again, I'm going to forward. He is seeking troubles, smiling, his identity dissolves.

629
00:49:30.159 --> 00:49:35.840
 So, Speculative Paths offers a model for engaging AI-generated media beyond the opposition between technical celebration and moral panic, treating the model neither as a creative subject nor

630
00:49:35.840 --> 00:49:42.960
 as a neutral tool, but as a historically situated apparatus that can be appropriated, redirected,

631
00:49:42.960 --> 00:49:48.000
 and made visible. The value of AI-generated media in intrinsic and artistic research

632
00:49:48.000 --> 00:49:52.000
 lies not only in its capacity to produce convincing images,

633
00:49:52.000 --> 00:49:58.000
 but in the opposite, in its capacity to make conviction itself unstable.

634
00:49:58.000 --> 00:50:03.000
 What the project cannot yet claim is that this instability is what audiences experience,

635
00:50:03.000 --> 00:50:05.280
 and that gap is the next problem

636
00:50:05.280 --> 00:50:07.880
 rather than a concluding flourish.

637
00:50:07.880 --> 00:50:10.380
 The past is never simply behind us.

638
00:50:10.380 --> 00:50:13.380
 It is organized, narrated, forgotten, weaponized,

639
00:50:13.380 --> 00:50:15.099
 and sometimes invented.

640
00:50:15.099 --> 00:50:18.579
 So to speculate on the past is not to abandon history,

641
00:50:18.579 --> 00:50:22.980
 but to recognize that history has always been mediated

642
00:50:22.980 --> 00:50:29.760
 and that the present demands forms capable of making that mediation

643
00:50:29.760 --> 00:50:45.000
 visible. So thank you very much. This are all available on my website.

644
00:50:46.340 --> 00:50:47.179
 Thank you so much.

645
00:50:54.000 --> 00:50:55.500
 It's fascinating to see these two

646
00:50:55.500 --> 00:50:57.239
 in juxtaposition to each other,

647
00:50:57.239 --> 00:50:58.559
 and if we have time, maybe at the end

648
00:50:58.559 --> 00:51:00.619
 we can have some more conversation.

649
00:51:01.780 --> 00:51:06.900
 The next talk is called Steering Through the Inner Residue, Feedback as an Artistic Strategy

650
00:51:06.900 --> 00:51:08.679
 Against Generative Images.

651
00:51:08.679 --> 00:51:13.239
 Our speakers are Ting-Chen Liu and Leon Etienne-Kerr.

652
00:51:13.239 --> 00:51:18.719
 Ting-Chen Liu and Leon Etienne-Kerr are artistic associate and lecturer and research associate

653
00:51:18.719 --> 00:51:24.219
 and co-director of the AI Lab at Bauhaus University Weimar and Offenbach University of Art and

654
00:51:24.219 --> 00:51:42.280
 Design.

655
00:51:42.800 --> 00:51:42.920
 Hello, hello.

656
00:51:43.480 --> 00:51:44.000
 Ah, yeah, no.

657
00:51:48.420 --> 00:51:53.139
 Yeah, thanks for coming to our paper presentation for steering through the inner residue, feedback as an artistic strategy against artificial generative images.

658
00:51:53.139 --> 00:51:56.920
 That was too long, that's why we titled the talk, Will the Future of the Internet Be Purple?

659
00:51:56.920 --> 00:51:59.219
 It will become clear what that means later.

660
00:51:59.219 --> 00:52:03.840
 So my name is Leonid Yankeur, like I already said, I'm a scientific associate and co-lead

661
00:52:03.840 --> 00:52:06.000
 of the AI Lab in HFG Offenbach.

662
00:52:06.000 --> 00:52:11.000
 My name is Ting-Chung Liu. I teach at Bauhaus Uni Weimar.

663
00:52:11.000 --> 00:52:30.920
 And so we work together as a duo where we start to do a lot of interaction from the inside to understand how generative system functions and today we're presenting the project steering through the inner residue which is a video installation that we work on trying to

664
00:52:30.920 --> 00:52:36.559
 understand what the mechanism of AI generated image but before that we want

665
00:52:36.559 --> 00:52:41.780
 to give some bit more context into where we reach this point does anyone of you

666
00:52:41.780 --> 00:52:45.039
 familiar with the piece from Alvin Dussire, I'm Sitting in the Room?

667
00:52:46.239 --> 00:52:48.699
 So, ah, I see hands.

668
00:52:48.699 --> 00:52:55.380
 So basically what happened was that he had two tape recorders and he speaks a long sentence,

669
00:52:55.380 --> 00:52:59.059
 or more like a passage, including the sentence I'm sitting in the room.

670
00:52:59.059 --> 00:53:04.559
 He recorded it, play it back, use another tape recorder, record it, play it back, record it,

671
00:53:04.559 --> 00:53:05.400
 play it back, record it, play it back. use another tape recorder, record it, play it back, record it, play it back,

672
00:53:05.400 --> 00:53:06.739
 record it, play it back.

673
00:53:06.739 --> 00:53:09.340
 As you might have guessed, the vibration

674
00:53:09.340 --> 00:53:11.539
 and the reverberation of the room

675
00:53:11.539 --> 00:53:14.780
 sort of records also into the final result.

676
00:53:14.780 --> 00:53:17.320
 In the end, you hear those humming sounds

677
00:53:17.320 --> 00:53:19.019
 coming out from the tape recorder.

678
00:53:19.019 --> 00:53:21.980
 We try to implement the same mechanism

679
00:53:21.980 --> 00:53:23.960
 on AI-generated image.

680
00:53:25.099 --> 00:53:33.940
 Another context that sort of informed us is the idea of cybernetic, how systems self-regulate.

681
00:53:33.940 --> 00:53:38.239
 At the same time, there is a concept that sort of pops out in 24, which is called model

682
00:53:38.239 --> 00:53:39.239
 collapse.

683
00:53:39.239 --> 00:53:46.980
 I think it becomes sort of mainstream nowadays about how when AI generated content being released into society and then

684
00:53:46.980 --> 00:53:54.300
 being trained back again is sort of sabotage or damage the generated outcomes.

685
00:53:54.300 --> 00:53:55.940
 And this is what we try to anticipate.

686
00:53:55.940 --> 00:53:59.880
 Of course, we're not trying to rebuild the whole training process, but we try to speculate

687
00:53:59.880 --> 00:54:09.320
 what could have happened in the image domain? Okay, so before we start with, don't worry, you don't have to read the code.

688
00:54:09.320 --> 00:54:11.960
 There's not much code there.

689
00:54:11.960 --> 00:54:16.340
 We want to start with a toy example before we start talking about generative AI coming

690
00:54:16.340 --> 00:54:18.500
 from feedback and complex systems.

691
00:54:18.500 --> 00:54:24.400
 So the idea is consider an image and an operation in this sense a blur, it doesn't matter which

692
00:54:24.400 --> 00:54:27.559
 one, and you just repeatedly apply this blur to an image and an operation in this sense a blur, it doesn't matter which one, and you just repeatedly apply this blur to an image.

693
00:54:27.559 --> 00:54:30.619
 I mean we can all guess what happens, right? The image becomes

694
00:54:31.579 --> 00:54:34.579
 progressively blurrier, but there's this very interesting

695
00:54:35.840 --> 00:54:40.460
 phenomenon if you add a second operation in this

696
00:54:41.099 --> 00:54:45.639
 case a sharpen filter over the image and you do this progressively.

697
00:54:45.639 --> 00:54:49.260
 I mean you could now think what happens, right? Do they cancel out? Does nothing happen?

698
00:54:49.519 --> 00:54:54.820
 But this is actually what happens, right? There are these patterns that appear called Turing patterns named after

699
00:54:55.500 --> 00:55:02.940
 Turing's paper on morphogenesis. And the interesting part becomes, and that also is the point that Turing made in his paper, is that

700
00:55:02.800 --> 00:55:07.000
 and that also is the point that Turing made in his paper, is that the instructions for the patterns

701
00:55:07.000 --> 00:55:09.000
 are not in the code themselves,

702
00:55:09.199 --> 00:55:11.179
 but they emerge from the instructions that,

703
00:55:11.539 --> 00:55:12.940
 in his case, he was talking about DNA,

704
00:55:13.400 --> 00:55:14.420
 but we talk about code.

705
00:55:14.880 --> 00:55:18.860
 So we can't really disentangle the patterns

706
00:55:18.860 --> 00:55:21.880
 from the instructions that make these patterns,

707
00:55:22.219 --> 00:55:24.760
 which is very important when talking about AI,

708
00:55:24.900 --> 00:55:25.059
 because if you ever opened up a generative AI system, that make these patterns, which is very important when talking about AI, because

709
00:55:25.059 --> 00:55:29.780
 if you ever opened up a generative AI system, you will see that there's lots of

710
00:55:29.780 --> 00:55:32.900
 recursion and feedback and different paths interacting with each other, right?

711
00:55:32.900 --> 00:55:37.599
 Autoregressive next token prediction, right, is feedback in itself. And diffusion

712
00:55:37.599 --> 00:55:41.659
 models that generate images or videos themselves also have this iterative

713
00:55:41.659 --> 00:55:45.179
 feedback structure. But we wanted to apply this to

714
00:55:46.000 --> 00:55:51.980
 to image models, but not to the inner workings of image models, but from the outside, so a system of systems.

715
00:55:51.980 --> 00:55:55.559
 So what we did, it's pretty basic. It's like the loop I showed you before. We took

716
00:55:56.079 --> 00:56:03.360
 image models and we gave them an initial starting image, and we just let it reproduce this image again, and we did this over and over again.

717
00:56:04.719 --> 00:56:06.400
 What happens you will see soon.

718
00:56:06.400 --> 00:56:07.840
 So this is what it looks like, right?

719
00:56:07.840 --> 00:56:09.340
 It's four lines of code, right?

720
00:56:09.340 --> 00:56:12.400
 There's some boilerplate hidden there, but it really is not much.

721
00:56:12.400 --> 00:56:15.360
 And important is that we remove the prompt, right?

722
00:56:15.360 --> 00:56:20.039
 Prompts are very deceptive because they encode also our expectation of what we expect.

723
00:56:20.039 --> 00:56:21.980
 So it's a lot of deception.

724
00:56:21.980 --> 00:56:25.760
 We self-deceive when we write, I want to see a cat wearing a hat.

725
00:56:25.760 --> 00:56:28.480
 And so I will also interpret the image in this sense.

726
00:56:28.480 --> 00:56:30.400
 So we removed this from the equation

727
00:56:30.400 --> 00:56:32.019
 and we left out the prompt

728
00:56:32.019 --> 00:56:35.360
 and only did it on images recursively.

729
00:56:35.360 --> 00:56:38.599
 So taking the output and putting it back in again.

730
00:56:38.599 --> 00:56:43.300
 So what you see here is a continuation of this operation.

731
00:56:43.300 --> 00:56:45.920
 We have this input, we use the output as an input again,

732
00:56:45.920 --> 00:56:47.460
 output as an input again.

733
00:56:47.460 --> 00:56:49.460
 And we try to speed it up,

734
00:56:49.460 --> 00:56:52.099
 and you see an interesting pattern sort of emerged.

735
00:56:53.340 --> 00:56:54.840
 That it became purple.

736
00:56:54.840 --> 00:56:57.780
 At the same time, because of those little tiny deviation,

737
00:56:57.780 --> 00:56:59.920
 it's sort of also animated.

738
00:56:59.920 --> 00:57:02.199
 Just because there's tiny little difference

739
00:57:02.199 --> 00:57:04.280
 from individual frames.

740
00:57:04.280 --> 00:57:09.320
 At the same time, we start to question where does this purple comes from? Like what does it mean?

741
00:57:09.320 --> 00:57:14.320
 Is it the same as the Turing pattern that we just saw? We try to have a different initial conditions as

742
00:57:14.760 --> 00:57:17.400
 the input, but all of them sort of

743
00:57:17.920 --> 00:57:19.920
 collapse into the same purple.

744
00:57:28.940 --> 00:57:33.559
 So, our question then became where does this come from, right? Is this something that comes from the training data or does it come from the models itself?

745
00:57:33.559 --> 00:57:37.139
 So we tried different models that kind of have similar architectures but not exactly

746
00:57:37.139 --> 00:57:39.360
 and similar training data but not exactly.

747
00:57:39.360 --> 00:57:43.539
 And what we could see very quickly is that the actual steering part that steers the image

748
00:57:43.539 --> 00:57:45.980
 into patterns, into special patterns

749
00:57:45.980 --> 00:57:49.320
 is the model assemblage themselves.

750
00:57:49.320 --> 00:57:58.159
 So you can see here, we tried, let's just let it run a little bit.

751
00:57:58.159 --> 00:58:01.699
 So it's very interesting to see that different models have different attractors that they

752
00:58:01.699 --> 00:58:07.980
 kind of converge to different patterns and this is very interesting because I think when we talk about

753
00:58:08.679 --> 00:58:13.619
 the steering that happens as a semantic steering that happens in generative AI we think a lot about

754
00:58:14.380 --> 00:58:16.179
 things like

755
00:58:16.179 --> 00:58:22.260
 how hegemonial structures right represent so it's if you would generate a flag it's always an American flag or something like this

756
00:58:22.360 --> 00:58:25.099
 but there's also something that's not in the training data, there's something that

757
00:58:25.099 --> 00:58:29.500
 is coming from the procedure itself, which you can see here. And

758
00:58:29.500 --> 00:58:35.059
 something very interesting is also that sometimes if you just tune the parameters

759
00:58:35.059 --> 00:58:39.000
 just right, you get kind of stability and attractors, conceptual attractors that

760
00:58:39.000 --> 00:58:42.760
 repeat over and over again. And you can see it's very exactly what you would

761
00:58:42.760 --> 00:58:53.679
 expect, it's very mundane patterns of advertising images and stock imagery, right? And in our experiments also, without any prompt,

762
00:58:53.679 --> 00:58:58.139
 we also, there came a series of images of politicians and even Donald Trump came there

763
00:58:58.139 --> 00:59:02.239
 without ever prompting it, right? Just because these conceptual attractors are so strong

764
00:59:02.239 --> 00:59:05.239
 that they kind of steer towards

765
00:59:07.500 --> 00:59:09.019
 kind of a mean of concepts.

766
00:59:12.480 --> 00:59:16.039
 What seems to be also really interesting for us is the idea to coming back to when we're trying,

767
00:59:16.039 --> 00:59:17.320
 when we're referencing Aving du Ciel's

768
00:59:17.320 --> 00:59:18.539
 I'm Sitting in the Room,

769
00:59:18.539 --> 00:59:22.500
 that in the end, what's being recorded as an outcome

770
00:59:22.500 --> 00:59:24.780
 is not just this humming voice,

771
00:59:24.780 --> 00:59:27.519
 it was the sort of attribute of the room.

772
00:59:27.900 --> 00:59:31.079
 If that performance was done in a different room, the end result would be different

773
00:59:31.079 --> 00:59:36.480
 just because how room sounds reflect as how I'm holding this microphone and speaking to you.

774
00:59:36.539 --> 00:59:38.659
 It might be different when I'm speaking in the toilet.

775
00:59:39.219 --> 00:59:42.059
 So it's the same situation happening with different models.

776
00:59:42.480 --> 00:59:46.500
 Individual models also carry specific attributes.

777
00:59:46.500 --> 00:59:51.000
 There is maybe different floating pool calculations, there may be different architecture how the

778
00:59:51.000 --> 00:59:56.099
 feedback was run, but using this sort of outside of the black box mechanism that we're trying

779
00:59:56.099 --> 01:00:02.539
 to apply here, we also see the tendency of different models, whether it behaves. Is there

780
01:00:02.539 --> 01:00:07.139
 something specific that it really likes?

781
01:00:10.639 --> 01:00:11.239
 So maybe that's also why the question about

782
01:00:14.260 --> 01:00:15.119
 will the internet be purple was that a lot of people talk about

783
01:00:20.300 --> 01:00:27.500
 model collapse in the sense of polluting the training data with synthetic data and then where will this lead, right? But our question was also just about influence, so just about running the models, right we take a picture with our phone, it's already run through an AI model.

784
01:00:27.500 --> 01:00:30.019
 If we post it to Instagram, it's probably run through an AI model.

785
01:00:30.019 --> 01:00:32.659
 YouTube is using AI now to compress our videos.

786
01:00:32.659 --> 01:00:37.659
 So kind of what happens, what's the steering that happens also algorithmically just by

787
01:00:37.659 --> 01:00:41.599
 AI's interpreting these images?

788
01:00:41.599 --> 01:00:44.960
 And so here, this is the final installation.

789
01:00:44.960 --> 01:00:47.699
 It's a continuous loop of like, I think it's 16,000 frames.

790
01:00:47.699 --> 01:00:50.800
 And the only thing that we do is that we change out the model.

791
01:00:50.800 --> 01:00:52.760
 So we have six starting images.

792
01:00:52.760 --> 01:00:54.199
 And the only thing that we change

793
01:00:54.199 --> 01:00:55.760
 is not the parameters or anything.

794
01:00:55.760 --> 01:00:56.940
 We just change out the model.

795
01:00:56.940 --> 01:00:59.500
 And you can see how it kind of iterates

796
01:00:59.500 --> 01:01:02.619
 through this possibility space of possible images

797
01:01:02.619 --> 01:01:05.000
 just through changing out the model.

798
01:01:05.000 --> 01:01:06.000
 That's it.

799
01:01:06.000 --> 01:01:07.440
 I think that's it.

800
01:01:07.440 --> 01:01:08.440
 Thank you.

801
01:01:08.440 --> 01:01:23.159
 I believe this is our last talk for today, so we have some time for questions.

802
01:01:23.159 --> 01:01:24.159
 There is one more, I think.

803
01:01:24.159 --> 01:01:25.000
 Are they here? Oh, you are here. Okay, I think. Are they here?

804
01:01:25.000 --> 01:01:26.000
 Oh, you are here.

805
01:01:26.000 --> 01:01:27.000
 Okay.

806
01:01:27.000 --> 01:01:28.000
 I'm sorry.

807
01:01:28.000 --> 01:01:29.000
 Okay.

808
01:01:29.000 --> 01:01:30.000
 So we don't have time for questions.

809
01:01:30.000 --> 01:01:33.000
 Well, maybe one or two questions and then we'll move on.

810
01:01:33.000 --> 01:01:34.000
 Sorry.

811
01:01:34.000 --> 01:01:51.260
 Apologies. Can this be related to certain data being more prevalent in connection to some topics?

812
01:01:52.880 --> 01:01:59.500
 Like, for instance, it was mentioned that Trump essentially generated himself out of nothing

813
01:01:59.500 --> 01:02:02.019
 because he was essentially hidden inside the model.

814
01:02:03.000 --> 01:02:10.860
 was essentially hidden inside the model. Yeah, I think the point we want to make is that we kind of, and this is kind of new with

815
01:02:10.860 --> 01:02:16.159
 generative AI I would say, you can't really differentiate anymore between training data

816
01:02:16.159 --> 01:02:20.760
 and also algorithm, because it kind of fuses together here, so you can't really get one

817
01:02:20.760 --> 01:02:21.760
 or the other.

818
01:02:21.760 --> 01:02:28.760
 And I think a lot of research was done about the training data right but in the last years it kind of it's not so prevalent

819
01:02:28.760 --> 01:02:32.900
 anymore because talking about training data it's really hard to predict what a

820
01:02:32.900 --> 01:02:37.699
 model will output even if you know all of the training data so the algorithm

821
01:02:37.699 --> 01:02:40.639
 plays an important part that was also kind of the the point of this experiment

822
01:02:40.639 --> 01:02:45.860
 and so it's it's impossible to differentiate the two. So we have to interact

823
01:02:45.860 --> 01:02:51.440
 with them. We have to run them actually to know what comes out and what the biases are.

824
01:02:51.440 --> 01:02:58.000
 And that's one way of using feedback to kind of not prompt qualitatively but quantitatively

825
01:02:58.000 --> 01:03:00.000
 interact with it.

826
01:03:00.000 --> 01:03:01.000
 Okay.

827
01:03:01.000 --> 01:03:12.659
 Do we have any other questions?

828
01:03:13.280 --> 01:03:13.559
 Thank you.

829
01:03:14.460 --> 01:03:15.239
 It's interesting.

830
01:03:16.300 --> 01:03:17.340
 It's not so much... I mean, there's a question in it,

831
01:03:17.420 --> 01:03:19.420
 but it's a quick FYI.

832
01:03:19.780 --> 01:03:20.599
 I'm Philippe Passy.

833
01:03:20.619 --> 01:03:22.679
 I'm a professor of creative AI in Vancouver.

834
01:03:23.159 --> 01:03:24.699
 Technically speaking, this is very well known,

835
01:03:24.760 --> 01:03:27.159
 and I could have predicted that before you started doing it.

836
01:03:27.360 --> 01:03:28.619
 So it's called fixed point theory,

837
01:03:28.739 --> 01:03:31.079
 and there's two types of fixed points you can attain,

838
01:03:31.260 --> 01:03:33.679
 you know, stable fixed point and unstable fixed point.

839
01:03:33.860 --> 01:03:35.019
 You get a great example of both.

840
01:03:35.059 --> 01:03:37.059
 Did you know about the fact that this is all known?

841
01:03:39.440 --> 01:03:42.059
 I mean, we had, like, references also from chaos,

842
01:03:42.059 --> 01:03:44.199
 and complex systems, and kind of this idea.

843
01:03:44.820 --> 01:03:46.400
 But what the

844
01:03:46.400 --> 01:03:49.639
 fixed points are, that still was unknown to us.

845
01:03:49.639 --> 01:03:54.119
 So that this can happen, that it loops or it gets into stable homeostasis or something

846
01:03:54.119 --> 01:03:55.119
 like this.

847
01:03:55.119 --> 01:03:58.780
 Yes, we were kind of predicting or we were kind of expecting something like this, but

848
01:03:58.780 --> 01:04:02.800
 what these points are here, because here they have semantic meaning, or semantic meaning,

849
01:04:02.800 --> 01:04:06.000
 but they say something if it is Trump that over-represents themselves.

850
01:04:06.000 --> 01:04:07.400
 That is what we didn't know, right?

851
01:04:07.400 --> 01:04:09.300
 So like, yeah, in complex systems or something like this,

852
01:04:09.300 --> 01:04:11.900
 sure you have divergence or convergence.

853
01:04:11.900 --> 01:04:14.599
 And I think we were expecting divergence in the beginning.

854
01:04:14.599 --> 01:04:16.300
 And then when it converged, and then also

855
01:04:16.300 --> 01:04:18.300
 into this pink noise, that was kind of interesting to us.

856
01:04:18.300 --> 01:04:20.199
 So really looking at what comes out,

857
01:04:20.199 --> 01:04:23.300
 not only that it converges, but what it converges

858
01:04:23.300 --> 01:04:26.480
 to was the question we were interested in.

859
01:04:29.760 --> 01:04:30.159
 Thank you.

860
01:04:42.480 --> 01:04:49.000
 Our final talk is called Double Blind, Sensibilities of Extinction, Inducing Commemorative Practices

861
01:04:49.000 --> 01:04:55.000
 of Countermapping and Distributed Archiving in an Extended Reality XR Environment.

862
01:04:55.000 --> 01:04:57.739
 Our speaker is Margarita Kroll.

863
01:04:57.739 --> 01:05:02.840
 She is from the head of department in design and lecturer at Vorarlberg University of Applied

864
01:05:02.840 --> 01:05:03.840
 Sciences.

865
01:05:03.840 --> 01:05:08.400
 Margarita Kroll is a transdisciplinary researcher and artist working at the intersection

866
01:05:08.400 --> 01:05:13.639
 of arts, eco-social, as well as science and technology studies, recently focusing

867
01:05:13.639 --> 01:05:32.760
 on the role of affect in more than human relationships. Thank you for the introduction.

868
01:05:32.760 --> 01:05:35.119
 I'm Margaretha Köhler.

869
01:05:35.119 --> 01:05:45.320
 I'm presenting an ongoing artistic research project which I developed together with my colleague David Altweger,

870
01:05:45.880 --> 01:05:50.480
 who is also here, but actually we cannot see him yet.

871
01:05:51.480 --> 01:05:52.579
 Just a moment.

872
01:05:52.579 --> 01:05:52.800
 Yeah.

873
01:05:54.000 --> 01:05:54.400
 No?

874
01:05:54.739 --> 01:05:54.860
 No?

875
01:05:57.619 --> 01:05:58.000
 Okay.

876
01:06:03.300 --> 01:06:06.960
 Because he is there in the online call at the moment.

877
01:06:08.219 --> 01:06:10.420
 But yeah, I'm going to start my talk now.

878
01:06:12.000 --> 01:06:16.360
 In this research project, we focus on biodiversity.

879
01:06:17.260 --> 01:06:22.800
 This is from an overall point of view.

880
01:06:23.480 --> 01:06:25.519
 It tackles the loss of biodiversity,

881
01:06:26.079 --> 01:06:30.739
 which is also integrated into the title.

882
01:06:30.739 --> 01:06:34.239
 So, double-blind sensibilities of extinction.

883
01:06:35.679 --> 01:06:37.699
 And I want to start with the title.

884
01:06:38.039 --> 01:06:40.440
 So, double-blind, what does it refer to?

885
01:06:41.400 --> 01:06:43.860
 While the project may initially appear

886
01:06:43.860 --> 01:06:46.119
 to address the loss of biodiversity in the

887
01:06:46.119 --> 01:06:52.619
 Austrian region of Vorarlberg, its deeper concern lies in the interrelationship between

888
01:06:52.619 --> 01:07:00.179
 hierarchies among species and the ways in which humans situate themselves in relation

889
01:07:00.179 --> 01:07:08.940
 to technologies and biological forms of life. So double-blind, we are all familiar with this term from academia.

890
01:07:10.139 --> 01:07:15.280
 Double-blind is kind of gold standard of experimental method

891
01:07:15.280 --> 01:07:22.219
 and it also guarantees in a way what we call objectivity.

892
01:07:22.219 --> 01:07:26.719
 So we borrow this term for an artistic research project

893
01:07:26.719 --> 01:07:29.599
 to question this notion of objectivity,

894
01:07:29.599 --> 01:07:34.639
 because we use this term to address the ambivalence

895
01:07:34.639 --> 01:07:39.380
 between safety and uncertainty, and also

896
01:07:39.380 --> 01:07:44.380
 to question methods of mapping and archiving, which

897
01:07:44.380 --> 01:07:49.280
 connects to the talk we heard before the last talk

898
01:07:49.280 --> 01:07:57.880
 and also this question of how do we represent biodiversity loss and also this process of

899
01:07:57.880 --> 01:08:20.800
 extinction. So I would like to start with the loss of biodiversity, which has been...

900
01:08:20.800 --> 01:08:28.640
 Approximately one million animal and plant species are at risk of extinction in the coming decades.

901
01:08:28.640 --> 01:08:35.199
 And there's an unprecedented acceleration of biodiversity loss in human history,

902
01:08:35.340 --> 01:08:42.239
 according to the IPBSE Global Assessment Report on Biodiversity and Ecosystem Services.

903
01:08:42.279 --> 01:08:43.720
 on biodiversity and ecosystem services.

904
01:08:48.020 --> 01:08:51.479
 So 75% of Earth's land surface and 66% of the marine environment

905
01:08:51.479 --> 01:08:55.260
 has already been significantly altered by human activity.

906
01:08:56.060 --> 01:08:59.340
 This is not only a story about tropical hot spots,

907
01:08:59.340 --> 01:09:02.680
 which is always framed like this in the media.

908
01:09:03.119 --> 01:09:05.100
 It's also unfolding in the Alpine region.

909
01:09:05.100 --> 01:09:12.560
 This project is made in Vorarlberg, seemingly stable central European

910
01:09:12.560 --> 01:09:21.000
 cultural landscape. Around 68,000 species have been recorded and more than 1,200

911
01:09:21.000 --> 01:09:25.920
 plant species are already listed as threatened according to the Umweltbundesamt.

912
01:09:26.500 --> 01:09:31.960
 So we work together intensively with a nature experience space in Vorarlberg,

913
01:09:32.180 --> 01:09:36.460
 which is called Innatura, and they also provide us with a lot of data.

914
01:09:36.899 --> 01:09:50.140
 And the typical farmland birds, for example, have also lost roughly 40% of their population. What is a pivotal point of this talk

915
01:09:50.140 --> 01:09:59.659
 is that not all species are treated equal in the recording of extinction. Attention, funding,

916
01:10:00.100 --> 01:10:07.000
 mourning, and also media coverage cluster around a small set of charismatic species.

917
01:10:07.979 --> 01:10:09.960
 So there's a kind of inequality.

918
01:10:11.020 --> 01:10:16.899
 And it's mainly the large mammals and animals that are visibly striking

919
01:10:16.899 --> 01:10:19.859
 and emotionally legible to us.

920
01:10:20.859 --> 01:10:37.800
 So mosses, fungi and microbiological organisms, for example, as well as insects and also countless unglamorous species or plants disappear barely noticed.

921
01:10:38.420 --> 01:10:42.819
 Despite comparable or even greater ecological impact.

922
01:10:49.060 --> 01:10:58.880
 or even greater ecological impact. So that's why we focus on these invisible or underrepresented species in the collective history or collective memory

923
01:10:58.880 --> 01:11:06.640
 and would like to give them greater ecological visibility and significance.

924
01:11:06.640 --> 01:11:12.859
 So that unevenness is one of the main points of this talk.

925
01:11:14.199 --> 01:11:22.640
 And that's why we turn to Achille Membe's notion of necropolitics.

926
01:11:23.159 --> 01:11:24.340
 of necropolitics.

927
01:11:26.079 --> 01:11:28.380
 So what is happening is that this biological crisis

928
01:11:28.380 --> 01:11:30.399
 is not only a biological one,

929
01:11:30.399 --> 01:11:35.399
 but also a socio-economical

930
01:11:36.140 --> 01:11:38.779
 and cultural historical one

931
01:11:38.779 --> 01:11:40.720
 or cultural epistemic one.

932
01:11:41.500 --> 01:11:44.180
 And this is also what is called semiocide

933
01:11:44.180 --> 01:11:47.859
 because if the diversity of species

934
01:11:47.859 --> 01:11:56.619
 is extinct, also our realm of memory and the way we can imagine a variety of species is

935
01:11:56.619 --> 01:12:05.619
 diminished. That's why we turned to creating a hybrid or living archive.

936
01:12:06.539 --> 01:12:08.899
 So the archive is something

937
01:12:08.899 --> 01:12:12.000
 that is not a stable storage for us.

938
01:12:12.539 --> 01:12:14.079
 It's a living archive

939
01:12:14.079 --> 01:12:19.060
 that is more something like a cultural memory,

940
01:12:19.359 --> 01:12:21.279
 as Asman calls it,

941
01:12:21.800 --> 01:12:24.340
 a productive repository,

942
01:12:24.960 --> 01:12:28.340
 which is living and where people or also

943
01:12:28.340 --> 01:12:32.279
 other species can contribute to and create

944
01:12:32.279 --> 01:12:36.380
 new meaning. So kind of counteracting this notion of

945
01:12:36.380 --> 01:12:40.380
 the dead storage. So let's have a look

946
01:12:40.380 --> 01:12:44.220
 in our hybrid systems lab which we built

947
01:12:44.220 --> 01:12:47.739
 at the university in Vorarlberg, which

948
01:12:47.739 --> 01:12:59.180
 is a kind of XR setting where you can enter this living archive by accessing it via distributed

949
01:12:59.180 --> 01:13:00.180
 maps.

950
01:13:00.180 --> 01:13:06.020
 So I think David will show it afterwards. We also have got a video where you can see this map

951
01:13:06.020 --> 01:13:12.819
 where you can enter the space of collected specimens,

952
01:13:14.100 --> 01:13:17.739
 but also non-existent species,

953
01:13:18.079 --> 01:13:22.180
 which are created using AI models,

954
01:13:23.100 --> 01:13:28.399
 which also links up to a project we presented two years ago here,

955
01:13:28.800 --> 01:13:33.479
 Sphagnum Denses, and this is kind of the continuation of this project.

956
01:13:34.619 --> 01:13:52.159
 So we can see this map here, which also is an open source archive. We work with local data from an open source repository called Vogis,

957
01:13:52.739 --> 01:13:57.000
 where you can access all the environmental data,

958
01:13:57.520 --> 01:14:02.520
 and you can also transform it into, for example, 3D models.

959
01:14:03.039 --> 01:14:09.359
 into, for example, 3D models. And you can also contribute to this map

960
01:14:09.359 --> 01:14:12.359
 with an accessible interface.

961
01:14:12.359 --> 01:14:14.680
 So maybe, David, you want to jump in now

962
01:14:14.680 --> 01:14:23.279
 how it's technically realized after I showed the video.

963
01:14:23.279 --> 01:14:28.000
 So for us, this map is a tool for counter mapping,

964
01:14:28.000 --> 01:14:30.000
 as Rebecca Solnit calls it,

965
01:14:31.000 --> 01:14:35.800
 where not only the so-called neutral data

966
01:14:35.800 --> 01:14:42.399
 or platform data is accessible or stored,

967
01:14:42.399 --> 01:14:47.279
 but where also effects and emotions or other

968
01:14:47.279 --> 01:14:56.880
 multisensory perceptive objects can be stored so it gets this more than human

969
01:14:56.880 --> 01:15:14.159
 dimensions. Yeah so let's have a look, I don't know if the video works, otherwise David will show us the setting.

970
01:15:14.159 --> 01:15:17.159
 I don't know.

971
01:16:08.500 --> 01:16:08.579
 Es hängt. David, hello. Hello.

972
01:16:08.579 --> 01:16:08.619
 Hello.

973
01:16:13.619 --> 01:16:47.420
 Not yet, just a moment. No, unfortunately it doesn't work.

974
01:16:55.420 --> 01:17:32.180
 Sorry, we cannot hear you. Okay. Sorry.

975
01:17:32.180 --> 01:17:52.899
 I'm going to continue. Wo ist der Kollege?

976
01:18:00.220 --> 01:18:11.680
 Sonst machen wir das jetzt einfach. Ja, das sieht gut aus. Okay.

977
01:18:12.560 --> 01:18:15.760
 Sorry.

978
01:18:16.000 --> 01:18:30.000
 Ja. I don't know why it doesn't work.

979
01:18:30.000 --> 01:19:20.640
 Okay. Sorry, I don't know. Okay, sorry for the interruption.

980
01:19:24.060 --> 01:19:25.979
 Yeah, I mean, we can

981
01:19:25.979 --> 01:19:27.260
 also have a look into

982
01:19:27.260 --> 01:19:28.920
 the lab

983
01:19:28.920 --> 01:19:31.920
 where

984
01:19:31.920 --> 01:19:33.020
 you can see

985
01:19:33.020 --> 01:19:35.920
 the map in the background and

986
01:19:35.920 --> 01:19:37.859
 yeah, you can enter

987
01:19:37.859 --> 01:19:39.119
 this space

988
01:19:39.119 --> 01:19:41.760
 with a head-up

989
01:19:41.760 --> 01:19:43.039
 display or

990
01:19:43.039 --> 01:19:46.680
 it's also accessible via touch.

991
01:19:46.680 --> 01:19:52.119
 And then you can have a look at the different hybrid creatures

992
01:19:52.119 --> 01:19:58.140
 where we also work with AI models

993
01:19:58.140 --> 01:20:04.220
 which are combined, two AI models are combined

994
01:20:04.220 --> 01:20:09.479
 to have these stereoscopic creatures where

995
01:20:09.479 --> 01:20:17.399
 people can also interact with in the room so actually it was created with a

996
01:20:17.399 --> 01:20:28.140
 specific touch designer patch where the two models are combined. And if people enter this space,

997
01:20:28.279 --> 01:20:32.600
 they can really interact with those speculative creatures

998
01:20:32.600 --> 01:20:37.880
 which are generated based on this historical data

999
01:20:37.880 --> 01:20:39.279
 about extinct species,

1000
01:20:40.180 --> 01:20:53.000
 but also still existing creatures or species and also speculative species.

1001
01:20:53.000 --> 01:21:06.920
 So actually we use play as kind of critical methods where people can kind of wander through this space, which draws on this notion of the Deriv,

1002
01:21:07.399 --> 01:21:13.239
 which is also a technique developed by the situationists

1003
01:21:13.239 --> 01:21:15.739
 where you explore your environment

1004
01:21:15.739 --> 01:21:24.479
 and focus on the emotion you have during these Derivs.

1005
01:21:25.000 --> 01:21:32.000
 So it's a kind of ludic assembling of species, audience, tests,

1006
01:21:32.000 --> 01:21:40.000
 and it also reconfigures the rules by which spaces, memories, and futures are organized.

1007
01:21:40.000 --> 01:22:12.000
 So in a way, the main focus of the project was also to kind of cultivate something that is called speculative literacy that enables participants to imagine other ways of living with loss and uncertainty, ways that acknowledge our implications in extinction and in colonial and also platform-driven histories of mapping and collecting.

1008
01:22:12.000 --> 01:22:29.000
 So, yeah, I think I just got an insight into this project which is still developing. And thank you for your attention and your patience also.

1009
01:22:33.000 --> 01:22:34.000
 Thank you.

1010
01:22:34.000 --> 01:22:36.000
 Do you want to answer any questions?

1011
01:22:36.000 --> 01:22:38.000
 Yeah?

1012
01:22:38.000 --> 01:22:41.000
 Do we have time for maybe a question or two?

1013
01:22:41.000 --> 01:22:43.000
 Yes?

1014
01:23:09.239 --> 01:23:09.960
 The phenomenon that less visually appealing species are less cared about when they are that can also be seen in species that are far away.

1015
01:23:15.039 --> 01:23:15.659
 Like, for instance, the blobfish is known for being also endangered,

1016
01:23:18.119 --> 01:23:21.020
 but people don't care about that either because it's not visually appealing.

1017
01:23:27.920 --> 01:23:31.899
 And my question is is there a

1018
01:23:31.899 --> 01:23:33.899
 way to do something against

1019
01:23:33.899 --> 01:23:34.319
 that?

1020
01:23:38.279 --> 01:23:40.000
 I can't

1021
01:23:40.000 --> 01:23:41.020
 get where you are.

1022
01:23:41.020 --> 01:23:41.859
 Sorry.

1023
01:23:42.619 --> 01:23:43.460
 Okay.

1024
01:23:45.000 --> 01:23:46.039
 So actually,

1025
01:23:46.800 --> 01:23:48.600
 this background of

1026
01:23:48.600 --> 01:23:50.539
 necropolitics

1027
01:23:50.539 --> 01:23:53.020
 mainly refers

1028
01:23:53.020 --> 01:23:53.659
 to

1029
01:23:53.659 --> 01:23:56.520
 racial

1030
01:23:56.520 --> 01:23:57.880
 inequality,

1031
01:23:58.899 --> 01:23:59.479
 that some

1032
01:23:59.479 --> 01:24:02.539
 groups of society

1033
01:24:02.539 --> 01:24:03.939
 are overrepresented,

1034
01:24:04.600 --> 01:24:08.600
 and also their extinction is not even mentioned.

1035
01:24:08.600 --> 01:24:17.100
 So if we view ecological systems with this perspective,

1036
01:24:17.880 --> 01:24:32.539
 I think it's very obvious that there are a lot of crucial species which are kind of not discussed in any journal, for example,

1037
01:24:32.539 --> 01:24:39.359
 or any nature archive, for example, the museum we work with.

1038
01:24:54.579 --> 01:24:54.779
 And our approach is to kind of get people to reflect on the different types of extinction and also the different species that are extinct.

1039
01:25:03.220 --> 01:25:03.779
 So that's kind of the way to approach it with this artistic living archive.

1040
01:25:09.680 --> 01:25:20.239
 I don't know if it answered your question but I tried to. But we don't see David. David, can we try, will it work? No, now we can't hear him anymore.

1041
01:25:21.000 --> 01:25:22.000
 Hello?.

1042
01:25:22.000 --> 01:25:23.000
.

1043
01:25:23.000 --> 01:25:24.000
.

1044
01:25:24.000 --> 01:25:25.000
.

1045
01:25:25.000 --> 01:25:26.000
.

1046
01:25:26.000 --> 01:25:27.000
.

1047
01:25:27.000 --> 01:25:28.000
.

1048
01:25:28.000 --> 01:25:29.000
.

1049
01:25:29.000 --> 01:25:30.000
.

1050
01:25:30.000 --> 01:25:31.000
.

1051
01:25:31.000 --> 01:25:32.000
.

1052
01:25:32.000 --> 01:25:33.000
.

1053
01:25:33.000 --> 01:25:34.000
.

1054
01:25:34.000 --> 01:25:35.000
.

1055
01:25:35.000 --> 01:25:36.000
.

1056
01:25:36.000 --> 01:25:37.000
.

1057
01:25:37.000 --> 01:25:38.000
.

1058
01:25:38.000 --> 01:25:39.000
.

1059
01:25:39.000 --> 01:25:40.000
.

1060
01:25:40.000 --> 01:25:41.000
.

1061
01:25:41.000 --> 01:25:42.000
.

1062
01:25:42.000 --> 01:25:43.000
.

1063
01:25:43.000 --> 01:25:44.000
.

1064
01:25:44.000 --> 01:25:45.000
...... Hi. It won't let us do it. Drag this. Let's try this again. Say it again.

1065
01:25:45.000 --> 01:25:46.000
 No, no, they turned it off completely.

1066
01:25:46.000 --> 01:25:47.000
 They don't like our idea.

1067
01:25:47.000 --> 01:25:48.000
 Okay, what's up?

1068
01:25:48.000 --> 01:25:49.000
 Do you want to do it again?

1069
01:25:49.000 --> 01:25:50.000
 No, no, no.

1070
01:25:50.000 --> 01:25:51.000
 I don't want to do it again.

1071
01:25:51.000 --> 01:25:52.000
 Okay, what's up?

1072
01:25:52.000 --> 01:25:53.000
 Do you want to do it again?

1073
01:25:53.000 --> 01:25:54.000
 No, no, no.

1074
01:25:54.000 --> 01:25:55.000
 I don't want to do it again.

1075
01:25:55.000 --> 01:25:56.000
 Okay, what's up?

1076
01:25:56.000 --> 01:25:57.000
 Do you want to do it again?

1077
01:25:57.000 --> 01:25:58.000
 No, no, no.

1078
01:25:58.000 --> 01:25:59.000
 I don't want to do it again.

1079
01:25:59.000 --> 01:26:00.000
 Okay, what's up?

1080
01:26:00.000 --> 01:26:01.000
 Do you want to do it again?

1081
01:26:01.000 --> 01:26:02.000
 No, no, no.

1082
01:26:02.000 --> 01:26:03.000
 I don't want to do it again.

1083
01:26:03.000 --> 01:26:04.000
 Okay, what's up?

1084
01:26:04.000 --> 01:26:05.000
 Do you want to do it again? No, no, no turned it off completely.

1085
01:26:05.000 --> 01:26:06.000
 They don't like our idea.

1086
01:26:06.000 --> 01:26:07.000
 Okay, what's up?

1087
01:26:07.000 --> 01:26:12.000
 Do you want to add anything so I just can?

1088
01:26:12.000 --> 01:26:13.000
 Yeah, then.

1089
01:26:13.000 --> 01:26:14.000
 Can you hear me now?

1090
01:26:14.000 --> 01:26:15.000
 A little bit.

1091
01:26:15.000 --> 01:26:16.000
 Not really.

1092
01:26:16.000 --> 01:26:17.000
 Da, da, da.

1093
01:26:17.000 --> 01:26:18.000
 Da, da, da, da, da, da, da, da, da, da.

1094
01:26:18.000 --> 01:26:19.000
 No, it doesn't work.

1095
01:26:19.000 --> 01:26:20.000
 Okay.

1096
01:26:20.000 --> 01:26:21.000
 That's a change.

1097
01:26:21.000 --> 01:26:22.000
 I think we should skip this experiment.

1098
01:26:22.000 --> 01:26:23.000
 So, I think we should just go back to the beginning.

1099
01:26:23.000 --> 01:26:24.000
 Okay.

1100
01:26:24.000 --> 01:26:30.000
 So, I think we should just go back to the beginning. Okay. So, I think we should skip this experiment.

1101
01:26:30.000 --> 01:26:32.000
 Can we turn the game all the way up?

1102
01:26:32.000 --> 01:26:35.000
 Hello, hello, hello?

1103
01:26:35.000 --> 01:26:37.000
 Yes?

1104
01:26:37.000 --> 01:26:39.000
 Yeah, you hear me?

1105
01:26:39.000 --> 01:26:40.000
 Yeah.

1106
01:26:40.000 --> 01:26:42.000
 Yeah, what's to add?

1107
01:26:42.000 --> 01:26:59.779
 I mean, in general, I would also like to add to what Margarita just said, in terms of, you know, there's hierarchies in all of these kind of like experiences of experiencing this loss of diversity and this loss of culture and this loss of communities.

1108
01:27:30.439 --> 01:27:35.119
 And what I would like to add is a political perspective that is really, really important and a crucial part of our work is that what we recognize a lot and what we see a lot is that a post-humanist perspective and the projects that come out of this way of thinking often engage with different species and different plants and different ecosystems and why why we see this absolutely central to not only developing empathy for other species but also never forget that this is about the destruction of cultures and bodies and

1109
01:27:35.119 --> 01:27:42.079
 and and systems that are also ecosystems and and I feel like it's so important

1110
01:27:42.079 --> 01:27:52.739
 you know in our post humanist perspective to include the solidarity with other communities and to not only focus on the nature part.

1111
01:27:52.739 --> 01:28:05.760
 And I feel like cultivating a practice and teaching people how to use tools and artificial intelligence in order to decolonize their workflows and their practices is really important.

1112
01:28:05.760 --> 01:28:10.579
 And to develop tools that are open source and to go away from software packages.

1113
01:28:14.020 --> 01:28:17.479
 For instance, OpenStreetMaps has been around for so long,

1114
01:28:17.739 --> 01:28:19.840
 and there's so much open source data,

1115
01:28:20.239 --> 01:28:24.560
 but there are so few applications that allow a broader public to experiment

1116
01:28:24.560 --> 01:28:26.119
 with what a map can be.

1117
01:28:26.279 --> 01:28:30.739
 Can it be a repository for poetry, for grievances?

1118
01:28:31.279 --> 01:28:39.539
 And we feel like AI and the hybrid systems at the moment really enables us to connect all of these technologies.

1119
01:28:48.579 --> 01:28:54.000
 And we feel like this is a really, really important time to reclaim all of these technologies that have been colonized by huge, huge technical conglomerates

1120
01:28:54.000 --> 01:28:58.460
 and use those technologies in order to make meaning

1121
01:28:58.460 --> 01:29:07.899
 and to make meaning out of this huge loss of richness that is happening around us and to be dogged about,

1122
01:29:08.020 --> 01:29:13.479
 you know, like holding up the practice of collecting and of archiving and of making

1123
01:29:13.479 --> 01:29:20.659
 memories and making meaning. And that's at the core of it. I say one more example, sorry,

1124
01:29:20.779 --> 01:29:25.359
 I'm probably talking way too long, but one more example is like by decolonizing our

1125
01:29:25.359 --> 01:29:32.800
 workflows i mean for instance we like ai enables us a lot in design to to automate certain things

1126
01:29:32.800 --> 01:29:39.520
 and like for students to be less reliable on software packages in order to create stuff

1127
01:29:39.520 --> 01:29:46.060
 and whenever we create some sort of freedom and space through the use of AI and AI-assisted

1128
01:29:46.060 --> 01:29:51.279
 workflows, we then take that space and fill it up with other practices that have been

1129
01:29:51.279 --> 01:29:51.560
 lost.

1130
01:29:52.140 --> 01:29:54.359
 So we reintroduced screen printing.

1131
01:29:54.560 --> 01:29:58.039
 We reintroduced, like, physical art making and design making.

1132
01:29:58.739 --> 01:30:01.000
 And then, so we work on both ends of the spectrum.

1133
01:30:01.220 --> 01:30:05.979
 And I feel like that's a really, really important thing to do, to not only automate, but also like, what are we going to do

1134
01:30:05.979 --> 01:30:07.899
 with that space that is

1135
01:30:07.899 --> 01:30:09.380
 being made by

1136
01:30:09.380 --> 01:30:11.680
 AI-assisted workflows?

1137
01:30:17.680 --> 01:30:19.960
 Thank you. I'm glad we were able to get you into the

1138
01:30:19.960 --> 01:30:20.760
 room, more or less.

1139
01:30:21.539 --> 01:30:23.100
 Thank you so much, and thank you for your patience.

1140
01:30:23.439 --> 01:30:30.640
 Of course, of course. Well, I think we're out of time, but please join me in thanking all of our panelists for their

1141
01:30:30.640 --> 01:31:20.000
 presentations today. We resume in 30 minutes for the next session. Maybe 29 minutes now. Thank you.