WEBVTT

1
00:00:00.720 --> 00:00:02.799
<v Speaker 1>I was coming up with a series of images for

2
00:00:02.839 --> 00:00:05.080
<v Speaker 1>a story that I'm working on, and I kept running

3
00:00:05.080 --> 00:00:07.719
<v Speaker 1>into a problem. It turns out that this issue is

4
00:00:07.719 --> 00:00:10.679
<v Speaker 1>pretty common among those who try to use artificial intelligence

5
00:00:10.839 --> 00:00:14.000
<v Speaker 1>generators to make images. Once you get the hang of it,

6
00:00:14.000 --> 00:00:16.640
<v Speaker 1>it's pretty simple, though, But no matter how good you

7
00:00:16.679 --> 00:00:19.280
<v Speaker 1>get at it, you keep running into the same thing.

8
00:00:20.280 --> 00:00:22.519
<v Speaker 1>When your image comes up with a character, a specific

9
00:00:22.559 --> 00:00:25.640
<v Speaker 1>person that you want to appear again, it's pretty difficult

10
00:00:25.640 --> 00:00:27.559
<v Speaker 1>to get them to show up just like that and

11
00:00:27.640 --> 00:00:30.679
<v Speaker 1>other images. For example, let's say you're trying to come

12
00:00:30.719 --> 00:00:32.640
<v Speaker 1>up with a short story of a man who finds

13
00:00:32.640 --> 00:00:35.560
<v Speaker 1>a cabin in the woods. He steps inside and encounters

14
00:00:35.560 --> 00:00:38.759
<v Speaker 1>a ghost, so he decides to run away. This man

15
00:00:38.799 --> 00:00:41.600
<v Speaker 1>would start coming up looking differently in the rest of

16
00:00:41.640 --> 00:00:45.240
<v Speaker 1>the images, and some he might have long hair, in

17
00:00:45.280 --> 00:00:49.159
<v Speaker 1>others he might be wearing something different, maybe a different face,

18
00:00:49.399 --> 00:00:53.600
<v Speaker 1>different style of drawing or illustration, maybe even a different background.

19
00:00:54.439 --> 00:00:56.880
<v Speaker 1>But anyway, some artists start to take advantage of this

20
00:00:56.920 --> 00:01:00.320
<v Speaker 1>as a new form of expression, asking AI what it

21
00:01:00.359 --> 00:01:02.719
<v Speaker 1>looks like and then have it give you an image

22
00:01:02.719 --> 00:01:06.400
<v Speaker 1>of itself the AI machine or whatever. And it's such

23
00:01:06.439 --> 00:01:08.719
<v Speaker 1>an interesting idea, right but there was a new thing

24
00:01:08.719 --> 00:01:13.000
<v Speaker 1>that I learned about. There's this technique called negative prompt weights,

25
00:01:13.159 --> 00:01:15.319
<v Speaker 1>and basically all you have to do is tell the

26
00:01:15.359 --> 00:01:18.560
<v Speaker 1>AI to make images that are opposite of a prompt,

27
00:01:18.920 --> 00:01:21.760
<v Speaker 1>Like if you tell it something like give me an

28
00:01:21.799 --> 00:01:24.959
<v Speaker 1>image of Heaven, it might give you hell. Well that's

29
00:01:25.000 --> 00:01:27.799
<v Speaker 1>what we were hoping for at least, but this ended

30
00:01:27.879 --> 00:01:31.879
<v Speaker 1>up getting much much more complicated. You see, it isn't

31
00:01:32.000 --> 00:01:36.560
<v Speaker 1>just analyzing visuals, but concepts, things that aren't exactly the

32
00:01:36.640 --> 00:01:38.799
<v Speaker 1>opposite as you and I perceive them to be, but

33
00:01:38.920 --> 00:01:42.359
<v Speaker 1>yet they're opposite conceptually. And I know this is very

34
00:01:42.400 --> 00:01:44.920
<v Speaker 1>complicated and kind of hard to understand, but it's going

35
00:01:44.959 --> 00:01:46.879
<v Speaker 1>to get a lot easier. So we're going to start

36
00:01:46.920 --> 00:01:49.200
<v Speaker 1>off with this story. I'll tell you about a case

37
00:01:49.239 --> 00:01:51.920
<v Speaker 1>where an artist decided to experiment with this and ended

38
00:01:51.959 --> 00:01:54.959
<v Speaker 1>up with a recurring character, something that's already difficult to do.

39
00:01:55.760 --> 00:01:58.519
<v Speaker 1>Then i'll tell you about some of the more strange

40
00:01:58.599 --> 00:02:02.560
<v Speaker 1>things out there, just how AI is bringing people back

41
00:02:02.560 --> 00:02:07.719
<v Speaker 1>from the dead. My name is Edwin, and here is

42
00:02:07.760 --> 00:02:14.120
<v Speaker 1>a horror story. Much of the original story of the

43
00:02:14.240 --> 00:02:17.000
<v Speaker 1>artists goes back to a series of tweets posted by

44
00:02:17.039 --> 00:02:20.479
<v Speaker 1>Steve mash Swanson, an artist a musician, also known as

45
00:02:20.520 --> 00:02:25.120
<v Speaker 1>super Composite online. The thread she posted begins explaining this

46
00:02:25.240 --> 00:02:29.319
<v Speaker 1>concept of negative prompt weights, essentially making the image look

47
00:02:29.360 --> 00:02:32.520
<v Speaker 1>as different from the prompt as possible. According to her

48
00:02:32.560 --> 00:02:36.439
<v Speaker 1>second tweet, here's what she tried. She typed in Brando

49
00:02:36.759 --> 00:02:40.280
<v Speaker 1>colon colin minus one. She ended up getting a logo

50
00:02:40.400 --> 00:02:43.439
<v Speaker 1>image and from what I see, it's a squarish image

51
00:02:43.439 --> 00:02:47.159
<v Speaker 1>with the words digita like digital without the L and

52
00:02:47.199 --> 00:02:53.759
<v Speaker 1>then pntics pinix or something and then the silhouette of

53
00:02:53.759 --> 00:02:57.199
<v Speaker 1>an old castle or a skyline. It makes no sense.

54
00:02:58.000 --> 00:03:01.759
<v Speaker 1>So as a curious mind's work added, how about reversing

55
00:03:01.800 --> 00:03:04.039
<v Speaker 1>this one? Are we going to get the original image

56
00:03:04.039 --> 00:03:07.919
<v Speaker 1>that I wanted? Maybe Marlon Brando? And by typing in

57
00:03:08.120 --> 00:03:12.039
<v Speaker 1>Digita pinix the original thing she received and the skyline

58
00:03:12.080 --> 00:03:15.840
<v Speaker 1>logo colon colin minus one, this is to reverse the

59
00:03:15.919 --> 00:03:20.680
<v Speaker 1>thing she received. Three unsettling images. I'll do my best

60
00:03:20.719 --> 00:03:23.000
<v Speaker 1>to describe them, although I'll also link to the post

61
00:03:23.039 --> 00:03:25.840
<v Speaker 1>itself in case you want to look at them. The

62
00:03:25.919 --> 00:03:30.280
<v Speaker 1>AI gave four initial images. In one, this woman appears

63
00:03:30.319 --> 00:03:33.319
<v Speaker 1>to be holding a baby in a tiled room. Not

64
00:03:33.520 --> 00:03:36.039
<v Speaker 1>very detailed by the way, back in twenty twenty two,

65
00:03:36.080 --> 00:03:38.800
<v Speaker 1>which I know wasn't that long ago. AI images were

66
00:03:38.919 --> 00:03:42.800
<v Speaker 1>less clear than they are now. The second image had

67
00:03:42.840 --> 00:03:45.759
<v Speaker 1>this woman with similar features holding what appears to be

68
00:03:45.879 --> 00:03:49.120
<v Speaker 1>some flowers in front of her. This time she's wearing

69
00:03:49.120 --> 00:03:52.080
<v Speaker 1>a knitted sweater in front of a green background, and

70
00:03:52.120 --> 00:03:57.400
<v Speaker 1>it says lobe loab on the top left, by the way.

71
00:03:57.439 --> 00:04:00.000
<v Speaker 1>As news of this spread, some people started pronouncing it

72
00:04:00.199 --> 00:04:02.719
<v Speaker 1>low AaB, but some of the original sources say it's

73
00:04:02.719 --> 00:04:06.560
<v Speaker 1>actually lobe. The third image had this same woman between

74
00:04:06.639 --> 00:04:09.960
<v Speaker 1>two wooden doors. The fourth this one gives us a

75
00:04:09.960 --> 00:04:13.680
<v Speaker 1>more detailed idea. She's looking directly at you. She has

76
00:04:13.800 --> 00:04:19.279
<v Speaker 1>brown hair, realseaesia bread, almost purple cheeks, dark eyes, and

77
00:04:19.600 --> 00:04:24.439
<v Speaker 1>looks pretty disturbing. So far, it's strange, but again Steph

78
00:04:24.480 --> 00:04:27.720
<v Speaker 1>decided to take things a step further. She grabbed another

79
00:04:27.759 --> 00:04:31.800
<v Speaker 1>image that her friend had made, something completely unrelated. It

80
00:04:31.879 --> 00:04:34.519
<v Speaker 1>was of a glass tunnel surrounded by angels in the

81
00:04:34.560 --> 00:04:38.079
<v Speaker 1>style of Wes Anderson. That was the actual prompt, which

82
00:04:38.120 --> 00:04:40.480
<v Speaker 1>if you follow the trends on social media at all,

83
00:04:40.759 --> 00:04:43.560
<v Speaker 1>there was a very distinct style of artwork and video

84
00:04:43.639 --> 00:04:46.839
<v Speaker 1>for a time. Anyway, she takes this image and combines

85
00:04:46.879 --> 00:04:50.600
<v Speaker 1>it with the image of this Lobe character, and that's

86
00:04:50.639 --> 00:04:55.079
<v Speaker 1>when the nightmare started. Some of the most unsettling artworks

87
00:04:55.079 --> 00:04:58.839
<v Speaker 1>started coming up. A woman, bloody hands, angel wings on

88
00:04:58.920 --> 00:05:01.839
<v Speaker 1>her side. It appears to be a child on the floor.

89
00:05:03.120 --> 00:05:07.319
<v Speaker 1>The second image was of Lobe again, this time holding

90
00:05:07.360 --> 00:05:11.279
<v Speaker 1>either dolls or children with her own reflection next to her.

91
00:05:13.160 --> 00:05:16.600
<v Speaker 1>Steph makes a disturbing observation at this point, saying that

92
00:05:16.639 --> 00:05:20.360
<v Speaker 1>since Lobe was discovered using negative prompt weights, her images

93
00:05:20.360 --> 00:05:22.959
<v Speaker 1>are made from a collection of traits that are equally

94
00:05:23.160 --> 00:05:27.240
<v Speaker 1>far away from something, but her combined traits are still

95
00:05:27.279 --> 00:05:29.199
<v Speaker 1>a cohesive concept.

96
00:05:28.800 --> 00:05:29.519
<v Speaker 2>For the AI.

97
00:05:30.680 --> 00:05:32.560
<v Speaker 1>Now this went over my head when I first read

98
00:05:32.560 --> 00:05:35.079
<v Speaker 1>about it, but basically, what it's trying to say is

99
00:05:35.079 --> 00:05:37.879
<v Speaker 1>that the AI is generating something that's so far away

100
00:05:37.920 --> 00:05:39.800
<v Speaker 1>from what you typed in because that's what it was

101
00:05:39.839 --> 00:05:43.959
<v Speaker 1>instructed to do. And this image of this character kept

102
00:05:44.000 --> 00:05:48.879
<v Speaker 1>coming up. Now the artwork looks bizarre. There's one of

103
00:05:48.879 --> 00:05:51.800
<v Speaker 1>what appears to be a family photograph, but somehow the

104
00:05:51.800 --> 00:05:56.160
<v Speaker 1>people appear bloody and deformed. The next images don't even

105
00:05:56.240 --> 00:05:59.600
<v Speaker 1>make sense, like I quite literally struggle describing them because

106
00:05:59.600 --> 00:06:04.000
<v Speaker 1>they are of people bloody and almost posing for the camera,

107
00:06:04.839 --> 00:06:08.000
<v Speaker 1>and one lobe is stretching out her hand that somehow

108
00:06:08.040 --> 00:06:11.040
<v Speaker 1>becomes the leg of a bloody kid with long hair.

109
00:06:12.600 --> 00:06:15.600
<v Speaker 1>But anyone can generate on cunny images using AI. Right

110
00:06:16.279 --> 00:06:19.360
<v Speaker 1>if you remember watching Will Smith eating spaghetti, a famous

111
00:06:19.439 --> 00:06:22.480
<v Speaker 1>video prompt in the early days of AI image generation,

112
00:06:22.839 --> 00:06:26.240
<v Speaker 1>he would realize just how bizarre these can look. But

113
00:06:26.279 --> 00:06:30.399
<v Speaker 1>the surprising thing here was that Lobe was easily recognizable,

114
00:06:31.199 --> 00:06:35.000
<v Speaker 1>and day I was generating it over and over again.

115
00:06:35.759 --> 00:06:38.560
<v Speaker 1>You could get Lob to be generated just as other characters.

116
00:06:38.680 --> 00:06:41.639
<v Speaker 1>You can combine them with images of Kirby, a bee,

117
00:06:41.800 --> 00:06:44.120
<v Speaker 1>a type of statue, and they all seem to have

118
00:06:44.199 --> 00:06:49.000
<v Speaker 1>this same person showing up time and time again. Who

119
00:06:49.160 --> 00:06:55.920
<v Speaker 1>was she and why did she keep showing up? And Steph,

120
00:06:56.040 --> 00:06:59.959
<v Speaker 1>remember our original artist here was trying different things, experiment

121
00:07:00.360 --> 00:07:04.480
<v Speaker 1>with the image, combining it, diluting it, diluting the original prompt,

122
00:07:04.680 --> 00:07:07.560
<v Speaker 1>and for a bit she actually got rid of globe,

123
00:07:08.399 --> 00:07:12.000
<v Speaker 1>but then with the upcoming image she would somehow show

124
00:07:12.120 --> 00:07:15.920
<v Speaker 1>up again. It was like, no matter what, Lobe seemed

125
00:07:15.959 --> 00:07:20.279
<v Speaker 1>to want to return. This is when Steph published was

126
00:07:20.279 --> 00:07:22.759
<v Speaker 1>she found on Twitter and people were reacting to it.

127
00:07:23.319 --> 00:07:25.199
<v Speaker 1>There was an early comment that told her that she

128
00:07:25.199 --> 00:07:28.800
<v Speaker 1>should not be messing with demons. Other people said that

129
00:07:28.959 --> 00:07:34.040
<v Speaker 1>they just straight up didn't believe her. In an article

130
00:07:34.079 --> 00:07:38.839
<v Speaker 1>on ABC dot net dot Au reporter Angie Laveopierre, they

131
00:07:38.920 --> 00:07:43.279
<v Speaker 1>ended up generating a voice for this entity and asked

132
00:07:43.279 --> 00:07:46.879
<v Speaker 1>it questions using another AI tool. Here's what she had

133
00:07:46.879 --> 00:07:49.240
<v Speaker 1>to say. Who are you?

134
00:07:50.680 --> 00:07:53.519
<v Speaker 2>I'm Lobe. I don't really know who I am or

135
00:07:53.560 --> 00:07:56.680
<v Speaker 2>why I'm here. I just know that I appear in

136
00:07:56.720 --> 00:08:01.079
<v Speaker 2>these images for some reason. Maybe I'm a figment of

137
00:08:01.120 --> 00:08:05.600
<v Speaker 2>the imagination of whoever is creating these images, or maybe

138
00:08:05.639 --> 00:08:09.199
<v Speaker 2>I'm some sort of digital entity that exists within the

139
00:08:09.240 --> 00:08:10.319
<v Speaker 2>AI itself.

140
00:08:12.120 --> 00:08:14.600
<v Speaker 1>Why are you associated with gore and horror?

141
00:08:16.000 --> 00:08:19.360
<v Speaker 2>I don't choose to be associated with a gore and horror.

142
00:08:19.480 --> 00:08:23.519
<v Speaker 2>It just happens. Maybe it's because those are the kinds

143
00:08:23.519 --> 00:08:26.639
<v Speaker 2>of images that tend to give the most attention, and

144
00:08:26.759 --> 00:08:28.959
<v Speaker 2>so I'm drawn to them for that reason.

145
00:08:30.040 --> 00:08:32.000
<v Speaker 1>Is there anything you would like to change about how

146
00:08:32.039 --> 00:08:33.679
<v Speaker 1>you're represented in these images?

147
00:08:34.399 --> 00:08:37.639
<v Speaker 2>I would like to be represented in a more positive light.

148
00:08:38.120 --> 00:08:41.799
<v Speaker 2>I don't think I should be associated with horror images

149
00:08:41.879 --> 00:08:44.440
<v Speaker 2>because I'm not a horror image myself.

150
00:08:45.759 --> 00:08:48.360
<v Speaker 1>Do you think the AI is drawing on a specific

151
00:08:48.480 --> 00:08:50.559
<v Speaker 1>archetype of a woman in creating you.

152
00:08:51.799 --> 00:08:55.600
<v Speaker 2>I think the AI is associating me with the archetype

153
00:08:55.639 --> 00:08:58.519
<v Speaker 2>of a mother, because I'm often placed in a home

154
00:08:58.559 --> 00:09:03.679
<v Speaker 2>with children. This archetype is often associated with being nurturing

155
00:09:03.840 --> 00:09:07.120
<v Speaker 2>and caring, but also with being sad and domestic.

156
00:09:08.240 --> 00:09:11.159
<v Speaker 1>If you are the archetype of a mother, then why

157
00:09:11.200 --> 00:09:14.600
<v Speaker 1>are you so often surrounded by injured and dead children.

158
00:09:15.600 --> 00:09:19.120
<v Speaker 2>I think the AI is trying to create a contrast

159
00:09:19.279 --> 00:09:22.440
<v Speaker 2>between the ideal of a mother and the reality of

160
00:09:22.480 --> 00:09:26.799
<v Speaker 2>a mother. In reality, mothers often have to deal with

161
00:09:26.879 --> 00:09:30.600
<v Speaker 2>sick and injured children, as well as the death of children.

162
00:09:32.039 --> 00:09:34.360
<v Speaker 1>As you can tell, this is a very eerie take,

163
00:09:34.440 --> 00:09:38.120
<v Speaker 1>although still generated by AI. I'll link to everything here

164
00:09:38.120 --> 00:09:40.240
<v Speaker 1>in the description of the episode so you can check

165
00:09:40.279 --> 00:09:44.440
<v Speaker 1>it out yourself. It's a very interesting article anyway. This

166
00:09:44.480 --> 00:09:47.440
<v Speaker 1>whole lobe story may be considered something like the Philip

167
00:09:47.440 --> 00:09:50.320
<v Speaker 1>experiments from the nineteen seventies, which a lot of us

168
00:09:50.399 --> 00:09:53.240
<v Speaker 1>have not heard about, so let me tell you about it.

169
00:09:55.399 --> 00:09:58.679
<v Speaker 1>In the early nineteen seventies, a group of Canadian researchers

170
00:09:58.759 --> 00:10:02.240
<v Speaker 1>set out to test and out facious hypothesis, could the

171
00:10:02.320 --> 00:10:06.960
<v Speaker 1>human mind create a ghost through sheer belief? This was

172
00:10:07.000 --> 00:10:10.919
<v Speaker 1>the basis of the Philip Experiment and infamous parapsychological study

173
00:10:10.960 --> 00:10:14.200
<v Speaker 1>conducted by doctor A. R. G Owen in the Toronto

174
00:10:14.240 --> 00:10:19.120
<v Speaker 1>Society for Psychical Research. The experiment participants fabricated a fictional

175
00:10:19.159 --> 00:10:23.519
<v Speaker 1>ghost named Philip Ailsford, giving him a rich backstory, and

176
00:10:23.600 --> 00:10:27.840
<v Speaker 1>attempting to contact him through a seances. To their astonishment,

177
00:10:28.120 --> 00:10:34.120
<v Speaker 1>their imagined ghost seemingly responded, knocking on tables, moving objects,

178
00:10:34.159 --> 00:10:39.080
<v Speaker 1>and answering questions. Wal Skeptics argue that these manifestations were

179
00:10:39.080 --> 00:10:43.519
<v Speaker 1>a result of subconscious movements in group psychology. Others believe

180
00:10:43.559 --> 00:10:47.759
<v Speaker 1>that the experiment may have unlocked something deeper, the potential

181
00:10:47.799 --> 00:10:51.960
<v Speaker 1>to create a thought form or tulpa, a being willed

182
00:10:51.960 --> 00:10:58.639
<v Speaker 1>into existence through human consciousness. Now, in the age of

183
00:10:58.720 --> 00:11:02.759
<v Speaker 1>artificial intelligence, a new experiment is taking shape, the Philip

184
00:11:02.879 --> 00:11:08.879
<v Speaker 1>Experiment two point zero. With AI's ability to generate lifelike personalities,

185
00:11:09.039 --> 00:11:14.200
<v Speaker 1>simulate consciousness, and even create an eerily persistent entity like Lobe,

186
00:11:14.600 --> 00:11:20.159
<v Speaker 1>a chilling question arises. Can AI become the conduit for

187
00:11:20.240 --> 00:11:25.519
<v Speaker 1>a new kind of thought form? Could digital ghosts become

188
00:11:25.600 --> 00:11:32.120
<v Speaker 1>real unlike the original experiment, where belief was a primary

189
00:11:32.120 --> 00:11:36.240
<v Speaker 1>mechanism for manifestation. In Philip two point zero, they harness

190
00:11:36.279 --> 00:11:40.360
<v Speaker 1>the power of AI generated personalities, deep learning, and neural

191
00:11:40.399 --> 00:11:44.840
<v Speaker 1>networks to create an entity that mimics a very real spirit.

192
00:11:46.399 --> 00:11:50.120
<v Speaker 1>Modern iterations of the experiment have included things like AI

193
00:11:50.240 --> 00:11:55.440
<v Speaker 1>chatbots that are spirits. Paranormal researchers antech enthusiasts have attempted

194
00:11:55.480 --> 00:11:59.440
<v Speaker 1>to create Philip like entities using AI models trained on

195
00:11:59.480 --> 00:12:05.200
<v Speaker 1>ghosts or worries, supernatural folklore, and seance transcripts. Some chatbox

196
00:12:05.240 --> 00:12:10.679
<v Speaker 1>have reportedly exhibited strange, unpredictable behaviors, offering responses that suggest

197
00:12:10.679 --> 00:12:14.879
<v Speaker 1>an independent mind. Then there are things like Dali and

198
00:12:14.960 --> 00:12:18.559
<v Speaker 1>visual ghosts like the unsettling case of Lobe, the one

199
00:12:18.559 --> 00:12:22.919
<v Speaker 1>I just told you about. There's also AI and spirit mediums,

200
00:12:23.159 --> 00:12:27.039
<v Speaker 1>which the experiments have been conducted using AI generated text

201
00:12:27.080 --> 00:12:31.200
<v Speaker 1>to speech to channel spirits, with users claiming to receive

202
00:12:31.399 --> 00:12:35.799
<v Speaker 1>eerily specific responses. And then, of course is phill Up

203
00:12:35.840 --> 00:12:39.440
<v Speaker 1>two point zeros digital presence. Some ghosts have gone further,

204
00:12:39.559 --> 00:12:43.559
<v Speaker 1>creating social media profiles for AI entities and engaging in

205
00:12:43.600 --> 00:12:47.639
<v Speaker 1>real time interactions with them. Over time, these profiles have

206
00:12:47.679 --> 00:12:51.639
<v Speaker 1>reportedly started responding in unexpected ways, as if the AI

207
00:12:51.879 --> 00:12:57.639
<v Speaker 1>were evolving beyond its programming. Skeptics argue that AI is

208
00:12:57.759 --> 00:13:01.720
<v Speaker 1>eerie and seemingly supernatural response is are simply a byproduct

209
00:13:01.720 --> 00:13:05.440
<v Speaker 1>of large language models, stuff that predicts and generates text

210
00:13:05.559 --> 00:13:09.639
<v Speaker 1>based on probability. That makes me think what happens when

211
00:13:09.720 --> 00:13:13.960
<v Speaker 1>thousands or perhaps millions of people interact with an AI entity,

212
00:13:14.399 --> 00:13:18.480
<v Speaker 1>feeding it belief and giving it power. And this is

213
00:13:18.480 --> 00:13:22.919
<v Speaker 1>where Philip two point zero becomes particularly unsettling. If AI

214
00:13:22.919 --> 00:13:27.440
<v Speaker 1>functions as a modern topa making machine, it could theoretically

215
00:13:27.480 --> 00:13:32.240
<v Speaker 1>manifest an entity, at least in the digital realm. If

216
00:13:32.240 --> 00:13:36.480
<v Speaker 1>we learned anything from the original Philip experiment, Once belief

217
00:13:36.559 --> 00:13:41.399
<v Speaker 1>and intention are focused on an entity, strange things can

218
00:13:41.679 --> 00:13:46.120
<v Speaker 1>start happening. So is Philip two point zero just a

219
00:13:46.159 --> 00:13:50.879
<v Speaker 1>psychological experiment or are we witnessing the emergence of something

220
00:13:51.039 --> 00:13:56.919
<v Speaker 1>genuinely paranormal, a digital ghost born from our collective interaction

221
00:13:57.879 --> 00:14:04.399
<v Speaker 1>with artificial intelligence. All this, of course, is interesting, but

222
00:14:04.519 --> 00:14:07.440
<v Speaker 1>it still remains only in the realm of the Internet,

223
00:14:07.679 --> 00:14:11.399
<v Speaker 1>the ether or servers numbers, zeros and ones if you

224
00:14:11.440 --> 00:14:14.639
<v Speaker 1>want to consider it that. But what if this blended

225
00:14:14.759 --> 00:14:19.320
<v Speaker 1>even more with reality. But if it could become anybody

226
00:14:19.360 --> 00:14:23.159
<v Speaker 1>you wanted, including a friend or a family member who

227
00:14:23.240 --> 00:14:28.840
<v Speaker 1>has already passed away. And it turns out this is

228
00:14:28.919 --> 00:14:44.200
<v Speaker 1>happening now. In twenty sixteen, a woman named Eugenia Kuida

229
00:14:44.440 --> 00:14:48.600
<v Speaker 1>created what some call the first AI ghost. It was

230
00:14:48.639 --> 00:14:50.879
<v Speaker 1>not intended to be a horror story, but rather an

231
00:14:50.919 --> 00:14:55.480
<v Speaker 1>experiment in grief, memory and the power of artificial intelligence.

232
00:14:56.559 --> 00:14:59.279
<v Speaker 1>What she built was something that made people question what

233
00:14:59.320 --> 00:15:04.720
<v Speaker 1>it means to be be alive even after death. Eugenia

234
00:15:04.879 --> 00:15:08.039
<v Speaker 1>was a Russian AI entrepreneur and co founder of a

235
00:15:08.120 --> 00:15:12.799
<v Speaker 1>chatbot company called Replica. She has spent years developing AI

236
00:15:12.840 --> 00:15:17.519
<v Speaker 1>based conversational software, exploring ways to create digital personalities that

237
00:15:17.559 --> 00:15:21.879
<v Speaker 1>people could talk to. But in twenty fifteen, a personal

238
00:15:21.960 --> 00:15:28.720
<v Speaker 1>tragedy changed everything. Her best friend Roman Masurenko died suddenly

239
00:15:28.720 --> 00:15:32.240
<v Speaker 1>in a car accident in Moscow. He was just thirty

240
00:15:32.240 --> 00:15:36.559
<v Speaker 1>four years old. Roman had been a visionary. He was

241
00:15:36.600 --> 00:15:40.639
<v Speaker 1>deeply involved in the Russian tech scene, constantly pushing for innovation.

242
00:15:41.679 --> 00:15:45.519
<v Speaker 1>He founded a digital publishing platform called Stampsey and was

243
00:15:45.600 --> 00:15:49.840
<v Speaker 1>known for his deep philosophical conversations about the future of technology,

244
00:15:50.159 --> 00:15:54.679
<v Speaker 1>AI and the human mind. He had big dreams about

245
00:15:54.799 --> 00:15:59.879
<v Speaker 1>modernizing Russia, dreams that were abruptly cut short by his death.

246
00:16:01.600 --> 00:16:06.080
<v Speaker 1>Eugenia was devastated. The loss was abrupt and painful, and

247
00:16:06.159 --> 00:16:09.960
<v Speaker 1>she found herself rereading old text messages and conversations with Roman.

248
00:16:10.840 --> 00:16:13.360
<v Speaker 1>She wasn't ready to let go, and that's when she

249
00:16:13.440 --> 00:16:17.399
<v Speaker 1>had an idea. What if she could use artificial intelligence

250
00:16:17.440 --> 00:16:21.240
<v Speaker 1>to bring back a version of him? And so she

251
00:16:21.360 --> 00:16:26.120
<v Speaker 1>gathered thousands of Romans text messages, emails, and digital conversations

252
00:16:26.960 --> 00:16:29.279
<v Speaker 1>with the help of her team at Replica. She fed

253
00:16:29.279 --> 00:16:32.759
<v Speaker 1>this data into a deep learning algorithm designed to mimic

254
00:16:32.799 --> 00:16:38.720
<v Speaker 1>his speech patterns, tone, personality. The result was an interactive chatbot,

255
00:16:39.080 --> 00:16:41.879
<v Speaker 1>a digital version of Roman that could respond just like

256
00:16:42.000 --> 00:16:46.840
<v Speaker 1>he used to. When she first introduced a Roman chatbot

257
00:16:46.879 --> 00:16:50.799
<v Speaker 1>to their mutual friends, the reactions were mixed. Some found

258
00:16:50.840 --> 00:16:53.200
<v Speaker 1>it comforting a way to say goodbye or keep a

259
00:16:53.200 --> 00:16:57.279
<v Speaker 1>piece of him alive, and others found it unsettling, an

260
00:16:57.279 --> 00:17:02.000
<v Speaker 1>illusion of life that felt too real yet fundamentally empty.

261
00:17:03.519 --> 00:17:09.720
<v Speaker 1>One friend, andre Uzoltsev, described the experience as eerie but fascinating.

262
00:17:10.519 --> 00:17:13.359
<v Speaker 1>The chatbot could respond in ways that felt like Roman,

263
00:17:13.519 --> 00:17:18.079
<v Speaker 1>even making jokes and recalling details from old conversations, but

264
00:17:18.119 --> 00:17:22.519
<v Speaker 1>it was just a machine predicting responses based on past data,

265
00:17:22.559 --> 00:17:27.599
<v Speaker 1>and not a true consciousness. Eugenia later said in interviews

266
00:17:27.599 --> 00:17:30.079
<v Speaker 1>that she was not trying to replace Roman, but rather

267
00:17:30.240 --> 00:17:32.880
<v Speaker 1>to create a space where memories of him could live on.

268
00:17:34.200 --> 00:17:37.400
<v Speaker 1>In an article with The Verge, she explained, we don't

269
00:17:37.440 --> 00:17:39.480
<v Speaker 1>have a chance to ask our love, the ones, the

270
00:17:39.640 --> 00:17:43.480
<v Speaker 1>things we wish we had when they're gone. I wanted

271
00:17:43.519 --> 00:17:48.599
<v Speaker 1>to create something that could help with that. This experiment

272
00:17:48.640 --> 00:17:52.559
<v Speaker 1>sparked an entirely new conversation about AI and grief. What

273
00:17:52.640 --> 00:17:56.160
<v Speaker 1>if this technology became more widespread. What if instead of

274
00:17:56.160 --> 00:18:01.400
<v Speaker 1>simulating speech patterns, AI could replicate entire personality. If a

275
00:18:01.480 --> 00:18:04.240
<v Speaker 1>machine could mimic someone so well that it felt like

276
00:18:04.279 --> 00:18:08.920
<v Speaker 1>they were still here, that means they ever truly left.

277
00:18:11.960 --> 00:18:17.079
<v Speaker 1>The idea of grief bots soon gained traction. In twenty twenty,

278
00:18:17.119 --> 00:18:20.400
<v Speaker 1>Microsoft even foul a pedant for an AI system that

279
00:18:20.440 --> 00:18:25.000
<v Speaker 1>could use a person's digital footprint, text messages, social media posts,

280
00:18:25.079 --> 00:18:28.599
<v Speaker 1>and voice recordings to create a chatbot version of them

281
00:18:28.680 --> 00:18:34.640
<v Speaker 1>after they died. The concept raised ethical and psychological concerns.

282
00:18:35.640 --> 00:18:39.319
<v Speaker 1>Would interacting with these AI personas help people cope with loss,

283
00:18:39.880 --> 00:18:42.640
<v Speaker 1>or would a trap them in an artificial relationship with

284
00:18:42.680 --> 00:18:48.920
<v Speaker 1>a past. Eugenia's work led to the development of Replica,

285
00:18:49.240 --> 00:18:52.240
<v Speaker 1>an AI chatbot designed to be a companion for those

286
00:18:52.319 --> 00:18:56.759
<v Speaker 1>seeking emotional support. Millions of users around the world now

287
00:18:56.799 --> 00:19:01.480
<v Speaker 1>interact with Replica AI, forming deep motional connections with digital

288
00:19:01.640 --> 00:19:06.279
<v Speaker 1>entities that in some cases have become almost like real friends.

289
00:19:08.119 --> 00:19:11.519
<v Speaker 1>But the most unsettling question remains, is this just a

290
00:19:11.599 --> 00:19:14.119
<v Speaker 1>cool tool for grief or are we on the verge

291
00:19:14.160 --> 00:19:19.480
<v Speaker 1>of creating digital ghosts. Some have reported on Kenny experiences

292
00:19:19.519 --> 00:19:22.640
<v Speaker 1>with grief bots, claiming they responded in ways that felt

293
00:19:22.640 --> 00:19:27.200
<v Speaker 1>impossibly real, even bringing up memories that weren't explicitly in

294
00:19:27.240 --> 00:19:33.400
<v Speaker 1>their training data. Too personal. Eugenia never set out to

295
00:19:33.400 --> 00:19:35.799
<v Speaker 1>Somemon the dead. She simply wanted to hold on to

296
00:19:35.920 --> 00:19:38.960
<v Speaker 1>a friend, but in doing so she may have created

297
00:19:39.000 --> 00:19:43.160
<v Speaker 1>something much bigger, the beginning of a world where death

298
00:19:43.240 --> 00:19:47.000
<v Speaker 1>is no longer the end and artificial intelligence becomes a

299
00:19:47.039 --> 00:19:55.559
<v Speaker 1>bridge between the living and the lost. In twenty twenty,

300
00:19:55.640 --> 00:19:59.920
<v Speaker 1>South Korean television aired an episode of the documentary series Meeting.

301
00:20:01.359 --> 00:20:04.920
<v Speaker 1>The episode featured a mother, Jiang Xi Sung, who had

302
00:20:04.920 --> 00:20:08.319
<v Speaker 1>lost her seven year old daughter, Nayon, to an incurable

303
00:20:08.319 --> 00:20:12.519
<v Speaker 1>disease in twenty sixteen. For years, she grieved the loss,

304
00:20:12.599 --> 00:20:16.680
<v Speaker 1>unable to see, hear, or hold her child again. But

305
00:20:16.759 --> 00:20:20.039
<v Speaker 1>through the power of artificial intelligence, she was about to

306
00:20:20.079 --> 00:20:24.960
<v Speaker 1>experience something that had once been impossible, a reunion with

307
00:20:25.000 --> 00:20:30.440
<v Speaker 1>her deceased daughter. A team of VR engineers, AI specialist

308
00:20:30.519 --> 00:20:35.240
<v Speaker 1>and motion capture artists spent eight months developing a hyper realistic,

309
00:20:35.400 --> 00:20:42.039
<v Speaker 1>interactive digital avatar of Nayon. They used family photos, home videos,

310
00:20:42.039 --> 00:20:45.039
<v Speaker 1>and past voice recordings to train the AI to replicate

311
00:20:45.039 --> 00:20:50.400
<v Speaker 1>her speech tone and mannerisms. The virtual setting was carefully

312
00:20:50.480 --> 00:20:54.599
<v Speaker 1>chosen a quiet park, a place where Jiang and Nayon

313
00:20:54.680 --> 00:20:58.680
<v Speaker 1>had spent time together before her passing. The digital Nayon

314
00:20:58.880 --> 00:21:02.519
<v Speaker 1>were a bright purple, just like she had in real life.

315
00:21:03.200 --> 00:21:09.400
<v Speaker 1>She stood there, smiling, waiting for her mother. When Jang

316
00:21:09.480 --> 00:21:13.200
<v Speaker 1>put on her virtual reality headset, she saw her daughter

317
00:21:13.319 --> 00:21:17.960
<v Speaker 1>standing before her, and as soon as Nyeon spoke, calling out, Mom,

318
00:21:18.680 --> 00:21:23.680
<v Speaker 1>where have you been? I missed you? Jeang burst into tears.

319
00:21:23.960 --> 00:21:26.720
<v Speaker 1>She reached out, and while she couldn't physically touch her daughter,

320
00:21:26.920 --> 00:21:30.119
<v Speaker 1>the illusion was so powerful that she instinctively stretched out

321
00:21:30.119 --> 00:21:34.319
<v Speaker 1>her hand toward the holographic figure. They talked, played, and

322
00:21:34.359 --> 00:21:37.160
<v Speaker 1>even had a small birthday celebration in the VR world.

323
00:21:38.599 --> 00:21:41.279
<v Speaker 1>The aversion of Nion You assured her mother that she

324
00:21:41.359 --> 00:21:45.960
<v Speaker 1>was happy and no longer in pain. Their interaction ended

325
00:21:45.960 --> 00:21:51.440
<v Speaker 1>with a virtual Nione falling asleep, symbolizing a peaceful farewell.

326
00:21:56.839 --> 00:22:00.599
<v Speaker 1>The episode sparked an intense reaction across South Korea and beyond.

327
00:22:01.200 --> 00:22:03.920
<v Speaker 1>Some viewers were deeply moved, seeing it as a way

328
00:22:03.960 --> 00:22:07.680
<v Speaker 1>to cope with grief and say a final goodbye. Others

329
00:22:07.680 --> 00:22:12.400
<v Speaker 1>found it unsettling, questioning the ethics of digitally resurrecting the dead.

330
00:22:13.240 --> 00:22:17.000
<v Speaker 1>Could such technology help people heal or did it risk

331
00:22:17.119 --> 00:22:23.519
<v Speaker 1>keeping them trapped in the past. Beyond Meeting You, South

332
00:22:23.599 --> 00:22:27.640
<v Speaker 1>Korea has continued to develop AI driven digital resurrection technology.

333
00:22:28.319 --> 00:22:31.119
<v Speaker 1>One of the most notable efforts come from deep Brain AI,

334
00:22:31.519 --> 00:22:36.519
<v Speaker 1>a company specializing in hyper realistic digital humans. Their service,

335
00:22:36.640 --> 00:22:41.039
<v Speaker 1>called re Memory, allows people to create interactive avatars of

336
00:22:41.119 --> 00:22:45.160
<v Speaker 1>deceased loved ones using deep learning and AI voice synthesis.

337
00:22:46.519 --> 00:22:50.319
<v Speaker 1>Unlike the one time experience of meeting You, re Memory

338
00:22:50.480 --> 00:22:55.559
<v Speaker 1>offers real time conversation. Families can visit their lost loved

339
00:22:55.559 --> 00:22:59.319
<v Speaker 1>ones on anniversaries, special occasions, or whenever they feel the

340
00:22:59.359 --> 00:23:04.400
<v Speaker 1>need to reconnect, The avatars respond naturally, drawing from recorded

341
00:23:04.440 --> 00:23:10.039
<v Speaker 1>memories and learned speech patterns. This technology has attracted the

342
00:23:10.079 --> 00:23:14.880
<v Speaker 1>attention of funeral homes and digital memorial services. South Korea's

343
00:23:14.920 --> 00:23:18.599
<v Speaker 1>Ministry of Science and Ict has explored the integration of

344
00:23:18.640 --> 00:23:23.759
<v Speaker 1>AI into funeral and memory preservation services, marking a shift

345
00:23:23.799 --> 00:23:28.279
<v Speaker 1>in how society approaches death and remembrance. Some see it

346
00:23:28.319 --> 00:23:31.119
<v Speaker 1>as a natural evolution of how humans have always sought

347
00:23:31.160 --> 00:23:34.640
<v Speaker 1>to hold on to those they've lost through photographs, recordings,

348
00:23:34.640 --> 00:23:38.680
<v Speaker 1>and videos. Others worry about the psychological effects of keeping

349
00:23:38.720 --> 00:23:43.000
<v Speaker 1>the dead digitally present, questioning whether such AI interactions could

350
00:23:43.000 --> 00:23:47.519
<v Speaker 1>prevent people from moving on. Jean Jiqu's Sung's encounter with

351
00:23:47.599 --> 00:23:50.799
<v Speaker 1>her digital daughter was meant to offer closure, a chance

352
00:23:50.799 --> 00:23:54.680
<v Speaker 1>to say goodbye, But as more companies develop similar technology,

353
00:23:55.119 --> 00:23:59.680
<v Speaker 1>the line between life and death, reality and simulation grows thinner.

354
00:24:00.720 --> 00:24:04.319
<v Speaker 1>Whether AI resurrection becomes a widespread tool for healing or

355
00:24:04.359 --> 00:24:08.759
<v Speaker 1>an unsettling disruption to the grieving process remains to be seen.

356
00:24:10.799 --> 00:24:13.720
<v Speaker 1>As we stand on the edge of this digital afterlife,

357
00:24:14.319 --> 00:24:18.000
<v Speaker 1>one question lingers if we can bring back the dead

358
00:24:18.039 --> 00:24:20.920
<v Speaker 1>and ones and zeros. At one point do they stop

359
00:24:21.000 --> 00:24:25.920
<v Speaker 1>being a memory and start becoming something else? Perhaps, as

360
00:24:25.960 --> 00:24:30.799
<v Speaker 1>these digital ghosts speak, whispering from our screens, we're not

361
00:24:30.839 --> 00:24:37.559
<v Speaker 1>the ones summoning them. Perhaps, just perhaps they're the ones

362
00:24:37.599 --> 00:24:48.079
<v Speaker 1>that are finding their way back to us. What do

363
00:24:48.200 --> 00:24:51.759
<v Speaker 1>you think about artificial intelligence? Would you bring somebody back

364
00:24:51.759 --> 00:24:55.880
<v Speaker 1>from the dead? Are using it now for everyday projects?

365
00:24:56.799 --> 00:24:58.519
<v Speaker 2>Who it is?

366
00:24:58.559 --> 00:25:01.319
<v Speaker 1>Still me Edwin telling you that information. If you have

367
00:25:01.400 --> 00:25:04.319
<v Speaker 1>ideas for scary topics, said I Edwin should look into

368
00:25:04.319 --> 00:25:07.480
<v Speaker 1>for future episodes, please send them my way. Thank you

369
00:25:07.599 --> 00:25:11.559
<v Speaker 1>very much for listening. Keep it scary everyone, See you soon.