1
00:00:00,120 --> 00:00:04,760
Speaker 1: My REGI welcome Stuff to Blow Your Mind, a production

2
00:00:04,800 --> 00:00:13,680
Speaker 1: of I Heart Radios How Stuff Works. Hey, welcome to

3
00:00:13,720 --> 00:00:15,960
Speaker 1: Stuff to Blow Your Mind. My name is Robert Lamb

4
00:00:16,280 --> 00:00:19,520
Speaker 1: and I'm Joe McCormick, and we're back with part three

5
00:00:19,680 --> 00:00:23,200
Speaker 1: of our Journey through Facial Recognition. In the first episode,

6
00:00:23,200 --> 00:00:25,640
Speaker 1: we talked a lot about the sort of the current

7
00:00:25,680 --> 00:00:29,520
Speaker 1: tech landscape of a company focusing on facial recognition, some

8
00:00:29,640 --> 00:00:32,080
Speaker 1: issues with that. In the last episode we focused primarily

9
00:00:32,080 --> 00:00:35,800
Speaker 1: on the biological domain of facial recognition, and now we're

10
00:00:35,840 --> 00:00:39,120
Speaker 1: bringing it back to technology to uh to finish up today.

11
00:00:39,880 --> 00:00:42,200
Speaker 1: So one of the things that we were looking at

12
00:00:42,600 --> 00:00:45,360
Speaker 1: coming into today's episode was an article that I thought

13
00:00:45,400 --> 00:00:48,520
Speaker 1: was really good in Wired magazine again from this month

14
00:00:48,520 --> 00:00:52,559
Speaker 1: from January of twenty by Seaun Revieve called the Secret

15
00:00:52,680 --> 00:00:56,040
Speaker 1: History of Facial Recognition, And I will say I was

16
00:00:56,160 --> 00:01:00,800
Speaker 1: surprised to find out how far back facial recognition projects go.

17
00:01:01,080 --> 00:01:05,160
Speaker 1: This this goes into the nineteen sixties. Yeah, yeah, this,

18
00:01:05,160 --> 00:01:07,759
Speaker 1: this this is a really good read. This article. I mean,

19
00:01:07,800 --> 00:01:10,480
Speaker 1: it's it's extremely well written. It almost has I would

20
00:01:10,480 --> 00:01:13,600
Speaker 1: say it has a very narrative flow to a beginning

21
00:01:13,840 --> 00:01:18,320
Speaker 1: with this scene in which Uh, an elderly researcher who

22
00:01:18,440 --> 00:01:19,959
Speaker 1: is you know at this point, I believe you can

23
00:01:20,000 --> 00:01:24,440
Speaker 1: find to a wheelchair is instructing his son to to

24
00:01:24,680 --> 00:01:28,920
Speaker 1: unlock some old, rotting files from the sixties and burn

25
00:01:29,000 --> 00:01:31,280
Speaker 1: them in front of him and the garbage can in

26
00:01:31,319 --> 00:01:34,000
Speaker 1: the garage or something. Yeah, with you know, with and

27
00:01:34,080 --> 00:01:36,319
Speaker 1: you can you can tell that there are things about

28
00:01:36,600 --> 00:01:41,319
Speaker 1: classified information, top secret or what have you on the documents.

29
00:01:41,600 --> 00:01:45,399
Speaker 1: So it's a wonder wonderful, um you know, ominous start

30
00:01:45,560 --> 00:01:48,960
Speaker 1: to this article, which of course deals predominantly with the

31
00:01:48,960 --> 00:01:52,720
Speaker 1: the origins of facial recognition and touches on other at

32
00:01:52,800 --> 00:01:59,280
Speaker 1: times inspiring and other times creepy and devastating scientific programs

33
00:01:59,400 --> 00:02:02,520
Speaker 1: that were going on or in like in full swing

34
00:02:02,720 --> 00:02:05,000
Speaker 1: during the sixties. Well. Yeah, one of the things that

35
00:02:05,080 --> 00:02:07,720
Speaker 1: really comes home in this article is that the creepiness

36
00:02:07,800 --> 00:02:11,040
Speaker 1: of facial recognition technology is not new. That's sort of

37
00:02:11,080 --> 00:02:13,120
Speaker 1: been there since the very beginning. It's not one of

38
00:02:13,160 --> 00:02:16,760
Speaker 1: these things where um, like even the author here mentions like,

39
00:02:16,800 --> 00:02:19,200
Speaker 1: you know, social media, where it seems great at first,

40
00:02:19,200 --> 00:02:21,600
Speaker 1: and it's not until it's quote in the wild that

41
00:02:21,720 --> 00:02:24,919
Speaker 1: we begin to realize, oh yeah, this is um civilization

42
00:02:25,000 --> 00:02:28,080
Speaker 1: wrecking awfulness. Uh. And not just a fun way to

43
00:02:28,120 --> 00:02:33,000
Speaker 1: share photos. No. At the time, there was a realization

44
00:02:33,200 --> 00:02:37,840
Speaker 1: that this was potentially problematic. Yeah. And so it might

45
00:02:37,880 --> 00:02:40,040
Speaker 1: not come as a big surprise, especially given the story

46
00:02:40,080 --> 00:02:43,359
Speaker 1: we mentioned about burning documents that some of the earliest

47
00:02:43,400 --> 00:02:47,720
Speaker 1: funding for facial recognition technology research clearly came from the

48
00:02:47,760 --> 00:02:50,760
Speaker 1: c i A and front companies set up to funnel

49
00:02:50,800 --> 00:02:54,600
Speaker 1: c i A money. Yeah, this was this was super interesting,

50
00:02:54,880 --> 00:02:58,840
Speaker 1: the CIA funding through these various phony companies, and and

51
00:02:59,120 --> 00:03:01,360
Speaker 1: due to the CI funding, some of the you know,

52
00:03:01,480 --> 00:03:04,760
Speaker 1: stuff was was secret. Some of the image material has

53
00:03:04,800 --> 00:03:07,040
Speaker 1: only come out, you know, due to Freedom of Information

54
00:03:07,480 --> 00:03:10,720
Speaker 1: Act filings, and and some of the work was just

55
00:03:10,800 --> 00:03:13,959
Speaker 1: never published. A lot of the work was never published. Uh.

56
00:03:14,280 --> 00:03:16,840
Speaker 1: To to drive home the creepiness though, one of the

57
00:03:16,840 --> 00:03:21,480
Speaker 1: companies involved was this company, Panoramic, which which was also

58
00:03:21,520 --> 00:03:25,480
Speaker 1: tied to other programs including UH the Project mk Ultra. Uh.

59
00:03:25,880 --> 00:03:28,959
Speaker 1: It was one of eighty organizations that worked on Project

60
00:03:29,120 --> 00:03:33,600
Speaker 1: mk Ultra in particular quote subprojects and ninety four on

61
00:03:33,639 --> 00:03:36,800
Speaker 1: the study of bacterial and fungal toxins and the remote

62
00:03:36,880 --> 00:03:42,880
Speaker 1: directional control of activities of selected species of animals Animal control. Yeah,

63
00:03:42,960 --> 00:03:44,320
Speaker 1: I was like, if I can, if I want to

64
00:03:44,320 --> 00:03:47,440
Speaker 1: sick tigers at you from the other side of the world. Yeah.

65
00:03:47,680 --> 00:03:50,120
Speaker 1: MK Ultra, just to remind everybody, was a c I

66
00:03:50,360 --> 00:03:57,040
Speaker 1: project that explored the potential for mind control using psychedelics

67
00:03:57,400 --> 00:04:00,480
Speaker 1: UH and other tactics like basically looking at ways to

68
00:04:00,600 --> 00:04:07,040
Speaker 1: take these mind expanding um UH agents and use them

69
00:04:07,040 --> 00:04:10,960
Speaker 1: to break down the human psyche and then inevitably built

70
00:04:11,040 --> 00:04:13,800
Speaker 1: something back up that it could be tightly control. Now,

71
00:04:13,800 --> 00:04:16,640
Speaker 1: the evidence today is that the mind control experiments of

72
00:04:16,760 --> 00:04:19,200
Speaker 1: m K culture didn't really work, but they were really

73
00:04:19,200 --> 00:04:22,560
Speaker 1: good at destroying the human mind. I mean because basically

74
00:04:22,720 --> 00:04:27,120
Speaker 1: the project was responsible for psychological torture. UH. Just a

75
00:04:27,279 --> 00:04:32,360
Speaker 1: horrible program in a real blight on the scientific history

76
00:04:32,480 --> 00:04:36,200
Speaker 1: of the United States. UM. You know, not the only blight,

77
00:04:36,279 --> 00:04:39,440
Speaker 1: but but I think an appalling one. They were not good.

78
00:04:39,480 --> 00:04:41,080
Speaker 1: They did not figure out a way how to read

79
00:04:41,160 --> 00:04:43,919
Speaker 1: and you know, to rebuild say an ideal sleeper agent

80
00:04:44,040 --> 00:04:47,279
Speaker 1: out of the psychological destruction that they wrought. Yeah. So

81
00:04:47,360 --> 00:04:49,880
Speaker 1: a lot of this article focuses on this one main

82
00:04:50,000 --> 00:04:53,919
Speaker 1: figure named Woody Bledso, who was a leader at this company.

83
00:04:54,160 --> 00:04:57,840
Speaker 1: A founder and leader at this company, Panoramic Research, Incorporated,

84
00:04:57,839 --> 00:05:00,960
Speaker 1: which you mentioned earlier, which got a lot of business

85
00:05:01,080 --> 00:05:04,279
Speaker 1: from c I A and c I A front organization

86
00:05:04,400 --> 00:05:08,640
Speaker 1: funding in the sixties to study things like facial recognition.

87
00:05:09,080 --> 00:05:11,640
Speaker 1: But if you put yourself back in the context to

88
00:05:11,640 --> 00:05:13,720
Speaker 1: the nineteen sixties, I think one thing that's kind of

89
00:05:13,760 --> 00:05:17,640
Speaker 1: funny is people then might not yet have realized how

90
00:05:17,680 --> 00:05:21,159
Speaker 1: difficult of a project recognizing a face would be for

91
00:05:21,240 --> 00:05:24,560
Speaker 1: a machine, because we'd have tons of sci fi goes

92
00:05:24,600 --> 00:05:28,120
Speaker 1: back decades before that, where of course robots, computers, whatever,

93
00:05:28,120 --> 00:05:31,000
Speaker 1: just recognize people easily. Yeah, And I think that's mainly

94
00:05:31,040 --> 00:05:33,200
Speaker 1: because we would just we generally would just have a

95
00:05:33,279 --> 00:05:35,880
Speaker 1: rough idea that a robot can do everything a person

96
00:05:36,080 --> 00:05:38,680
Speaker 1: can do. A robot is a mechanical person, and you

97
00:05:38,720 --> 00:05:41,120
Speaker 1: didn't have to think too hard about all the complexities

98
00:05:41,160 --> 00:05:45,159
Speaker 1: involved there. I mean, even like the best example of

99
00:05:45,240 --> 00:05:48,839
Speaker 1: nineteen sixties science fiction and really one of the twenty

100
00:05:48,920 --> 00:05:51,640
Speaker 1: and twenty one centuries best examples of science fiction two

101
00:05:51,680 --> 00:05:56,160
Speaker 1: thou one of Space Odyssey, it does reference facial recognition

102
00:05:56,200 --> 00:05:59,719
Speaker 1: capabilities for how but it doesn't I wouldn't say that

103
00:05:59,760 --> 00:06:02,080
Speaker 1: it really it goes in depth about what that means.

104
00:06:03,120 --> 00:06:06,360
Speaker 1: But but how does is able to recognize faces, and

105
00:06:06,400 --> 00:06:09,040
Speaker 1: even like can recognize faces when it is a sketch

106
00:06:09,360 --> 00:06:12,080
Speaker 1: as opposed to video feed or a photo. Yeah, and

107
00:06:12,279 --> 00:06:15,480
Speaker 1: I think if you're not well versed in the computer

108
00:06:15,560 --> 00:06:18,760
Speaker 1: technology world, it might not be immediately apparent what's so

109
00:06:18,839 --> 00:06:22,440
Speaker 1: difficult about making a computer recognize a face? But you know,

110
00:06:22,520 --> 00:06:25,240
Speaker 1: our facial recognition systems, the things going on in our

111
00:06:25,279 --> 00:06:28,400
Speaker 1: brain have amazing capabilities and their analog right. You know,

112
00:06:28,440 --> 00:06:31,360
Speaker 1: a face, face has all kinds of variables that move

113
00:06:31,400 --> 00:06:34,719
Speaker 1: around all the time. It can be extremely difficult to

114
00:06:35,120 --> 00:06:38,520
Speaker 1: reduce a face to a set of nu miracle values,

115
00:06:38,600 --> 00:06:40,479
Speaker 1: which are what you need to do in order to

116
00:06:40,560 --> 00:06:44,320
Speaker 1: have a computer recognize a face. Yeah, as as I

117
00:06:44,360 --> 00:06:47,960
Speaker 1: think we've explored in the previous two episodes, it's um yeah,

118
00:06:48,000 --> 00:06:50,640
Speaker 1: there's there's a lot going on in facial recognition. There

119
00:06:50,640 --> 00:06:52,480
Speaker 1: are a number of challenges to it, and it is

120
00:06:52,520 --> 00:06:55,880
Speaker 1: still not a perfected technology by any means. Yeah, that's true,

121
00:06:55,880 --> 00:06:58,560
Speaker 1: And so I think it's reasonable to think of what

122
00:06:58,680 --> 00:07:02,240
Speaker 1: he bled So and his colleague as legitimate AI pioneers,

123
00:07:02,279 --> 00:07:05,320
Speaker 1: even with their work in the nineteen sixties here. Oh absolutely,

124
00:07:05,400 --> 00:07:08,000
Speaker 1: but I'm Bledso was working with a number of you know,

125
00:07:08,080 --> 00:07:14,000
Speaker 1: highly talented individuals, sometimes on projects like involving atomic weaponry,

126
00:07:14,040 --> 00:07:18,679
Speaker 1: for instance, before he became more focused on AI. Another

127
00:07:18,720 --> 00:07:22,680
Speaker 1: individual that he worked on concerning facial recognition was a

128
00:07:22,720 --> 00:07:27,120
Speaker 1: Helen shan Wolf. Um Wolf was involved in the development

129
00:07:27,120 --> 00:07:30,440
Speaker 1: of Shaky, which is a robot which DARPA describes as

130
00:07:30,560 --> 00:07:33,880
Speaker 1: quote the first mobile robot with enough artificial intelligence to

131
00:07:33,960 --> 00:07:36,640
Speaker 1: navigate on its own through a set of rooms. I

132
00:07:36,680 --> 00:07:39,120
Speaker 1: think somebody at the company also worked on a robot

133
00:07:39,160 --> 00:07:43,160
Speaker 1: called Mobot, which mode lawns and a random and unattended pattern.

134
00:07:44,480 --> 00:07:46,080
Speaker 1: I'm not joking, by the way, where that is the

135
00:07:46,160 --> 00:07:49,240
Speaker 1: thing Revieve mentioned. Yeah, it just always that That makes

136
00:07:49,240 --> 00:07:52,040
Speaker 1: me think of the the old Gumby short where the

137
00:07:52,080 --> 00:07:55,120
Speaker 1: Gumby family have the robots that are doing lawn care

138
00:07:55,200 --> 00:07:58,040
Speaker 1: and home repair and they just go perserk and the

139
00:07:58,120 --> 00:08:00,720
Speaker 1: Gumby family has to as to at them down. I

140
00:08:00,720 --> 00:08:02,280
Speaker 1: don't think I know that one. Oh, it's good. There

141
00:08:02,320 --> 00:08:04,480
Speaker 1: was an MST three k riff five eight years back.

142
00:08:04,720 --> 00:08:06,960
Speaker 1: That sounds horrible. It is. It's horrifying. At the end,

143
00:08:06,960 --> 00:08:09,560
Speaker 1: there's like a robot's head on the wall. It's it's

144
00:08:10,400 --> 00:08:14,200
Speaker 1: terrifying material. But one of the things that ended up

145
00:08:14,240 --> 00:08:16,520
Speaker 1: being the case if if you're trying to think, okay,

146
00:08:16,560 --> 00:08:19,320
Speaker 1: in the sixties, how would you even begin to get

147
00:08:19,320 --> 00:08:24,000
Speaker 1: a computer to recognize a face across different images? Uh.

148
00:08:24,360 --> 00:08:26,840
Speaker 1: One of the problems is, of course, the lack of

149
00:08:26,920 --> 00:08:30,880
Speaker 1: existing digitized imagery at the time, right. I mean, we

150
00:08:30,920 --> 00:08:34,080
Speaker 1: live in a world where digital imagery is ubiquitous. It

151
00:08:34,120 --> 00:08:35,880
Speaker 1: was not at all back then. There was almost none

152
00:08:35,920 --> 00:08:38,440
Speaker 1: of it. Yeah. Yeah, Today when you read about official

153
00:08:38,520 --> 00:08:41,840
Speaker 1: recognition research that are often working off of digital databases

154
00:08:41,880 --> 00:08:46,199
Speaker 1: containing hundreds and not thousands of photographs. Uh. And at

155
00:08:46,200 --> 00:08:50,319
Speaker 1: the time, yeah, how many digital photographs were there in

156
00:08:50,360 --> 00:08:52,600
Speaker 1: the world, right, So they actually had to have some

157
00:08:52,720 --> 00:08:55,960
Speaker 1: kind of digitization process, Like they had to be able

158
00:08:56,000 --> 00:08:59,640
Speaker 1: to take a photo and turn that photo into some

159
00:08:59,800 --> 00:09:03,280
Speaker 1: no numbers that could then be interpreted by the machine

160
00:09:03,800 --> 00:09:06,760
Speaker 1: to try to recognize a person or you know, say that, yeah,

161
00:09:06,760 --> 00:09:09,000
Speaker 1: that's the same person or not. And so a lot

162
00:09:09,000 --> 00:09:12,480
Speaker 1: of these methods involved number one, recognition rules based on

163
00:09:12,640 --> 00:09:16,240
Speaker 1: explicit measures, not machine learning. They didn't have the machine

164
00:09:16,320 --> 00:09:18,840
Speaker 1: learning methods we have today back then. They would have

165
00:09:18,920 --> 00:09:23,439
Speaker 1: to have explicit rules like the distance between randomly selected

166
00:09:23,520 --> 00:09:26,600
Speaker 1: features of the face. So in this image, what's the

167
00:09:26,640 --> 00:09:29,920
Speaker 1: difference between the left eyebrow and the right ear and

168
00:09:29,960 --> 00:09:31,959
Speaker 1: the you know, left eye and the corner the right

169
00:09:32,000 --> 00:09:35,720
Speaker 1: corner of the mouth or something. And furthermore, to get

170
00:09:35,760 --> 00:09:38,319
Speaker 1: those measures, they often had to resort to what they

171
00:09:38,320 --> 00:09:42,439
Speaker 1: called a man machine approach, which was it would pair

172
00:09:42,480 --> 00:09:47,000
Speaker 1: initial measurements made by humans. Uh. It would take those measurements,

173
00:09:47,000 --> 00:09:50,480
Speaker 1: turn them into numerical values, and then train the computer

174
00:09:50,559 --> 00:09:55,760
Speaker 1: to match based on those values, which is quite laborious. UH.

175
00:09:55,760 --> 00:09:59,000
Speaker 1: And the man machine approach was apparently necessary to input

176
00:09:59,000 --> 00:10:03,120
Speaker 1: the measures based really until the seventies, until um Review

177
00:10:03,160 --> 00:10:06,520
Speaker 1: writes quote. In nineteen seventy three, a Japanese computer scientists

178
00:10:06,600 --> 00:10:11,040
Speaker 1: named Takeo Kannate made a major leap in facial recognition technology.

179
00:10:11,160 --> 00:10:14,040
Speaker 1: Using what was then a very rare commodity, a database

180
00:10:14,240 --> 00:10:19,120
Speaker 1: of eight hundred and fifty digitized photographs eight hundred and

181
00:10:19,160 --> 00:10:24,200
Speaker 1: fifty taken from the nineteen seventy World's Fair in sweet To, Japan,

182
00:10:24,880 --> 00:10:28,280
Speaker 1: Kannada developed a program that could extract facial features such

183
00:10:28,320 --> 00:10:31,920
Speaker 1: as the nose, mouth, and eyes without human input, and

184
00:10:31,960 --> 00:10:34,800
Speaker 1: so by doing that, Cannadi was able to finally eliminate

185
00:10:35,160 --> 00:10:38,480
Speaker 1: the man part of the man machine approach the human

186
00:10:38,520 --> 00:10:41,600
Speaker 1: input of the measuring and input of the values. But

187
00:10:41,760 --> 00:10:44,000
Speaker 1: throughout all this period, I mean there were still huge

188
00:10:44,160 --> 00:10:48,200
Speaker 1: problems with machine recognition of faces. Like they would sometimes

189
00:10:48,200 --> 00:10:51,719
Speaker 1: get to the point where the system developed by by

190
00:10:51,760 --> 00:10:55,760
Speaker 1: Panoramic would be it would be more efficient than humans

191
00:10:55,800 --> 00:10:59,000
Speaker 1: at matching faces under ideal conditions. So if you could

192
00:10:59,000 --> 00:11:02,320
Speaker 1: get all the faces like oriented the same way, looking

193
00:11:02,440 --> 00:11:05,400
Speaker 1: right into the camera and all that, the machine would

194
00:11:05,400 --> 00:11:08,160
Speaker 1: be better than humans trying to match photos. But if

195
00:11:08,200 --> 00:11:10,719
Speaker 1: you just pollute the imagery a little bit and make

196
00:11:10,760 --> 00:11:14,920
Speaker 1: people like turn their heads or change the lighting, etcetera. Yeah,

197
00:11:14,960 --> 00:11:17,800
Speaker 1: any kind of problems like that, suddenly the machine loses

198
00:11:17,840 --> 00:11:20,440
Speaker 1: all its advantages and the humans are better again. Right,

199
00:11:20,440 --> 00:11:22,920
Speaker 1: And some of the other problems that are still issues

200
00:11:23,000 --> 00:11:25,960
Speaker 1: today we're problems at the time, like like depending too

201
00:11:26,040 --> 00:11:31,040
Speaker 1: much on say, uh an all lighte male database, you know,

202
00:11:31,280 --> 00:11:33,080
Speaker 1: or something to that effect. You know, where you just

203
00:11:33,160 --> 00:11:36,400
Speaker 1: do not have you don't have a broad enough sample

204
00:11:36,640 --> 00:11:40,360
Speaker 1: of of human appearances to to really have a robust

205
00:11:40,720 --> 00:11:44,400
Speaker 1: facial recognition system. Right. If it's not trained on humans generally,

206
00:11:44,400 --> 00:11:47,240
Speaker 1: it's not gonna work on humans generally, right, Uh. And

207
00:11:47,320 --> 00:11:50,079
Speaker 1: so yeah, like the racial bias problems that show up

208
00:11:50,080 --> 00:11:53,960
Speaker 1: an existing facial recognition technologies today, we're basically they're right

209
00:11:54,000 --> 00:11:56,319
Speaker 1: from the beginning. But towards the end of his article,

210
00:11:56,360 --> 00:11:59,560
Speaker 1: Revieve writes, quote only in the past ten years or so,

211
00:11:59,600 --> 00:12:03,000
Speaker 1: as face show recognition started to become capable of dealing

212
00:12:03,040 --> 00:12:07,040
Speaker 1: with real world imperfection, says A. Neil K. Jane, a

213
00:12:07,080 --> 00:12:10,320
Speaker 1: computer scientist at a Michigan State University and co editor

214
00:12:10,440 --> 00:12:14,000
Speaker 1: of handbook on Face Recognition, nearly all of the obstacles

215
00:12:14,000 --> 00:12:17,200
Speaker 1: that would he encountered, in fact, have fallen away. And

216
00:12:17,240 --> 00:12:19,800
Speaker 1: one of the big ones they point to is not

217
00:12:19,880 --> 00:12:22,920
Speaker 1: just the fact that you can get some digital images now,

218
00:12:23,160 --> 00:12:25,520
Speaker 1: but you can get so many of them that you know,

219
00:12:25,679 --> 00:12:29,360
Speaker 1: it provides these vast databases for for machine learning and

220
00:12:29,679 --> 00:12:32,920
Speaker 1: data sets for training of neural networks and stuff. That's right,

221
00:12:32,960 --> 00:12:35,679
Speaker 1: you can just basically crawl, um, you know, any given

222
00:12:35,679 --> 00:12:37,960
Speaker 1: social media side. And well, we mentioned a few already.

223
00:12:37,960 --> 00:12:40,880
Speaker 1: We'll mention a few different ones as we proceed here. Well,

224
00:12:40,920 --> 00:12:43,480
Speaker 1: maybe we should pivot to talk about the ways that

225
00:12:43,559 --> 00:12:48,400
Speaker 1: facial recognition technology is being used today. Uh And there

226
00:12:48,480 --> 00:12:50,640
Speaker 1: was one thing that I was looking at, uh So

227
00:12:50,679 --> 00:12:52,200
Speaker 1: it was a piece in the New York Times. It

228
00:12:52,240 --> 00:12:55,560
Speaker 1: was an opinion piece a guy named Bruce Schneier, who

229
00:12:55,640 --> 00:12:58,480
Speaker 1: is a fellow at the Harvard Kennedy School, that was

230
00:12:58,559 --> 00:13:01,200
Speaker 1: making a point about facial recognition that I'm not sure

231
00:13:01,200 --> 00:13:05,000
Speaker 1: I fully agree with the framing of. But within making this,

232
00:13:05,320 --> 00:13:08,400
Speaker 1: within writing this article, he articulates something that I think

233
00:13:08,520 --> 00:13:11,000
Speaker 1: is important and clarifying for us to remember as we

234
00:13:11,040 --> 00:13:14,040
Speaker 1: go on. So to first acknowledge his main point, Uh,

235
00:13:14,040 --> 00:13:17,360
Speaker 1: he writes, quote, facial recognition bands are the wrong way

236
00:13:17,360 --> 00:13:22,000
Speaker 1: to fight against modern surveillance. Focusing on one particular identification

237
00:13:22,080 --> 00:13:25,880
Speaker 1: method misconstrues the nature of the surveillance society we're in

238
00:13:25,920 --> 00:13:29,840
Speaker 1: the process of building. So he's arguing facial recognition, you know,

239
00:13:29,880 --> 00:13:33,640
Speaker 1: it's just one technology among many, and banning it doesn't

240
00:13:33,640 --> 00:13:37,600
Speaker 1: necessarily stop all of these other surveillance technologies from doing

241
00:13:37,640 --> 00:13:40,680
Speaker 1: effectively the same thing. Uh. And we can talk later

242
00:13:40,720 --> 00:13:43,960
Speaker 1: about why. I'm probably not really convinced by this argument,

243
00:13:44,080 --> 00:13:47,240
Speaker 1: But Uh, the point that I think is clarifying is

244
00:13:47,280 --> 00:13:50,080
Speaker 1: that he, you know, he says what we call facial

245
00:13:50,160 --> 00:13:53,560
Speaker 1: recognition is not just one act, but it's at least

246
00:13:53,600 --> 00:13:57,880
Speaker 1: three different acts, each coming with their own challenges. Quote

247
00:13:58,040 --> 00:14:04,640
Speaker 1: in all cases. Modern mass r valance has three broad components, identification, correlation,

248
00:14:04,880 --> 00:14:09,000
Speaker 1: and discrimination. So of course identification is just the first step.

249
00:14:09,080 --> 00:14:13,840
Speaker 1: It recognizes who you are. Correlation then associates that ide

250
00:14:14,080 --> 00:14:18,760
Speaker 1: with other information about you, and then finally discrimination treats

251
00:14:18,760 --> 00:14:21,600
Speaker 1: you differently because of that I D or because of

252
00:14:21,640 --> 00:14:25,280
Speaker 1: the associated information. Now, this in and of itself, of course,

253
00:14:25,360 --> 00:14:28,040
Speaker 1: is something that that humans are. Humans are perfectly capable

254
00:14:28,040 --> 00:14:29,760
Speaker 1: of carrying out all three of these tasks without the

255
00:14:29,800 --> 00:14:32,800
Speaker 1: aigive machines. But what we're talking about here is that

256
00:14:33,040 --> 00:14:35,640
Speaker 1: it would be automated. It would be something that would

257
00:14:35,640 --> 00:14:40,200
Speaker 1: be happening by default to everybody, as opposed to something

258
00:14:40,200 --> 00:14:44,040
Speaker 1: that might take place uh and perhaps even laboriously uh

259
00:14:44,240 --> 00:14:47,760
Speaker 1: in specific scenarios such as, you know, once you've been

260
00:14:47,800 --> 00:14:51,520
Speaker 1: flagged at a security checkpoint or something like that, they

261
00:14:51,800 --> 00:14:53,440
Speaker 1: they ask you to pull out your I D and

262
00:14:53,440 --> 00:14:56,240
Speaker 1: then look you up, to correlate you with other information

263
00:14:56,360 --> 00:14:58,600
Speaker 1: and then maybe treat you differently based on whatever they

264
00:14:58,640 --> 00:15:01,320
Speaker 1: find out. That we're dealing with the scenario here where

265
00:15:01,360 --> 00:15:03,880
Speaker 1: that this would all be happening in real time and

266
00:15:04,000 --> 00:15:06,720
Speaker 1: would happen perhaps with very little delay. And I think

267
00:15:06,760 --> 00:15:10,200
Speaker 1: most importantly would be the scale and pervasiveness. It would

268
00:15:10,200 --> 00:15:14,480
Speaker 1: be happening everywhere all the time. We know from experience

269
00:15:14,480 --> 00:15:19,680
Speaker 1: how quickly new digital technology pervades all spaces, so basically,

270
00:15:20,200 --> 00:15:22,280
Speaker 1: you know the one of his arguments is that these

271
00:15:22,320 --> 00:15:26,240
Speaker 1: are three broad components that it might be easier to

272
00:15:26,320 --> 00:15:31,280
Speaker 1: regulate individually, as opposed to saying, let's not do facial recognition. Well, instead,

273
00:15:31,360 --> 00:15:33,760
Speaker 1: let's maybe we go after each of these three things.

274
00:15:34,000 --> 00:15:36,920
Speaker 1: I think that's not necessarily a bad idea to to

275
00:15:37,000 --> 00:15:40,400
Speaker 1: think about a framework for overall regulation of surveillance and

276
00:15:40,760 --> 00:15:44,360
Speaker 1: preservation of privacy. I mean, ultimately, I think I agree

277
00:15:44,440 --> 00:15:48,200
Speaker 1: that it's important to better understand and regulate the entire

278
00:15:48,320 --> 00:15:52,920
Speaker 1: process recognizing the three different components of identification, correlation, and

279
00:15:52,960 --> 00:15:56,840
Speaker 1: discrimination individually. I just don't think it makes a lot

280
00:15:56,920 --> 00:15:59,920
Speaker 1: of sense to frame this is an argument against banning

281
00:16:00,000 --> 00:16:03,080
Speaker 1: a regulating facial recognition, because to me, that sounds kind

282
00:16:03,080 --> 00:16:06,400
Speaker 1: of like saying, well, international treaties banning the use of

283
00:16:06,440 --> 00:16:09,440
Speaker 1: smallpox virus as a weapon of war miss the point

284
00:16:09,480 --> 00:16:11,960
Speaker 1: that we need to rethink our entire concept of war

285
00:16:12,080 --> 00:16:15,280
Speaker 1: and defense, and we need to regulate international conflict in

286
00:16:15,320 --> 00:16:19,280
Speaker 1: a more comprehensive way. Um. I mean, like, yes, that

287
00:16:19,360 --> 00:16:21,800
Speaker 1: would be true. But if you don't know when and

288
00:16:21,840 --> 00:16:24,920
Speaker 1: whether that full comprehensive thing is going to be accomplished,

289
00:16:25,120 --> 00:16:27,560
Speaker 1: and you do currently have a consensus to ban the

290
00:16:27,640 --> 00:16:31,440
Speaker 1: use of germ warfare, why wouldn't you do it? Absolutely? Um,

291
00:16:31,520 --> 00:16:33,640
Speaker 1: here's another quote from that that article though that I

292
00:16:33,720 --> 00:16:36,240
Speaker 1: that I was really taken by quote. The point is

293
00:16:36,280 --> 00:16:38,920
Speaker 1: that it doesn't matter which technology is used to identify

294
00:16:39,040 --> 00:16:42,640
Speaker 1: people that they're currently is. No comprehensive database of heartbeats

295
00:16:42,720 --> 00:16:45,600
Speaker 1: or gates doesn't make the technologies that gather them any

296
00:16:45,680 --> 00:16:48,080
Speaker 1: less effective, And most of the time it doesn't matter

297
00:16:48,120 --> 00:16:51,040
Speaker 1: if identification isn't tied to a real name. What's important

298
00:16:51,280 --> 00:16:54,960
Speaker 1: is that we can be consistently identified over time. And

299
00:16:55,000 --> 00:16:57,360
Speaker 1: then on on top of that though, you know, once

300
00:16:57,400 --> 00:17:00,560
Speaker 1: your real name is then attached to that information, there's

301
00:17:00,600 --> 00:17:03,160
Speaker 1: no turning back, right. The system is building a picture

302
00:17:03,200 --> 00:17:06,639
Speaker 1: of you, move by move, purchased by purchase, search by search,

303
00:17:06,920 --> 00:17:10,800
Speaker 1: over the course of years, if not decades. Yeah. Uh.

304
00:17:10,840 --> 00:17:13,119
Speaker 1: And of course, because it's the Internet, what kind of

305
00:17:13,200 --> 00:17:15,919
Speaker 1: system are we talking about? I mean, I would argue

306
00:17:15,960 --> 00:17:18,320
Speaker 1: that it is a dystopia A little bit less like

307
00:17:18,480 --> 00:17:22,200
Speaker 1: nineteen eighty four and more like Terry Gilliam's Brazil, where

308
00:17:22,320 --> 00:17:24,560
Speaker 1: you know, there's no one person in charge that you

309
00:17:24,600 --> 00:17:28,199
Speaker 1: can rebel against. No one person seems to understand how

310
00:17:28,240 --> 00:17:30,720
Speaker 1: it all works or call all the shots. It is

311
00:17:30,800 --> 00:17:35,199
Speaker 1: a terrible will emanating from the void that is expressed

312
00:17:35,200 --> 00:17:39,320
Speaker 1: through millions of antlike functionaries, each doing it's bidding without

313
00:17:39,400 --> 00:17:43,240
Speaker 1: being bidden. I mean, isn't the oppression scarier when there's

314
00:17:43,280 --> 00:17:45,159
Speaker 1: no boss in charge of it all that you can

315
00:17:45,200 --> 00:17:48,719
Speaker 1: rebel against? Absolutely, And especially when it's uh when it's

316
00:17:48,760 --> 00:17:51,399
Speaker 1: a case like facial recognition where a lot of the

317
00:17:52,440 --> 00:17:54,320
Speaker 1: a lot of the the advancements that have been made,

318
00:17:54,320 --> 00:17:56,960
Speaker 1: and then more specifically a lot of the cases where

319
00:17:57,320 --> 00:18:02,080
Speaker 1: it has been or is being um rolled out. You know,

320
00:18:02,119 --> 00:18:04,720
Speaker 1: it's often it's it's not as a situation where people

321
00:18:04,720 --> 00:18:07,240
Speaker 1: are getting to vote on it or even necessarily having

322
00:18:07,320 --> 00:18:11,600
Speaker 1: any kind of really broad discussion about it before it

323
00:18:11,600 --> 00:18:14,080
Speaker 1: takes place. It just sort of happens and then here

324
00:18:14,119 --> 00:18:16,280
Speaker 1: we are. You know, So not only is there no

325
00:18:16,440 --> 00:18:20,479
Speaker 1: key individual in place that you can blame and rebell against,

326
00:18:20,520 --> 00:18:22,600
Speaker 1: there's like they're not even necessarily a set point in

327
00:18:22,680 --> 00:18:24,520
Speaker 1: time where you could say, we have to go back

328
00:18:24,560 --> 00:18:27,359
Speaker 1: and change this, you know, getting a time machine and

329
00:18:27,400 --> 00:18:30,840
Speaker 1: try and prevent facial recognition from being rolled out. And

330
00:18:30,840 --> 00:18:33,320
Speaker 1: where do you go, Who do you try and stop?

331
00:18:33,359 --> 00:18:36,040
Speaker 1: Who do you try and speak to except the whole world?

332
00:18:36,640 --> 00:18:38,080
Speaker 1: You know? Another thing I would say is to go

333
00:18:38,119 --> 00:18:41,560
Speaker 1: back to my germ warfare analogy. I think it's possible

334
00:18:41,720 --> 00:18:44,960
Speaker 1: that that he's not quite right, that the different methods

335
00:18:45,000 --> 00:18:48,280
Speaker 1: of ideaing you and tracking you are indistinguishable from each other.

336
00:18:48,600 --> 00:18:52,160
Speaker 1: I think it's possible that facial recognition is an especially

337
00:18:52,320 --> 00:18:57,159
Speaker 1: insidious and corrupting type of automated identification compared to some

338
00:18:57,200 --> 00:18:59,640
Speaker 1: other methods, like you know, ideas of credit card numbers

339
00:18:59,680 --> 00:19:03,720
Speaker 1: or math addresses on your phone. Because our lives are

340
00:19:03,800 --> 00:19:07,400
Speaker 1: built around faces, our social existence and our cultures are

341
00:19:07,440 --> 00:19:11,120
Speaker 1: built around interfacing with faces. I mean, it seems like

342
00:19:11,200 --> 00:19:15,560
Speaker 1: somehow a more dangerous kind of well of information to

343
00:19:15,800 --> 00:19:19,080
Speaker 1: poison in the culture than to say, like, well, you know,

344
00:19:19,160 --> 00:19:21,040
Speaker 1: your phone knows where you are. I mean, you could

345
00:19:21,119 --> 00:19:25,280
Speaker 1: potentially smash your phone with a hammer. Yeah, and again,

346
00:19:25,720 --> 00:19:27,879
Speaker 1: I think I've mentioned this already. So many of the

347
00:19:27,880 --> 00:19:31,480
Speaker 1: things involving facial recognition on the surface, it doesn't seem threatening.

348
00:19:31,480 --> 00:19:33,520
Speaker 1: It seems like, well, we wanted machines to do what

349
00:19:33,520 --> 00:19:35,840
Speaker 1: people could do, so we we've created a way for

350
00:19:35,920 --> 00:19:38,199
Speaker 1: them to do it. That's what we always wanted. How

351
00:19:38,280 --> 00:19:42,120
Speaker 1: is that that terrifying? And I think part of it

352
00:19:42,160 --> 00:19:44,399
Speaker 1: is haven't you always wanted a person who is a

353
00:19:44,440 --> 00:19:47,399
Speaker 1: boss who can watch you every minute of the day. Well,

354
00:19:47,440 --> 00:19:49,679
Speaker 1: I mean, yes, when you get do specific examples of it.

355
00:19:49,680 --> 00:19:51,399
Speaker 1: But but I think I think one of the keys

356
00:19:51,520 --> 00:19:54,560
Speaker 1: is that when we're talking about automating this sort of thing,

357
00:19:54,800 --> 00:19:57,600
Speaker 1: and when we're talking about the machine version of it,

358
00:19:57,800 --> 00:19:59,879
Speaker 1: we're talking about a version of it that exists on

359
00:20:00,119 --> 00:20:03,600
Speaker 1: scale and on a scope that is beyond human. And

360
00:20:03,680 --> 00:20:05,920
Speaker 1: that's that's the thing. It's like, we're not talking about

361
00:20:05,920 --> 00:20:10,040
Speaker 1: a human capability anymore. We're talking about an inhuman capability

362
00:20:10,480 --> 00:20:12,000
Speaker 1: the same way that we've you know, we've talked about

363
00:20:12,000 --> 00:20:14,440
Speaker 1: consciousness before and we sort of post the question, well,

364
00:20:15,160 --> 00:20:18,520
Speaker 1: is part of consciousness? The limitation of what we can

365
00:20:18,560 --> 00:20:21,760
Speaker 1: focus our attention on? Is there is part of the

366
00:20:21,840 --> 00:20:25,159
Speaker 1: key to being human? Uh? Is part of it? The

367
00:20:25,200 --> 00:20:27,520
Speaker 1: limitations on what our brains can do, on what our

368
00:20:27,560 --> 00:20:31,200
Speaker 1: senses can do. And we're talking about models that are

369
00:20:31,280 --> 00:20:34,000
Speaker 1: not limited by those senses or those uh, you know,

370
00:20:34,040 --> 00:20:37,200
Speaker 1: those computational resources as beautifully put. And I think you're

371
00:20:37,240 --> 00:20:40,400
Speaker 1: exactly right. I mean, if you could be conscious of everything,

372
00:20:40,440 --> 00:20:42,800
Speaker 1: I'm not sure you would be conscious anymore. I'm not

373
00:20:42,840 --> 00:20:45,359
Speaker 1: sure you'd be a person. Yeah, yeah, I don't think.

374
00:20:45,880 --> 00:20:50,000
Speaker 1: You know, this is something that that far more specialized

375
00:20:50,280 --> 00:20:53,560
Speaker 1: people than myself probably have better arguments about. But yeah,

376
00:20:53,560 --> 00:20:56,760
Speaker 1: to to what extents extent would a super consciousness not

377
00:20:56,880 --> 00:20:59,600
Speaker 1: be a consciousness? It would be beyond what we think

378
00:20:59,640 --> 00:21:02,720
Speaker 1: of a consciousness. Okay, we need to take a quick break,

379
00:21:02,760 --> 00:21:05,080
Speaker 1: but when we come back, we will discuss how facial

380
00:21:05,080 --> 00:21:08,520
Speaker 1: recognition technology is already being used in several countries around

381
00:21:08,520 --> 00:21:14,680
Speaker 1: the world. Alright, we're back, all right, So let's let's

382
00:21:14,680 --> 00:21:16,360
Speaker 1: look at just this is just going to be kind

383
00:21:16,359 --> 00:21:19,040
Speaker 1: of a snapshot at a few different a few different

384
00:21:19,040 --> 00:21:22,560
Speaker 1: stories covering facial recognition as it is being rolled out

385
00:21:22,680 --> 00:21:25,400
Speaker 1: right now as of this recording in at the tail

386
00:21:25,520 --> 00:21:28,960
Speaker 1: end of January. So the first a couple of sources

387
00:21:28,960 --> 00:21:32,040
Speaker 1: of looking at uh. There's a an opinion piece by

388
00:21:32,240 --> 00:21:37,400
Speaker 1: Frederic califooner UH in The Guardian titled Facial Recognition cameras

389
00:21:37,400 --> 00:21:41,119
Speaker 1: will put us all in an identity parade. And this

390
00:21:41,160 --> 00:21:44,359
Speaker 1: piece also refers in places to reporting from the same

391
00:21:44,400 --> 00:21:48,000
Speaker 1: month and publication by Vicram Dodd, a police and crime

392
00:21:48,040 --> 00:21:53,080
Speaker 1: correspondent for The Guardian. So basically, London's Metro Police announced

393
00:21:53,119 --> 00:21:57,119
Speaker 1: their intention to launch live facial recognition cameras in London,

394
00:21:57,480 --> 00:22:00,239
Speaker 1: a city already known for a mass surveillance survey ence

395
00:22:00,320 --> 00:22:03,960
Speaker 1: via their CCTV system. It's a move condemned by civil

396
00:22:04,000 --> 00:22:08,439
Speaker 1: liberty groups as as a quote breathtaking assault on human rights. Uh.

397
00:22:08,520 --> 00:22:11,880
Speaker 1: The police, however, claimed that eight of people surveyed back

398
00:22:12,000 --> 00:22:13,880
Speaker 1: the move and they would only be used to catch

399
00:22:13,960 --> 00:22:17,119
Speaker 1: violent criminals and find missing people. Uh. They also stressed

400
00:22:17,160 --> 00:22:19,480
Speaker 1: that it will be properly posted and it will be

401
00:22:19,600 --> 00:22:22,840
Speaker 1: rolled out with clarity and transparency. Uh and and only

402
00:22:23,200 --> 00:22:27,720
Speaker 1: you know, following outreach in affected communities. Now, you know,

403
00:22:27,720 --> 00:22:29,480
Speaker 1: we've kind of talked about about this already. If you

404
00:22:29,640 --> 00:22:32,840
Speaker 1: if you stand by the notion that, Okay, I trust

405
00:22:32,880 --> 00:22:35,439
Speaker 1: the government as as it exists now, and I trust

406
00:22:35,480 --> 00:22:38,480
Speaker 1: whatever model it will take in the future. I trust

407
00:22:38,480 --> 00:22:42,560
Speaker 1: the keepers of this information and my personal information and uh,

408
00:22:42,560 --> 00:22:44,919
Speaker 1: and I don't have anything to hide anyway. Then okay,

409
00:22:44,960 --> 00:22:47,080
Speaker 1: I guess you can easily get on board with something

410
00:22:47,119 --> 00:22:49,760
Speaker 1: like that, but a lot of folks have have issues

411
00:22:49,880 --> 00:22:53,520
Speaker 1: with this rollout in particular, as well as with facial

412
00:22:53,560 --> 00:22:58,120
Speaker 1: recognition technology in general. So for starters, this comes as

413
00:22:58,200 --> 00:23:01,240
Speaker 1: the European Union is consider during a temporary ban on

414
00:23:01,280 --> 00:23:04,399
Speaker 1: facial recognition, and of course this is also occurring as

415
00:23:04,440 --> 00:23:08,480
Speaker 1: the UK continues to extract itself from the EU. Also,

416
00:23:08,760 --> 00:23:14,000
Speaker 1: the only independent review of the Metro facial recognition public trials,

417
00:23:14,040 --> 00:23:17,679
Speaker 1: by one Pete to Fussy of Essex University, found that

418
00:23:17,720 --> 00:23:21,199
Speaker 1: it was only verifiably accurate in just nineteen percent of

419
00:23:21,280 --> 00:23:24,360
Speaker 1: the cases that opposed to the uh TOO. I think

420
00:23:24,359 --> 00:23:27,399
Speaker 1: it was something like a seventy success rate that the

421
00:23:27,440 --> 00:23:31,840
Speaker 1: Metro police were claiming. Yeah, I mean we saw in

422
00:23:31,880 --> 00:23:34,240
Speaker 1: the first episode we talked about how the company clear

423
00:23:34,320 --> 00:23:37,919
Speaker 1: View AI claimed that they found correct matches up to

424
00:23:37,960 --> 00:23:41,600
Speaker 1: seventy percent of the time. But we've definitely read reviews

425
00:23:41,680 --> 00:23:45,200
Speaker 1: that that placed correct to numbers way way lower. Right,

426
00:23:45,240 --> 00:23:47,080
Speaker 1: And and the other thing is there's no there's no

427
00:23:47,160 --> 00:23:49,800
Speaker 1: opt out here. This is not even though they're talking

428
00:23:49,840 --> 00:23:53,760
Speaker 1: about targeted uses of it, the technology is not targeted.

429
00:23:53,840 --> 00:23:57,040
Speaker 1: All faces are scanned that are at all scannable, and

430
00:23:57,040 --> 00:24:01,280
Speaker 1: therefore everyone is in a virtual criminal line up whenever

431
00:24:01,400 --> 00:24:04,480
Speaker 1: they're in the sights of the camera. Um. This is

432
00:24:04,920 --> 00:24:08,399
Speaker 1: a quote from that opinion piece quote. Given that the

433
00:24:08,400 --> 00:24:12,479
Speaker 1: system inevitably processes the biometric data of everyone live, facial

434
00:24:12,520 --> 00:24:16,960
Speaker 1: recognition has the potential to fundamentally change the power relationship

435
00:24:17,000 --> 00:24:20,520
Speaker 1: between people and the police, and even alter the very

436
00:24:20,600 --> 00:24:24,439
Speaker 1: meaning of public space. So the you know, the standard

437
00:24:24,440 --> 00:24:29,040
Speaker 1: criticisms come up as well, including misidentification, but not only misidentification,

438
00:24:29,080 --> 00:24:33,480
Speaker 1: but also automated misidentification, which the author charges shifts the

439
00:24:33,520 --> 00:24:38,359
Speaker 1: burden of proof onto the falsely recognized individual because there's

440
00:24:38,359 --> 00:24:41,520
Speaker 1: no human accuser, right, it's just the machine said you

441
00:24:41,560 --> 00:24:43,360
Speaker 1: did this, and it's it's up to you to prove

442
00:24:43,400 --> 00:24:45,520
Speaker 1: that you are not that face, that that face is

443
00:24:45,560 --> 00:24:49,399
Speaker 1: not yours. Again and and and this is and I

444
00:24:49,400 --> 00:24:53,160
Speaker 1: want to stress that this is problematic even if that

445
00:24:53,600 --> 00:24:56,080
Speaker 1: even if the success rate were not so low, like

446
00:24:56,359 --> 00:24:59,200
Speaker 1: because obviously the technology is improving and that is not

447
00:24:59,400 --> 00:25:01,480
Speaker 1: the ants or that is part of the problem. No,

448
00:25:01,560 --> 00:25:04,560
Speaker 1: I'm more scared when the success rate gets verifiably higher,

449
00:25:04,600 --> 00:25:07,160
Speaker 1: because then if the trust and it just keeps going

450
00:25:07,240 --> 00:25:10,800
Speaker 1: up and up. The fewer cases where people still get misidentified,

451
00:25:10,840 --> 00:25:13,439
Speaker 1: which always is going to continue to happen to some degree,

452
00:25:13,840 --> 00:25:16,719
Speaker 1: are are going to get more and more like, you know, terrified.

453
00:25:16,720 --> 00:25:19,080
Speaker 1: They're going to be more and more under the knife, right,

454
00:25:19,119 --> 00:25:21,159
Speaker 1: I mean you end up with it again, it changes

455
00:25:21,200 --> 00:25:23,720
Speaker 1: what is a public space. Well, a public space then

456
00:25:23,880 --> 00:25:25,760
Speaker 1: I can't go to the park without being in a

457
00:25:25,840 --> 00:25:28,959
Speaker 1: virtual uh like a lot of crime line up, Like

458
00:25:29,040 --> 00:25:32,399
Speaker 1: I am always going to be profiled as to be

459
00:25:32,600 --> 00:25:37,879
Speaker 1: as a potential criminal. So yeah, it's a it's a

460
00:25:37,880 --> 00:25:40,320
Speaker 1: good piece, worth a worth reading. So Caul Throner, you know,

461
00:25:40,400 --> 00:25:42,000
Speaker 1: sums this up by saying, you know, look, the stakes

462
00:25:42,040 --> 00:25:45,159
Speaker 1: are very high with this kind of invasive technology. And

463
00:25:45,160 --> 00:25:48,720
Speaker 1: while these rollouts are not without their critics. Obviously we've

464
00:25:48,760 --> 00:25:52,760
Speaker 1: been talking about the various criticisms of it, but they

465
00:25:52,840 --> 00:25:56,199
Speaker 1: argue that we still haven't really seen them subjected to

466
00:25:56,320 --> 00:26:00,320
Speaker 1: true public debate. The technology is happening faster than the

467
00:26:00,320 --> 00:26:03,160
Speaker 1: awareness of it is exactly. Yeah. Yeah, And I would

468
00:26:03,200 --> 00:26:06,080
Speaker 1: point out another thing about it, which is just that, Okay,

469
00:26:06,119 --> 00:26:08,240
Speaker 1: if you suddenly put a pause on this kind of

470
00:26:08,240 --> 00:26:11,119
Speaker 1: technology and say we can't use it, and then you

471
00:26:11,200 --> 00:26:14,280
Speaker 1: go under a review, when you review and review and review,

472
00:26:14,320 --> 00:26:17,199
Speaker 1: and you're sure in the end, okay, it definitely the

473
00:26:17,240 --> 00:26:21,119
Speaker 1: benefits outweigh the harms. Then you could still release it

474
00:26:21,160 --> 00:26:24,000
Speaker 1: in the future. But I think you can't really go

475
00:26:24,160 --> 00:26:26,800
Speaker 1: the other way, right, once you live in the facial

476
00:26:26,840 --> 00:26:31,400
Speaker 1: recognition world, that animals not getting back in the cage, right, Yeah,

477
00:26:31,440 --> 00:26:35,400
Speaker 1: once you've sort of eroded the norms of of privacy,

478
00:26:35,440 --> 00:26:37,080
Speaker 1: like how do you go back? You know, you and

479
00:26:37,280 --> 00:26:40,919
Speaker 1: quickly enter an age where people who don't want to

480
00:26:40,960 --> 00:26:44,240
Speaker 1: be a part of the database are going to be

481
00:26:44,280 --> 00:26:47,040
Speaker 1: seen as the outsiders in the fringe cases and the

482
00:26:47,359 --> 00:26:49,520
Speaker 1: you know, the people who live out in the hut

483
00:26:49,520 --> 00:26:52,040
Speaker 1: in the woods. And perhaps that would be also because

484
00:26:52,040 --> 00:26:57,080
Speaker 1: they would be effectively ostracized from so many of these systems. Yeah. Uh,

485
00:26:57,160 --> 00:26:59,800
Speaker 1: And again we're mainly talking about like state uses or

486
00:27:00,080 --> 00:27:02,680
Speaker 1: lease uses or whatever, but there's a whole other world

487
00:27:02,720 --> 00:27:07,159
Speaker 1: to consider of just like, uh, distributed public use of

488
00:27:07,200 --> 00:27:12,480
Speaker 1: facial recognition systems, which could be hugely socially important, could

489
00:27:12,600 --> 00:27:16,280
Speaker 1: lead to new types of social ostracism and stuff. Yeah,

490
00:27:16,440 --> 00:27:19,040
Speaker 1: or situations where oh well, I'm I'm opting out of

491
00:27:19,040 --> 00:27:22,359
Speaker 1: facial recognition technology. Oops. Looks like it's gonna make it

492
00:27:22,400 --> 00:27:25,199
Speaker 1: really difficult for me to pay my bills because I

493
00:27:25,240 --> 00:27:28,439
Speaker 1: need my face for identification. On that, we'll get to

494
00:27:28,480 --> 00:27:31,240
Speaker 1: some examples of of that currently in use here in

495
00:27:31,280 --> 00:27:34,720
Speaker 1: a bit. Or how about I mean this might sound petty,

496
00:27:34,800 --> 00:27:38,800
Speaker 1: but imagine a world where, uh, anytime somebody invites you

497
00:27:38,880 --> 00:27:41,560
Speaker 1: to do something and you say no, I can't, they

498
00:27:41,560 --> 00:27:45,120
Speaker 1: can also very likely find out wherever you were at

499
00:27:45,119 --> 00:27:48,880
Speaker 1: the time you couldn't do the thing they invited you to. Die. Yeah,

500
00:27:48,920 --> 00:27:50,800
Speaker 1: I was just having a conversation with my wife about

501
00:27:50,800 --> 00:27:54,919
Speaker 1: smartphone technology, about how I had recently been to a

502
00:27:55,040 --> 00:27:58,680
Speaker 1: concert venue and uh, and I'm not going to name

503
00:27:58,680 --> 00:28:02,440
Speaker 1: the organization that was handling a concert, but you had

504
00:28:02,480 --> 00:28:04,920
Speaker 1: to have a mobile entry ticket, which, as far as

505
00:28:04,920 --> 00:28:06,480
Speaker 1: I know, that meant it had to be I could

506
00:28:06,480 --> 00:28:08,440
Speaker 1: be wrong on this, but my understanding was you had

507
00:28:08,480 --> 00:28:11,919
Speaker 1: to have like a smartphone version of your ticket to

508
00:28:11,920 --> 00:28:14,360
Speaker 1: get in, which I just kind of blew me agg

509
00:28:14,400 --> 00:28:15,879
Speaker 1: And I'll what if you don't have a smartphone? Not

510
00:28:15,960 --> 00:28:18,720
Speaker 1: everyone has one? Does this mean like only smartphone people

511
00:28:18,720 --> 00:28:20,720
Speaker 1: are getting into this thing? And my wife had a

512
00:28:20,760 --> 00:28:22,800
Speaker 1: similar situation with a parking deck where you had to

513
00:28:22,840 --> 00:28:26,040
Speaker 1: have an app on your phone to get and and

514
00:28:26,080 --> 00:28:29,600
Speaker 1: so you can easily imagine this kind of scenario extrapolated

515
00:28:29,880 --> 00:28:32,679
Speaker 1: to to facial recognition, like, oh, you're not opting in

516
00:28:32,680 --> 00:28:35,119
Speaker 1: on facial recognition, Well, I'm sorry. You can't come to

517
00:28:35,160 --> 00:28:37,720
Speaker 1: this this this concert because you have to have a

518
00:28:37,720 --> 00:28:39,960
Speaker 1: face scandy get in and your face will be scanned

519
00:28:40,000 --> 00:28:42,400
Speaker 1: of course while you're in the venue by the security systems.

520
00:28:42,920 --> 00:28:44,920
Speaker 1: But the UK, of course is by far not the

521
00:28:44,960 --> 00:28:47,760
Speaker 1: only place that's already trying to roll out some kind

522
00:28:47,760 --> 00:28:50,560
Speaker 1: of public form of facial recognition technology. I've read about

523
00:28:50,640 --> 00:28:53,880
Speaker 1: uses in China. I've read about uses in Russia. Yeah,

524
00:28:53,960 --> 00:28:57,080
Speaker 1: even in Russia, though, civil rights activists are criticizing facial

525
00:28:57,080 --> 00:29:00,320
Speaker 1: recognition technology is a threat to privacy and human rights,

526
00:29:00,840 --> 00:29:05,160
Speaker 1: specifically for the technology's ability to say, identify individuals at protests,

527
00:29:05,200 --> 00:29:07,920
Speaker 1: store them a database, and then track them. Right, you

528
00:29:07,920 --> 00:29:10,000
Speaker 1: could easily be put on the you know, the government's

529
00:29:10,080 --> 00:29:15,440
Speaker 1: undesirables list. Here's a quote from Natalia's Virgina, Amnesty International

530
00:29:15,520 --> 00:29:20,680
Speaker 1: Russia's director. She's quoted in the January BBC article Russia's

531
00:29:20,760 --> 00:29:24,959
Speaker 1: use of facial recognition challenged in court. Quote, facial recognition

532
00:29:24,960 --> 00:29:28,880
Speaker 1: technology is by nature deeply intrusive, as it enables the

533
00:29:29,200 --> 00:29:33,600
Speaker 1: widespread and bulk monitoring, collection, storage, and analysis of sensitive

534
00:29:33,640 --> 00:29:38,520
Speaker 1: personal data without individualized reasonable suspicion. Again, everybody's in the

535
00:29:38,520 --> 00:29:41,000
Speaker 1: criminal lineup. Yeah, you don't have to do anything in

536
00:29:41,040 --> 00:29:44,880
Speaker 1: particular to arouse suspicion of yourself just by being in public,

537
00:29:45,360 --> 00:29:49,400
Speaker 1: that's enough. According to the article that PBC article, Moscow

538
00:29:49,480 --> 00:29:53,360
Speaker 1: has about a hundred and sixty thousand CCTV cameras in

539
00:29:53,400 --> 00:29:56,840
Speaker 1: operation in the city and this month month they expanded

540
00:29:56,840 --> 00:30:00,400
Speaker 1: the number using facial recognition with no x a nation

541
00:30:00,440 --> 00:30:04,800
Speaker 1: of how privacy and human rights would be insured or

542
00:30:05,000 --> 00:30:07,160
Speaker 1: I mean it seems if they would being insured, because

543
00:30:07,160 --> 00:30:09,840
Speaker 1: it seems like the basic system would would be an

544
00:30:09,840 --> 00:30:13,760
Speaker 1: athema to that. The BBC article also pointed out that

545
00:30:13,800 --> 00:30:17,520
Speaker 1: the Moscow depart Department of Information Technology has reportedly signed

546
00:30:17,520 --> 00:30:20,800
Speaker 1: a deal with the Russian firm intech Lab to provide

547
00:30:20,840 --> 00:30:23,560
Speaker 1: the need to pride the needed technology, and this firm

548
00:30:23,600 --> 00:30:27,280
Speaker 1: had previously rolled out the the Fine Face app, which

549
00:30:27,440 --> 00:30:30,880
Speaker 1: used data from a website that is often referred to

550
00:30:30,920 --> 00:30:33,840
Speaker 1: as Russia's Facebook. It is not an actual Facebook website,

551
00:30:33,840 --> 00:30:36,880
Speaker 1: but is often held up as the equivalent. Yeah, there

552
00:30:36,880 --> 00:30:39,880
Speaker 1: are a lot of different sort of other countries facebooks,

553
00:30:40,680 --> 00:30:43,640
Speaker 1: and then there's China to consider. So China has also

554
00:30:43,800 --> 00:30:47,000
Speaker 1: rolled out facial recognition, and there are a number of

555
00:30:47,040 --> 00:30:50,400
Speaker 1: different uses and implicatement implementations that you can find just

556
00:30:50,440 --> 00:30:52,360
Speaker 1: all over the place. But you can find it in

557
00:30:52,440 --> 00:30:57,920
Speaker 1: train stations, airports, stores, hotels, gated communities, and more. I

558
00:30:58,000 --> 00:31:00,640
Speaker 1: think I had seen allegations that it had been used

559
00:31:00,800 --> 00:31:04,760
Speaker 1: in um in attacks on the weaker communities. Basically, yes,

560
00:31:05,080 --> 00:31:08,120
Speaker 1: that's a big one, the tracking of the alleged tracking

561
00:31:08,120 --> 00:31:13,000
Speaker 1: of ethnic minorities in China using these systems. But you

562
00:31:13,080 --> 00:31:15,640
Speaker 1: also you also find it another place, like it's used

563
00:31:15,680 --> 00:31:20,760
Speaker 1: for things that are almost comical but also troubling by

564
00:31:20,800 --> 00:31:24,520
Speaker 1: just how tedious and small steaks they seem, such as

565
00:31:25,320 --> 00:31:27,600
Speaker 1: you know, using it to catch toilet tissue thiefs at

566
00:31:27,680 --> 00:31:31,200
Speaker 1: bathroom um as. Some of what I was reading about

567
00:31:31,200 --> 00:31:33,640
Speaker 1: it came from a New York Times article by any

568
00:31:33,800 --> 00:31:36,400
Speaker 1: chin Uh that's about q I in it. In case

569
00:31:36,440 --> 00:31:38,360
Speaker 1: you want to look around, I'm not gonna name I'm

570
00:31:38,400 --> 00:31:40,400
Speaker 1: not gonna mention the title of the article because it

571
00:31:40,440 --> 00:31:42,200
Speaker 1: will spoil the fun of what I'm about to share.

572
00:31:42,960 --> 00:31:46,400
Speaker 1: But anyway, she in this she described China as a

573
00:31:46,480 --> 00:31:51,040
Speaker 1: quote a country accustomed to surveillance, which I think is

574
00:31:51,080 --> 00:31:53,360
Speaker 1: a is an interesting way of looking at because I

575
00:31:53,400 --> 00:31:55,920
Speaker 1: wonder if it is if China's case in some way

576
00:31:55,960 --> 00:31:58,800
Speaker 1: provides a model of where some of these other countries

577
00:31:58,840 --> 00:32:01,320
Speaker 1: we've mentioned could be head it very soon, Like it

578
00:32:01,480 --> 00:32:06,080
Speaker 1: is just given a a culture that is in very

579
00:32:06,120 --> 00:32:12,760
Speaker 1: broadly um predisposed to to being okay with surveillance. Uh,

580
00:32:13,040 --> 00:32:15,200
Speaker 1: it might provide a future look at where where we're

581
00:32:15,200 --> 00:32:18,000
Speaker 1: going in these other countries. Uh. For for example, you

582
00:32:18,040 --> 00:32:20,200
Speaker 1: see quite a bit of facial recognition used to do

583
00:32:20,280 --> 00:32:23,400
Speaker 1: stuff like open your phone or make a payment, something

584
00:32:23,440 --> 00:32:26,080
Speaker 1: that the South China Morning Post points out has been

585
00:32:26,120 --> 00:32:29,760
Speaker 1: disrupted by the current coronavirus outbreak. So in the wake

586
00:32:29,800 --> 00:32:33,040
Speaker 1: of this current global health threat which has impacted China

587
00:32:33,120 --> 00:32:36,680
Speaker 1: the most thus far, lots of people are wearing surgical masks,

588
00:32:37,320 --> 00:32:39,760
Speaker 1: and they were already in many cases wearing these masks

589
00:32:39,760 --> 00:32:43,719
Speaker 1: and high numbers due to pollution concerns. But the software

590
00:32:43,800 --> 00:32:47,000
Speaker 1: that is used in many of these instances is apparently

591
00:32:47,000 --> 00:32:50,600
Speaker 1: struggling to deal with such little facial information, requiring other

592
00:32:50,680 --> 00:32:55,040
Speaker 1: biometric data uh, in order to still I D and individual.

593
00:32:55,960 --> 00:32:59,320
Speaker 1: So so Chen's article goes into this about like people

594
00:32:59,320 --> 00:33:02,800
Speaker 1: basically have to make a choice between uh, being able

595
00:33:02,840 --> 00:33:06,880
Speaker 1: to easily make payments at stores using facial recognition or

596
00:33:07,080 --> 00:33:11,240
Speaker 1: feeling like they are adequately protected against a dangerous illness.

597
00:33:11,320 --> 00:33:13,520
Speaker 1: This is a great example of something that I was

598
00:33:13,520 --> 00:33:15,960
Speaker 1: searching for an example of later in the episode. So

599
00:33:16,040 --> 00:33:18,720
Speaker 1: we'll have to remember this later. Now I mentioned something

600
00:33:18,960 --> 00:33:21,320
Speaker 1: kind of fun but also still scary, and that it

601
00:33:21,440 --> 00:33:24,840
Speaker 1: seems small steaks and uh. And that is the central

602
00:33:25,000 --> 00:33:28,200
Speaker 1: point of Chen's recent article, and that's the use of

603
00:33:28,240 --> 00:33:32,880
Speaker 1: facial recognition technology to root out quote uncivilized behavior in

604
00:33:33,160 --> 00:33:37,120
Speaker 1: in an Hua Province in eastern China. So using facial

605
00:33:37,120 --> 00:33:41,160
Speaker 1: recognition data, she writes, the urban management the Urban Management

606
00:33:41,200 --> 00:33:46,680
Speaker 1: Department of SUSO published surveillance photos of individuals engaged in

607
00:33:47,160 --> 00:33:51,360
Speaker 1: said uncivilized behavior with partial exposure of their names. So

608
00:33:51,400 --> 00:33:54,880
Speaker 1: like a public shaming or public outing, kind of like saying,

609
00:33:54,920 --> 00:33:57,520
Speaker 1: look at these people, her her name is, you know,

610
00:33:57,560 --> 00:34:00,560
Speaker 1: Amy or whatever. Uh, here they are in gauged in

611
00:34:00,920 --> 00:34:03,480
Speaker 1: uncivilized behavior. They should not do this. What was the

612
00:34:03,520 --> 00:34:06,200
Speaker 1: uncivilized behavior? I know you you would think Okay, is

613
00:34:06,200 --> 00:34:11,320
Speaker 1: it defacing you know, a public place, stealing the toilet

614
00:34:11,320 --> 00:34:14,239
Speaker 1: paper even is low stakes? Is that same? No, it's

615
00:34:14,239 --> 00:34:17,120
Speaker 1: even lower stakes. It's the wearing of full length pajamas

616
00:34:17,160 --> 00:34:21,239
Speaker 1: in public. What yeah, So, I mean I've seen people

617
00:34:21,280 --> 00:34:25,239
Speaker 1: wear pajamas in public. I am currently wearing clothes as

618
00:34:25,239 --> 00:34:27,839
Speaker 1: I record this. I'm wearing clothes that I also sleep in.

619
00:34:28,320 --> 00:34:30,759
Speaker 1: So I think the last time you came to my

620
00:34:30,800 --> 00:34:32,640
Speaker 1: house to let me borrow a book, I answered the

621
00:34:32,680 --> 00:34:35,160
Speaker 1: door in pajamas and didn't realize until you were there.

622
00:34:35,480 --> 00:34:38,000
Speaker 1: Because I'm sorry, man, No, no, because you don't have

623
00:34:38,080 --> 00:34:41,200
Speaker 1: to apologize, because this is the universal truth. Chen also

624
00:34:42,120 --> 00:34:45,879
Speaker 1: describes it thus in her article, pajamas are comfortable. They're

625
00:34:46,000 --> 00:34:49,840
Speaker 1: very comfortable, and public pajama wearing is common in China,

626
00:34:49,920 --> 00:34:53,759
Speaker 1: especially among older women and especially in cold months and regions.

627
00:34:54,520 --> 00:34:56,680
Speaker 1: And to add another level to the injustice is the

628
00:34:56,719 --> 00:34:59,360
Speaker 1: sort of thing that is praised as fashionable when celebrities

629
00:34:59,400 --> 00:35:01,799
Speaker 1: do it, but to this sort of backlash when it's

630
00:35:01,840 --> 00:35:04,640
Speaker 1: common people. And it's been a point of contention um

631
00:35:05,360 --> 00:35:07,680
Speaker 1: for for a little while here, with government officials trying

632
00:35:07,719 --> 00:35:10,640
Speaker 1: to ban it in some places, but the people pushing

633
00:35:10,680 --> 00:35:12,400
Speaker 1: back and say no, you know this, this is a

634
00:35:12,440 --> 00:35:15,040
Speaker 1: bridge too far. I want to wear my comfy pajamas

635
00:35:15,040 --> 00:35:16,799
Speaker 1: and if I want to wear them outside of my home,

636
00:35:16,840 --> 00:35:19,879
Speaker 1: I would do so. It's not uncivilized to do this.

637
00:35:20,320 --> 00:35:23,080
Speaker 1: And in this case, it's a rare it's a rare

638
00:35:23,080 --> 00:35:25,960
Speaker 1: example of of people in China pushing back against facial

639
00:35:26,000 --> 00:35:30,600
Speaker 1: recognition technology in a place again where the technology has

640
00:35:30,600 --> 00:35:35,320
Speaker 1: already been highly established and is is recognized and appreciated

641
00:35:35,680 --> 00:35:38,279
Speaker 1: for the things that it makes easier in life, such

642
00:35:38,320 --> 00:35:41,799
Speaker 1: as making payments. So when you're purchasing something at the store, yeah,

643
00:35:41,840 --> 00:35:45,160
Speaker 1: I mean the process there is so clear suddenly, how

644
00:35:45,239 --> 00:35:48,319
Speaker 1: like you are lured in with convenience, you know, you're

645
00:35:48,400 --> 00:35:52,160
Speaker 1: lured in with immediate concrete benefits, and then there are

646
00:35:52,200 --> 00:35:57,000
Speaker 1: these consequences that just come later. Right, Yeah, And again

647
00:35:57,000 --> 00:35:59,879
Speaker 1: it shows also it should shows you what happens when

648
00:36:01,160 --> 00:36:03,839
Speaker 1: the government or the police, you know, change their mind

649
00:36:03,840 --> 00:36:06,720
Speaker 1: about how specific they need to be with the laws

650
00:36:06,760 --> 00:36:09,440
Speaker 1: that they readily enforce. You know. It's it's like, it's

651
00:36:09,480 --> 00:36:11,960
Speaker 1: it's similar case to you know that we often encounter

652
00:36:12,000 --> 00:36:14,759
Speaker 1: with say traffic cameras, like how are we supposed to

653
00:36:14,800 --> 00:36:17,440
Speaker 1: feel about like, yes, we don't want people just you know,

654
00:36:17,600 --> 00:36:19,919
Speaker 1: going through the traffic running every red light. There needs

655
00:36:19,920 --> 00:36:22,759
Speaker 1: to be order, uh, And I guess you know, there

656
00:36:22,760 --> 00:36:24,839
Speaker 1: needs to be there needs to be this idea that, yes,

657
00:36:24,880 --> 00:36:27,000
Speaker 1: if you break these rules, you will you know, you

658
00:36:27,040 --> 00:36:29,520
Speaker 1: could be pulled over, you get a ticket, there could

659
00:36:29,520 --> 00:36:31,480
Speaker 1: be some sort of a punishment in place. But when

660
00:36:31,480 --> 00:36:35,359
Speaker 1: you have this situation where any any transgression, uh, you know,

661
00:36:35,440 --> 00:36:37,840
Speaker 1: for you know, for even the smallest thing can be

662
00:36:37,920 --> 00:36:41,600
Speaker 1: met with an automated fine. Uh. You know, that gets

663
00:36:41,600 --> 00:36:45,439
Speaker 1: into an area that feels more like dystopia than order. Yeah,

664
00:36:45,480 --> 00:36:47,279
Speaker 1: it certainly can. I mean again, it's one of those

665
00:36:47,280 --> 00:36:49,560
Speaker 1: things where it starts to get harder to argue with,

666
00:36:49,600 --> 00:36:52,640
Speaker 1: Like how do you argue with the machine? The camera

667
00:36:52,719 --> 00:36:55,080
Speaker 1: said you were speeding and you're like, I wasn't speeding.

668
00:36:55,400 --> 00:36:57,759
Speaker 1: What do you do? Yeah? So anyway, these I think

669
00:36:57,800 --> 00:37:00,480
Speaker 1: these these examples, I think they just they just helped

670
00:37:00,480 --> 00:37:03,160
Speaker 1: provide a more a more nuanced idea of like what's

671
00:37:03,160 --> 00:37:06,160
Speaker 1: going on in the world right now with facial recognition, uh,

672
00:37:06,200 --> 00:37:09,880
Speaker 1: and what the many pain points are and and and

673
00:37:09,920 --> 00:37:12,919
Speaker 1: what people are where people are fighting back against him. Yeah.

674
00:37:12,960 --> 00:37:15,840
Speaker 1: So I think now that we better understand how the

675
00:37:15,880 --> 00:37:18,480
Speaker 1: technology is being deployed. Of course, in the first episode

676
00:37:18,840 --> 00:37:21,040
Speaker 1: we talked about how it's being deployed even in the

677
00:37:21,120 --> 00:37:24,640
Speaker 1: United States. The question is how should we react, Like,

678
00:37:24,680 --> 00:37:27,480
Speaker 1: what can we do? Uh? And I think maybe it

679
00:37:27,480 --> 00:37:31,520
Speaker 1: would be best to first talk about individual countermeasures and

680
00:37:31,520 --> 00:37:34,600
Speaker 1: then come back to broader action after that. Alright, so

681
00:37:34,680 --> 00:37:37,880
Speaker 1: individual countermeasures so something that an individual person can do

682
00:37:38,080 --> 00:37:42,479
Speaker 1: in the face of facial recognition technology. Yes. So. A

683
00:37:42,520 --> 00:37:45,920
Speaker 1: bunch of the existing knowledge about how to confuse and

684
00:37:45,960 --> 00:37:49,480
Speaker 1: confound facial recognition tech was summarized in a really good

685
00:37:49,560 --> 00:37:52,160
Speaker 1: article I was reading and wired by at least Thomas

686
00:37:52,239 --> 00:37:55,800
Speaker 1: from February of nineteen called how to Hack Your Face

687
00:37:55,880 --> 00:37:59,880
Speaker 1: and Dodge The Rise of Facial Recognition Tech, And unfortunately,

688
00:38:00,080 --> 00:38:02,319
Speaker 1: Thomas points out that the best way to foil a

689
00:38:02,360 --> 00:38:06,320
Speaker 1: facial recognition system is to know what method of facial

690
00:38:06,360 --> 00:38:10,520
Speaker 1: recognition is being used. Thomas quotes a privacy advocate named

691
00:38:10,560 --> 00:38:13,239
Speaker 1: Lily Ryan, who says, quote, you really need to know

692
00:38:13,360 --> 00:38:16,600
Speaker 1: what's under the hood to know what's most likely to work,

693
00:38:16,920 --> 00:38:19,160
Speaker 1: And it can be very hard for the average person

694
00:38:19,200 --> 00:38:22,000
Speaker 1: to know what kind of facial recognition is being used

695
00:38:22,040 --> 00:38:25,120
Speaker 1: on them at any particular time. So it's worth pausing

696
00:38:25,160 --> 00:38:28,920
Speaker 1: to look at different kinds of rebellion against the recognizing machine.

697
00:38:29,560 --> 00:38:31,480
Speaker 1: I think the first one is the simpler one. That

698
00:38:31,680 --> 00:38:35,759
Speaker 1: the first would be an anonymity defense, some method of

699
00:38:35,800 --> 00:38:40,760
Speaker 1: making your actual identity unrecognizable and presenting as an unknown,

700
00:38:40,880 --> 00:38:45,600
Speaker 1: unscannable person. This would essentially trying to become faceless, yeah, exactly.

701
00:38:46,040 --> 00:38:48,680
Speaker 1: The second path would go beyond that into what are

702
00:38:48,960 --> 00:38:53,400
Speaker 1: called presentation attacks in some of the literature. For example,

703
00:38:53,480 --> 00:38:56,960
Speaker 1: there's an article that's linked by Thomas by Raga, Vendra

704
00:38:57,080 --> 00:39:02,240
Speaker 1: Rama Chandra and Kristoff Bush called Presentation Attack Detection Methods

705
00:39:02,280 --> 00:39:07,160
Speaker 1: for Face Recognition Systems. A comprehensive survey published in in

706
00:39:07,280 --> 00:39:10,120
Speaker 1: a c M Computing Surveys and so this is a

707
00:39:10,160 --> 00:39:13,840
Speaker 1: survey of known presentation attacks, also known as direct attacks

708
00:39:13,960 --> 00:39:17,440
Speaker 1: or spoof attacks the author's right quote. The goal of

709
00:39:17,440 --> 00:39:21,440
Speaker 1: a presentation attack is to subvert the face recognition system

710
00:39:21,480 --> 00:39:26,040
Speaker 1: by presenting a facial biometric artifact. So if you go

711
00:39:26,200 --> 00:39:28,759
Speaker 1: in front of a facial recognition scanner and you hold

712
00:39:28,840 --> 00:39:31,520
Speaker 1: up a picture of Nicolas Cage, or you wear a

713
00:39:31,640 --> 00:39:35,840
Speaker 1: Richard Nixon mask or something, you are conducting a presentation attack.

714
00:39:36,160 --> 00:39:39,400
Speaker 1: You're not just trying not to be recognized as yourself,

715
00:39:39,680 --> 00:39:43,960
Speaker 1: but actively trying to be recognized as someone else. Common

716
00:39:43,960 --> 00:39:47,320
Speaker 1: methods here would include like presenting a photo of someone

717
00:39:47,600 --> 00:39:50,239
Speaker 1: to a scanner that has actually worked in a lot

718
00:39:50,280 --> 00:39:53,840
Speaker 1: of cases, uh, playing a video of the target face,

719
00:39:54,000 --> 00:39:57,040
Speaker 1: or wearing a three D mask of somebody else's head.

720
00:39:57,840 --> 00:40:00,360
Speaker 1: All of these methods have had some success us, but

721
00:40:00,440 --> 00:40:05,040
Speaker 1: the researchers here describe presentation attack detection algorithms or p

722
00:40:05,200 --> 00:40:09,959
Speaker 1: A D algorithms that are countermeasures against the countermeasures. Now,

723
00:40:10,360 --> 00:40:12,680
Speaker 1: a question you might be wondering is like, well, why

724
00:40:12,760 --> 00:40:14,960
Speaker 1: would you want to present as someone else instead of

725
00:40:15,000 --> 00:40:17,879
Speaker 1: just being anonymous? Well, I mean there there are all

726
00:40:17,880 --> 00:40:19,880
Speaker 1: sorts of reasons for that. I mean, there's certainly nefarious

727
00:40:19,880 --> 00:40:21,839
Speaker 1: reasons for that. If you get into a situation where

728
00:40:21,920 --> 00:40:25,680
Speaker 1: saying facial recognition is required to interrogated community, well, then

729
00:40:25,719 --> 00:40:28,520
Speaker 1: if you wanted to break into sedigated community, it would, uh,

730
00:40:28,840 --> 00:40:31,520
Speaker 1: it would behoove you to have another person's face to wear.

731
00:40:31,600 --> 00:40:34,160
Speaker 1: Perhaps you know, print it out exactly right, That could

732
00:40:34,160 --> 00:40:36,480
Speaker 1: be the direct reason. Maybe you want access to a

733
00:40:36,520 --> 00:40:41,240
Speaker 1: location or a device, and access is granted to specific

734
00:40:41,280 --> 00:40:45,279
Speaker 1: people based on facial recognition, So you use an authorized

735
00:40:45,320 --> 00:40:48,960
Speaker 1: person's face in order to get in. But I can

736
00:40:49,000 --> 00:40:54,120
Speaker 1: also see another idea, which is perhaps rampant presentation attacks

737
00:40:54,760 --> 00:40:58,560
Speaker 1: could be an effective method for fighting the facial recognition

738
00:40:58,600 --> 00:41:01,640
Speaker 1: reign of terror. Beca is it would not deny these

739
00:41:01,760 --> 00:41:04,759
Speaker 1: data collecting systems the data they want, not just do that.

740
00:41:05,080 --> 00:41:07,960
Speaker 1: It would go further and come up the databases with

741
00:41:08,040 --> 00:41:12,560
Speaker 1: lots of confusing, incorrect information, which might in fact make

742
00:41:12,600 --> 00:41:17,279
Speaker 1: them less useful overall. Yeah. I mean another application here

743
00:41:17,320 --> 00:41:20,920
Speaker 1: that is awful to think about is the use of

744
00:41:21,920 --> 00:41:24,320
Speaker 1: you We're talking about sort of the automated guilt machine

745
00:41:24,520 --> 00:41:28,359
Speaker 1: that could exist with facial recognition technology. Say you had

746
00:41:28,400 --> 00:41:31,279
Speaker 1: it in for somebody you know you're mad at, you know,

747
00:41:31,840 --> 00:41:34,799
Speaker 1: a coworker or a schoolmate. Then you go and you

748
00:41:34,880 --> 00:41:39,000
Speaker 1: do something illegal with with a mask of that person's face.

749
00:41:39,400 --> 00:41:41,600
Speaker 1: You know, not enough to say send them away forever,

750
00:41:41,680 --> 00:41:45,399
Speaker 1: but enough to say, uh, you know, to to cause

751
00:41:45,440 --> 00:41:47,000
Speaker 1: them a lot of grief in the short term, in

752
00:41:47,040 --> 00:41:49,040
Speaker 1: the very least. I mean, you don't know how well

753
00:41:49,120 --> 00:41:51,359
Speaker 1: their their defense would be. I mean, maybe it would

754
00:41:51,400 --> 00:41:53,359
Speaker 1: send them away forever. I mean, I don't know how

755
00:41:53,440 --> 00:41:55,840
Speaker 1: much faith the criminal justice system is going to end

756
00:41:55,920 --> 00:41:58,720
Speaker 1: up putting in the verdicts of these machines. I wouldn't

757
00:41:58,760 --> 00:42:01,440
Speaker 1: be surprised if it's too much. We've seen that before.

758
00:42:01,480 --> 00:42:05,120
Speaker 1: We've certainly seen models of that with some of the

759
00:42:05,200 --> 00:42:09,480
Speaker 1: past episodes where where we've discussed a forensic science. Yeah, exactly, Yeah,

760
00:42:09,640 --> 00:42:11,600
Speaker 1: there are a lot of methods of forensic science that

761
00:42:11,680 --> 00:42:17,200
Speaker 1: have been vastly overestimated in their in their confidence. Now, finally,

762
00:42:17,239 --> 00:42:19,560
Speaker 1: another reason I was thinking it might be useful to

763
00:42:19,719 --> 00:42:22,960
Speaker 1: present as another human instead of just trying to make

764
00:42:23,000 --> 00:42:26,680
Speaker 1: yourself anonymous is it's not hard at all to imagine

765
00:42:26,719 --> 00:42:31,680
Speaker 1: scenarios were being unscannable or being anonymous, will itself be

766
00:42:31,760 --> 00:42:36,759
Speaker 1: a problem, will restrict your rights, make you a target, etcetera. Like,

767
00:42:36,840 --> 00:42:41,359
Speaker 1: the anonymity could attract attention rather than discouraging it. So

768
00:42:41,440 --> 00:42:45,000
Speaker 1: the alternative would be to appear as a real scannable person,

769
00:42:45,120 --> 00:42:48,880
Speaker 1: but not yourself. Right, Wearing a faceless mask in public

770
00:42:49,200 --> 00:42:53,319
Speaker 1: is gonna draw more attention than looking like someone else. Yes,

771
00:42:53,520 --> 00:42:55,640
Speaker 1: even even if it wasn't like that great a mask

772
00:42:55,760 --> 00:42:58,120
Speaker 1: you know, uh it would, it would still it was

773
00:42:58,160 --> 00:43:01,640
Speaker 1: still potentially draw less attention. Now back to Thomas's article,

774
00:43:01,680 --> 00:43:04,000
Speaker 1: Thomas writes, that. Of course, you know, the simplest method

775
00:43:04,000 --> 00:43:07,719
Speaker 1: of fighting facial recognition is what would normally be called occlusion,

776
00:43:07,960 --> 00:43:11,239
Speaker 1: hiding all or part of your face. But again, this

777
00:43:11,320 --> 00:43:14,200
Speaker 1: is more difficult and more complicated than it sounds. So

778
00:43:14,560 --> 00:43:16,279
Speaker 1: let's say you want to walk around in public with

779
00:43:16,280 --> 00:43:19,319
Speaker 1: your face completely hidden behind a cloth or a zip

780
00:43:19,400 --> 00:43:21,799
Speaker 1: up hood or something. You know that there are there

781
00:43:21,840 --> 00:43:26,120
Speaker 1: are actually people selling basically backwards hoodies, you know, front

782
00:43:26,200 --> 00:43:29,520
Speaker 1: zip hoodies that cover your entire head. First of all,

783
00:43:29,680 --> 00:43:32,000
Speaker 1: is that legal in a lot of places? No, in

784
00:43:32,040 --> 00:43:34,560
Speaker 1: a lot of places in for example, Europe and Canada.

785
00:43:34,600 --> 00:43:37,520
Speaker 1: In the US, it's illegal to cover your entire face

786
00:43:37,560 --> 00:43:40,600
Speaker 1: in public. But even if these laws were changed or

787
00:43:40,640 --> 00:43:42,759
Speaker 1: you're in a place where that's not illegal, is this

788
00:43:42,880 --> 00:43:46,040
Speaker 1: socially feasible. We'll come back to that in a minute.

789
00:43:46,440 --> 00:43:49,560
Speaker 1: But okay, let's say you decided it is not practical

790
00:43:49,719 --> 00:43:53,080
Speaker 1: to completely occlude your entire face. What if you just

791
00:43:53,160 --> 00:43:57,480
Speaker 1: cover part of your face? Unfortunately, Thomas writes that a

792
00:43:57,480 --> 00:44:00,640
Speaker 1: lot of facial recognition software is good enough now to

793
00:44:00,760 --> 00:44:04,040
Speaker 1: make partial covering of the face ineffective as a defense.

794
00:44:04,560 --> 00:44:08,000
Speaker 1: Uh quote. For example, a bala clava, which leaves the

795
00:44:08,000 --> 00:44:11,600
Speaker 1: most important facial features exposed the eyes, the mouth, the

796
00:44:11,680 --> 00:44:14,520
Speaker 1: nose may not actually do much to prevent a person

797
00:44:14,560 --> 00:44:17,920
Speaker 1: from being identified. Researchers have found that by using a

798
00:44:18,000 --> 00:44:21,759
Speaker 1: deep learning framework trained on fourteen key facial points, they

799
00:44:21,800 --> 00:44:25,640
Speaker 1: were able to accurately identify partially occluded faces most of

800
00:44:25,719 --> 00:44:30,640
Speaker 1: the time. This includes wearing glasses, scarves, hats, or fake beards. Yeah.

801
00:44:30,680 --> 00:44:32,520
Speaker 1: I mean when you're getting down to things like like

802
00:44:32,600 --> 00:44:35,319
Speaker 1: the measurement between your eyes, you know, stuff like that.

803
00:44:35,400 --> 00:44:37,759
Speaker 1: I mean, it's it's probably going to be visible. And

804
00:44:37,840 --> 00:44:42,520
Speaker 1: most of these facial inclusion methods and it's it's not

805
00:44:42,640 --> 00:44:45,239
Speaker 1: something you can easily mess with, you know, I mean

806
00:44:46,080 --> 00:44:50,560
Speaker 1: short of like massive facial injury. Uh, I can't think

807
00:44:50,560 --> 00:44:53,440
Speaker 1: of anything much that's going to alter that measurement. Yeah,

808
00:44:53,480 --> 00:44:56,040
Speaker 1: and they're apparently multiple measures like that of a face.

809
00:44:56,120 --> 00:44:58,040
Speaker 1: You know, as long as you want to have your

810
00:44:58,080 --> 00:45:01,319
Speaker 1: eyes and your mouth and stuff exposed, is there are

811
00:45:01,360 --> 00:45:03,920
Speaker 1: probably going to be systems, especially in the near future,

812
00:45:04,360 --> 00:45:07,960
Speaker 1: that will be fairly accurate identifying you. Anyway. Now, another thing,

813
00:45:08,000 --> 00:45:10,480
Speaker 1: as I mentioned a minute ago, there's some evidence that

814
00:45:10,600 --> 00:45:14,000
Speaker 1: three D printed masks based on other people's faces can

815
00:45:14,040 --> 00:45:18,080
Speaker 1: be pretty effective, but they might not remain effective for long. Remember,

816
00:45:18,080 --> 00:45:21,400
Speaker 1: of course, we've got these pa D algorithms, the presentation

817
00:45:21,440 --> 00:45:25,759
Speaker 1: attack detections, and they're developing quickly. And also, I mean,

818
00:45:25,880 --> 00:45:28,719
Speaker 1: is that anything close to practical for regular people? Like

819
00:45:28,800 --> 00:45:30,840
Speaker 1: it seems more like an option that might be available

820
00:45:30,880 --> 00:45:34,080
Speaker 1: to a professional spy, but not just as somebody trying

821
00:45:34,120 --> 00:45:36,880
Speaker 1: to live their life. Now, there's another interesting method that

822
00:45:37,120 --> 00:45:40,640
Speaker 1: at least Thomas mentions in her article, which is confusing

823
00:45:40,680 --> 00:45:43,440
Speaker 1: the computer into believing it is not looking at a

824
00:45:43,480 --> 00:45:47,880
Speaker 1: face at all, attacking not the facial recognition stage, but

825
00:45:47,960 --> 00:45:51,480
Speaker 1: the facial detection stage. I hadn't so much thought about that,

826
00:45:51,520 --> 00:45:54,200
Speaker 1: and I think that's a really good point. One solution

827
00:45:54,239 --> 00:45:57,440
Speaker 1: along these lines is widely known as c V dazzle,

828
00:45:57,520 --> 00:46:00,960
Speaker 1: which stands for Computer Vision Dazzle dazzle kind of in

829
00:46:01,000 --> 00:46:03,800
Speaker 1: the same sense it's used in like a military context.

830
00:46:03,800 --> 00:46:05,760
Speaker 1: You would put dazzle on the side of a ship

831
00:46:06,239 --> 00:46:09,800
Speaker 1: in a context of warfare, which are like different lines

832
00:46:09,880 --> 00:46:12,319
Speaker 1: going in different ways that apparently make it harder to

833
00:46:12,400 --> 00:46:15,200
Speaker 1: look at the ship and determine its speed and it's bearing,

834
00:46:15,680 --> 00:46:20,280
Speaker 1: like like some of the more like lightning bolty camouflage designs.

835
00:46:20,280 --> 00:46:23,640
Speaker 1: One sees not not necessarily be dazzled, which is the

836
00:46:23,640 --> 00:46:25,640
Speaker 1: first thing that came to my mind, like just bedazzle

837
00:46:25,719 --> 00:46:29,080
Speaker 1: your face. Well that it's funny because that does kind

838
00:46:29,120 --> 00:46:31,320
Speaker 1: of come in but but yeah, I think it's based

839
00:46:31,360 --> 00:46:34,080
Speaker 1: on the idea of a visual dazzle, and so what

840
00:46:34,160 --> 00:46:37,279
Speaker 1: it involves for people is altering the appearance of your

841
00:46:37,320 --> 00:46:41,799
Speaker 1: head to stop facial detection algorithms from flagging it as

842
00:46:41,800 --> 00:46:44,200
Speaker 1: a face. And the most common ways to do this

843
00:46:44,239 --> 00:46:48,160
Speaker 1: are with makeup, with hair styling, and coloring with facial

844
00:46:48,200 --> 00:46:51,319
Speaker 1: accessories like hair clips and stick on rhinestone. So there

845
00:46:51,400 --> 00:46:54,719
Speaker 1: is some bedazzling going on. Robert I included a few

846
00:46:54,719 --> 00:46:57,279
Speaker 1: examples here for you to look at. These are from

847
00:46:57,280 --> 00:47:01,400
Speaker 1: a website cv dazzle dot com. We offers some explicit

848
00:47:01,520 --> 00:47:04,439
Speaker 1: style guides that people can use. Oh wow, these are great.

849
00:47:04,480 --> 00:47:07,640
Speaker 1: I mean these look like futuristic hair and makeup designs

850
00:47:07,640 --> 00:47:10,200
Speaker 1: that you might see in like Blade Runner or something. Yes,

851
00:47:10,880 --> 00:47:14,759
Speaker 1: now again here, clearly, the the purpose of these designs

852
00:47:15,200 --> 00:47:17,840
Speaker 1: is not just to make your face look different. It

853
00:47:18,000 --> 00:47:20,960
Speaker 1: is to try to make your face look like something

854
00:47:21,000 --> 00:47:23,439
Speaker 1: other than a face, which could but I can see

855
00:47:23,520 --> 00:47:27,000
Speaker 1: proving potentially difficult coming back again to that that blog

856
00:47:27,040 --> 00:47:29,560
Speaker 1: post that we discussed in one of the earlier episodes,

857
00:47:29,840 --> 00:47:34,840
Speaker 1: um about about trying to to fool the Skype facial

858
00:47:34,880 --> 00:47:38,239
Speaker 1: recognition software. Yeah, the background blurring thing. Yeah yeah, so

859
00:47:38,400 --> 00:47:41,440
Speaker 1: like you know, even the stuff giraffe was getting recognized

860
00:47:41,480 --> 00:47:45,040
Speaker 1: as mostly a face. Uh you. So it's it's a

861
00:47:45,080 --> 00:47:48,640
Speaker 1: more difficult challenge than than one might think. But but again,

862
00:47:48,680 --> 00:47:50,840
Speaker 1: these visual examples from c v Dazzle dot com I

863
00:47:50,880 --> 00:47:53,560
Speaker 1: think will be very informative if you can't quite picture it,

864
00:47:53,600 --> 00:47:56,239
Speaker 1: look these up, yeah totally. Um. Now, a lot of

865
00:47:56,280 --> 00:48:00,080
Speaker 1: them involved things like, um, sort of different streaks of

866
00:48:00,200 --> 00:48:03,360
Speaker 1: light and dark in the hair and in lines across

867
00:48:03,440 --> 00:48:07,279
Speaker 1: the face, and makeup hair partially covering the face in

868
00:48:07,320 --> 00:48:10,200
Speaker 1: a lot of cases, things stuck on the face that

869
00:48:10,280 --> 00:48:13,200
Speaker 1: kind of make it make the shape look different. Thomas

870
00:48:13,280 --> 00:48:16,280
Speaker 1: quotes Kristoff Bush, one of the authors of that study

871
00:48:16,320 --> 00:48:19,400
Speaker 1: we mentioned earlier, and he says, quote from an academic

872
00:48:19,440 --> 00:48:23,840
Speaker 1: research perspective, the makeup attack is gaining more attention. However,

873
00:48:24,280 --> 00:48:27,719
Speaker 1: this kind of attack demands good makeup skills in order

874
00:48:27,760 --> 00:48:30,080
Speaker 1: to be successful, and that's a really good point. In

875
00:48:30,160 --> 00:48:33,080
Speaker 1: other words, it's not good enough just to do some

876
00:48:33,160 --> 00:48:36,640
Speaker 1: unusual things to your hair and put different colored makeup

877
00:48:36,680 --> 00:48:39,759
Speaker 1: on at random. The face design needs to be specially

878
00:48:39,840 --> 00:48:43,600
Speaker 1: tailored to obscure and break up specific like lines and

879
00:48:43,800 --> 00:48:47,040
Speaker 1: shapes and points on the face in order to make

880
00:48:47,080 --> 00:48:50,120
Speaker 1: it sabotage the detection stage. So if if you want

881
00:48:50,120 --> 00:48:53,040
Speaker 1: to know what these specific techniques are, you can find

882
00:48:53,080 --> 00:48:56,279
Speaker 1: guides online from people who study the issue. Now, I

883
00:48:56,560 --> 00:48:59,200
Speaker 1: think this is really cool, and you know, being being

884
00:48:59,280 --> 00:49:02,440
Speaker 1: somebody who's into I guess weirdness myself, like I like

885
00:49:02,600 --> 00:49:05,520
Speaker 1: these styles. Like if I saw somebody wearing these styles,

886
00:49:05,560 --> 00:49:08,239
Speaker 1: I would think it was cool. But I I think

887
00:49:08,280 --> 00:49:10,719
Speaker 1: we should be real and say these styles are in

888
00:49:10,920 --> 00:49:14,960
Speaker 1: daily life simply not socially feasible for a lot of people,

889
00:49:15,280 --> 00:49:17,960
Speaker 1: maybe most people. Well, for one thing, socially, they're going

890
00:49:18,000 --> 00:49:20,399
Speaker 1: to have the opposite effect. Instead, you're going to draw

891
00:49:20,440 --> 00:49:23,719
Speaker 1: attention to yourself from humans, even if you are obscuring

892
00:49:23,840 --> 00:49:26,879
Speaker 1: the attention of the machine. Yeah, that's exactly right. And

893
00:49:26,960 --> 00:49:29,480
Speaker 1: in addition to that, a lot of people's friends, families,

894
00:49:29,600 --> 00:49:33,600
Speaker 1: especially workplaces will probably not be okay with them styling

895
00:49:33,640 --> 00:49:36,680
Speaker 1: their hair and makeup deliberately to make their face look

896
00:49:36,680 --> 00:49:39,799
Speaker 1: as unlike a face as possible. I'm okay with it

897
00:49:39,840 --> 00:49:42,520
Speaker 1: around here. Yeah, me too. I don't know if the

898
00:49:42,520 --> 00:49:44,680
Speaker 1: bosses would be okay, right, You can easily imagine like

899
00:49:44,680 --> 00:49:49,319
Speaker 1: a very conservative or a governmental kind of employer, who

900
00:49:49,360 --> 00:49:52,120
Speaker 1: would you know, we're gonna have a very firm idea

901
00:49:52,120 --> 00:49:54,520
Speaker 1: of what your hair and your your makeup should consist of.

902
00:49:54,840 --> 00:49:57,279
Speaker 1: And again, if you're not allowed to wear, say, pajamas

903
00:49:57,680 --> 00:50:00,799
Speaker 1: in public, I can't imagine this would lie either. And

904
00:50:00,840 --> 00:50:04,480
Speaker 1: there's another complication actually that that makes this even more difficult.

905
00:50:04,880 --> 00:50:09,320
Speaker 1: So the CV dazzle method is only effective at fighting

906
00:50:09,360 --> 00:50:13,600
Speaker 1: face detection technologies that rely on visible light, and not

907
00:50:13,840 --> 00:50:17,880
Speaker 1: all do Thomas sites. For example, Apple's Face i D

908
00:50:18,080 --> 00:50:22,080
Speaker 1: which act which actually uses infrared light. The system detects

909
00:50:22,080 --> 00:50:25,520
Speaker 1: sort of more about the underlying contours and like bone

910
00:50:25,680 --> 00:50:28,960
Speaker 1: structure of your face, and it is not easily thrown

911
00:50:28,960 --> 00:50:32,120
Speaker 1: off by unusual patterns of light and dark colors. The

912
00:50:32,120 --> 00:50:36,399
Speaker 1: CV dazzle wouldn't necessarily affect a system like that, Uh,

913
00:50:36,440 --> 00:50:39,120
Speaker 1: though there are other methods you could maybe use. Apparently,

914
00:50:39,160 --> 00:50:41,840
Speaker 1: you might be able to protect your face from infrared

915
00:50:41,880 --> 00:50:45,400
Speaker 1: detection by, for instance, wearing a hat that projects infrared

916
00:50:45,520 --> 00:50:48,839
Speaker 1: light on your face in weird patterns, as demonstrated by

917
00:50:48,840 --> 00:50:52,880
Speaker 1: one study from Chinese and American researchers in But again, like,

918
00:50:53,120 --> 00:50:55,759
Speaker 1: is that realistic that that people would be able to

919
00:50:55,800 --> 00:50:59,640
Speaker 1: do that? Thomas mentions another method that I like overwhelming

920
00:50:59,719 --> 00:51:03,360
Speaker 1: to re action. Uh, kind of like a visual denial

921
00:51:03,440 --> 00:51:06,880
Speaker 1: of service attack on facial recognition. The solution here is

922
00:51:06,880 --> 00:51:10,879
Speaker 1: pretty simple. You cover yourself in lots of images of faces,

923
00:51:11,280 --> 00:51:15,160
Speaker 1: shirt scarf, ear rings, all with pictures of faces on them.

924
00:51:15,719 --> 00:51:18,480
Speaker 1: Would this always work? Probably not, but it will be

925
00:51:18,520 --> 00:51:21,879
Speaker 1: effective with some systems. And she ends her article by

926
00:51:21,920 --> 00:51:25,880
Speaker 1: mentioning again like major problems with existing facial recognition technology

927
00:51:25,880 --> 00:51:29,280
Speaker 1: that we've already alluded to, like huge numbers of false matches,

928
00:51:29,520 --> 00:51:32,439
Speaker 1: errors that skew along race and gender lines, all kinds

929
00:51:32,440 --> 00:51:35,560
Speaker 1: of problems like that. She ends up saying, quote, the

930
00:51:35,640 --> 00:51:39,560
Speaker 1: real solution to issues around facial recognition the tech community

931
00:51:39,600 --> 00:51:42,600
Speaker 1: working with other industries and sectors to strike an appropriate

932
00:51:42,600 --> 00:51:47,160
Speaker 1: balance between security and privacy in public spaces. Uh, Which

933
00:51:47,600 --> 00:51:49,680
Speaker 1: that may be an answer, but I mean I wonder

934
00:51:49,840 --> 00:51:52,120
Speaker 1: will that be good enough. In the first episode of

935
00:51:52,160 --> 00:51:55,799
Speaker 1: the series, we discussed several figures who have called for

936
00:51:55,880 --> 00:52:01,480
Speaker 1: either strong regulations or even outright bands on face recognition technology.

937
00:52:01,560 --> 00:52:03,319
Speaker 1: So I think next we should look at that as

938
00:52:03,320 --> 00:52:05,360
Speaker 1: a solution, maybe after we come back from a break.

939
00:52:05,800 --> 00:52:11,719
Speaker 1: All right, we'll be right back, thank thank alright, we're back.

940
00:52:11,760 --> 00:52:14,719
Speaker 1: So we were just talking about the the individual approaches

941
00:52:14,760 --> 00:52:17,440
Speaker 1: to fighting facial recognition, like things you could do to

942
00:52:17,680 --> 00:52:23,080
Speaker 1: disrupt your own image and or potentially full facial recognition device.

943
00:52:23,680 --> 00:52:26,680
Speaker 1: But now we're getting more into bands and regulations, the

944
00:52:26,719 --> 00:52:30,960
Speaker 1: broader governmental legislative moves that could be made to keep

945
00:52:31,000 --> 00:52:33,239
Speaker 1: this kind of technology from getting out of control. Yeah,

946
00:52:33,280 --> 00:52:35,000
Speaker 1: And there was one article I was reading that I

947
00:52:35,000 --> 00:52:37,560
Speaker 1: thought was pretty straightforward and made a very good case.

948
00:52:37,600 --> 00:52:41,520
Speaker 1: It was by Evans Sellinger and Woodrow heart Zog, published

949
00:52:41,560 --> 00:52:44,920
Speaker 1: in The New York Times in October seventeen nineteen. It

950
00:52:45,000 --> 00:52:48,400
Speaker 1: was called, what happens when employers can read your facial expressions?

951
00:52:48,560 --> 00:52:51,880
Speaker 1: That's that's a great question, um uh. So Sellinger is

952
00:52:51,920 --> 00:52:55,520
Speaker 1: a professor of philosophy at the Rochester Institute of Technology,

953
00:52:55,719 --> 00:52:58,040
Speaker 1: and heart Soog is a professor of law and computer

954
00:52:58,080 --> 00:53:01,520
Speaker 1: science at Northeastern University. And they are responding to the

955
00:53:01,520 --> 00:53:03,960
Speaker 1: fact that, of course many many are calling for a

956
00:53:04,000 --> 00:53:07,160
Speaker 1: band on this technology. Uh, and this is one of

957
00:53:07,200 --> 00:53:10,680
Speaker 1: these rare cases left in in the US politics, at

958
00:53:10,760 --> 00:53:15,600
Speaker 1: least where there is actually some bipartisan agreement. Apparently, some

959
00:53:15,719 --> 00:53:19,720
Speaker 1: right wing politicians like Jim Jordans have expressed concern. Speaking

960
00:53:19,760 --> 00:53:22,919
Speaker 1: to NPR in July twent nineteen, he said he thought

961
00:53:22,920 --> 00:53:25,240
Speaker 1: it was time for a time out on this technology

962
00:53:25,280 --> 00:53:27,600
Speaker 1: and that we needed to put safeguards in place before

963
00:53:27,680 --> 00:53:30,799
Speaker 1: we went forward with developing it. Meanwhile, you've got left

964
00:53:30,800 --> 00:53:34,840
Speaker 1: wing leaders Alexandria Cassio Cortez and Bernie Sanders have called

965
00:53:34,960 --> 00:53:38,160
Speaker 1: have called for regulation or bands on facial recognition. I

966
00:53:38,160 --> 00:53:40,640
Speaker 1: think uh Sanders announced the band as part of a

967
00:53:40,640 --> 00:53:44,879
Speaker 1: criminal justice reform agenda for his presidential campaign. So all

968
00:53:44,880 --> 00:53:47,880
Speaker 1: over the map, people are throwing up flags and saying

969
00:53:47,920 --> 00:53:50,400
Speaker 1: we stop this is this is scary, We need to

970
00:53:50,440 --> 00:53:53,040
Speaker 1: do something about it, and that at least is reassuring.

971
00:53:53,239 --> 00:53:56,160
Speaker 1: Let's just hope that that can maintain that. You know

972
00:53:56,160 --> 00:53:59,080
Speaker 1: that there there remains bipartisan support. It doesn't wind up

973
00:53:59,080 --> 00:54:02,000
Speaker 1: politicized in way or the other. But but certainly, the

974
00:54:02,040 --> 00:54:05,719
Speaker 1: idea of your face not becoming part of a massive database,

975
00:54:05,800 --> 00:54:08,680
Speaker 1: the idea of not being in a perpetual police lineup,

976
00:54:09,000 --> 00:54:12,440
Speaker 1: I feel like like that is going to strike a

977
00:54:12,520 --> 00:54:17,640
Speaker 1: chord with with with with with with most demographics in America. Well,

978
00:54:17,680 --> 00:54:19,399
Speaker 1: I think it's one of those weird things where there

979
00:54:19,520 --> 00:54:22,400
Speaker 1: is some bipartisan support, but there's also just not nearly

980
00:54:22,480 --> 00:54:26,440
Speaker 1: enough awareness. So like a few people on different parts

981
00:54:26,440 --> 00:54:28,840
Speaker 1: of the political spectrum are all sort of in agreement

982
00:54:28,880 --> 00:54:30,920
Speaker 1: about this, like wait a minute, we need to do something,

983
00:54:31,200 --> 00:54:34,239
Speaker 1: But it's not a lot of people overall, right, and

984
00:54:34,280 --> 00:54:36,319
Speaker 1: if all, and if the only thing you've really heard

985
00:54:36,360 --> 00:54:39,360
Speaker 1: has been like say, pitched from say a company that

986
00:54:39,480 --> 00:54:42,880
Speaker 1: is specializing in this, perhaps with a law enforcement focus,

987
00:54:42,920 --> 00:54:44,480
Speaker 1: like you just you just might think, oh, well that

988
00:54:44,520 --> 00:54:49,040
Speaker 1: sounds fine, say, you know, find missing people and stop criminals. Sure,

989
00:54:49,040 --> 00:54:50,759
Speaker 1: I'll sign off on that those are good things. These

990
00:54:50,800 --> 00:54:53,080
Speaker 1: are good things, but it's not the complete picture, right

991
00:54:53,480 --> 00:54:55,799
Speaker 1: uh So they also the authors here note that many

992
00:54:55,880 --> 00:55:00,320
Speaker 1: local governments have already formally restricted their government agencies, including police,

993
00:55:00,400 --> 00:55:03,040
Speaker 1: from using it, so nothing really strong has happened at

994
00:55:03,040 --> 00:55:08,320
Speaker 1: the national level, but like San Francisco, Oakland, Berkeley, Somerville, Massachusetts,

995
00:55:08,320 --> 00:55:10,960
Speaker 1: and some other local areas have have put a ban

996
00:55:11,080 --> 00:55:14,040
Speaker 1: in place. So the authors of this piece, Hertzog and

997
00:55:14,120 --> 00:55:17,360
Speaker 1: Sellinger here argue that it is not enough just to

998
00:55:17,560 --> 00:55:21,239
Speaker 1: limit the ways in which facial recognition is used, or

999
00:55:21,280 --> 00:55:24,719
Speaker 1: to say, restrict government agencies such as police departments, from

1000
00:55:24,719 --> 00:55:28,799
Speaker 1: buying these tools. They say that, unfortunately, the only way

1001
00:55:28,840 --> 00:55:32,279
Speaker 1: to actually protect ourselves is to enact a complete and

1002
00:55:32,320 --> 00:55:36,320
Speaker 1: total ban on the technology. Quote, we must ban facial

1003
00:55:36,360 --> 00:55:40,360
Speaker 1: recognition in both public and private sectors before we grow

1004
00:55:40,480 --> 00:55:43,840
Speaker 1: so dependent on it that we accept its inevitable harms

1005
00:55:43,880 --> 00:55:48,759
Speaker 1: as necessary for progress. Perhaps over time, appropriate policies can

1006
00:55:48,800 --> 00:55:52,440
Speaker 1: be enacted that justify lifting a ban, but we doubt it.

1007
00:55:53,040 --> 00:55:55,120
Speaker 1: So I think they make a pretty good case, and

1008
00:55:55,120 --> 00:55:57,080
Speaker 1: and we'll get to a little bit more about it

1009
00:55:57,120 --> 00:55:59,200
Speaker 1: in a second. But you know, you might be wondering,

1010
00:55:59,239 --> 00:56:02,800
Speaker 1: like if there is actually some bipartisan agreement that facial

1011
00:56:02,880 --> 00:56:06,040
Speaker 1: recognition could be devastating to our basic liberties, like what's

1012
00:56:06,080 --> 00:56:09,239
Speaker 1: the hold up? You know, what's the problem. And so

1013
00:56:09,320 --> 00:56:12,080
Speaker 1: of course, the authors note that in general, the United

1014
00:56:12,160 --> 00:56:15,799
Speaker 1: States is very reticent to enact bands on technology like

1015
00:56:15,840 --> 00:56:18,600
Speaker 1: with with one of the few counter examples being various

1016
00:56:18,600 --> 00:56:21,279
Speaker 1: types of malware. And I think that's good because I

1017
00:56:21,320 --> 00:56:24,440
Speaker 1: think it makes sense to think of facial recognition technology

1018
00:56:24,480 --> 00:56:28,360
Speaker 1: as a form of cultural malware seeps in makes copies

1019
00:56:28,400 --> 00:56:30,799
Speaker 1: of itself. You know, you get it as a byproduct

1020
00:56:30,800 --> 00:56:34,600
Speaker 1: of something you wanted to download. But of course there's

1021
00:56:34,600 --> 00:56:37,480
Speaker 1: another thing. The authors don't really speculate about these motives,

1022
00:56:37,520 --> 00:56:40,040
Speaker 1: but obviously one big hurdle is that there's just a

1023
00:56:40,080 --> 00:56:42,279
Speaker 1: lot of money to be made in this sector, and

1024
00:56:42,360 --> 00:56:46,960
Speaker 1: even worse, there are significant sunk costs. Like many extremely

1025
00:56:47,040 --> 00:56:52,200
Speaker 1: talented people and powerful institutions have already devoted significant resources

1026
00:56:52,680 --> 00:56:57,040
Speaker 1: and time to improving facial recognition software. And we know

1027
00:56:57,120 --> 00:57:00,799
Speaker 1: that humans do not like abandoning sunk costs. Once you've

1028
00:57:00,800 --> 00:57:05,000
Speaker 1: already invested in something, you're kind of your psychologically stuck

1029
00:57:05,040 --> 00:57:06,879
Speaker 1: with it. You know, do you want to throw all

1030
00:57:06,920 --> 00:57:09,640
Speaker 1: the all that work in the trash now? Right? Right?

1031
00:57:09,880 --> 00:57:11,799
Speaker 1: You know, companies that have we have created this kind

1032
00:57:11,840 --> 00:57:14,359
Speaker 1: of technology that are looking for ways to expand new

1033
00:57:14,440 --> 00:57:17,160
Speaker 1: markets that they can they can get into new new uses,

1034
00:57:17,200 --> 00:57:19,760
Speaker 1: and certainly when you have uses that in and off

1035
00:57:19,800 --> 00:57:26,600
Speaker 1: themselves are advantageous things like making payments, personal security um

1036
00:57:26,680 --> 00:57:28,800
Speaker 1: and just the you know, the broad broadly speaking, the

1037
00:57:28,920 --> 00:57:33,200
Speaker 1: idea of finding missing persons and catching criminals. Yeah, I

1038
00:57:33,200 --> 00:57:36,880
Speaker 1: mean that makes all those things in isolation make they

1039
00:57:37,000 --> 00:57:40,240
Speaker 1: make a good case, right, And that's actually the first

1040
00:57:40,280 --> 00:57:42,960
Speaker 1: to main argument. So the authors of this piece outline

1041
00:57:43,040 --> 00:57:45,800
Speaker 1: three major arguments that are used by the advocates of

1042
00:57:45,840 --> 00:57:49,680
Speaker 1: facial recognition, and the first one is exactly that is that, well,

1043
00:57:50,160 --> 00:57:52,920
Speaker 1: there might be some harms, but they argue the potential

1044
00:57:52,960 --> 00:57:56,040
Speaker 1: benefits outweigh the harms. Uh. You know, again, think about

1045
00:57:56,040 --> 00:57:58,400
Speaker 1: the benefits to law enforcement alone. Think of all the

1046
00:57:58,480 --> 00:58:00,560
Speaker 1: violent criminals that could be called, Think of all the

1047
00:58:00,600 --> 00:58:03,400
Speaker 1: missing persons that could be found. Uh. And on top

1048
00:58:03,440 --> 00:58:06,840
Speaker 1: of that, think about some of the obvious consumer demand

1049
00:58:07,000 --> 00:58:09,560
Speaker 1: that there would be for public versions of the software.

1050
00:58:09,920 --> 00:58:11,720
Speaker 1: I mean, it's the kind of thing that, like, we

1051
00:58:11,800 --> 00:58:14,520
Speaker 1: would be terrified about the thought of people using it

1052
00:58:14,560 --> 00:58:17,520
Speaker 1: on us, but might be really excited to use it

1053
00:58:17,520 --> 00:58:20,520
Speaker 1: on others. You know. It's just like this, uh, this

1054
00:58:20,680 --> 00:58:24,400
Speaker 1: basic failure to like reverse the situation and apply the

1055
00:58:24,440 --> 00:58:26,920
Speaker 1: same rules you would want applied to yourself to other people.

1056
00:58:27,320 --> 00:58:30,680
Speaker 1: So under this mindset, instead of being banned, the people

1057
00:58:30,680 --> 00:58:32,720
Speaker 1: who say, you know, the benefits outweigh the harms, they

1058
00:58:32,720 --> 00:58:36,680
Speaker 1: would probably say, well, facial recognition should be lightly regulated.

1059
00:58:36,760 --> 00:58:40,040
Speaker 1: You know, maybe we could require transparency and make sure

1060
00:58:40,080 --> 00:58:43,280
Speaker 1: consumers are aware when face data is being harvested for

1061
00:58:43,320 --> 00:58:46,800
Speaker 1: recognition purposes. You know you can't do it surreptitiously. Uh

1062
00:58:46,800 --> 00:58:50,200
Speaker 1: And the authors here disagree, they write, quote, notice and

1063
00:58:50,320 --> 00:58:54,760
Speaker 1: choice has been an abysmal failure. Social media companies, airlines,

1064
00:58:54,760 --> 00:58:58,320
Speaker 1: and retailers overhyped the short term benefits of facial recognition

1065
00:58:58,560 --> 00:59:02,600
Speaker 1: while using unreadable privacy policies and vague disclaimers that make

1066
00:59:02,640 --> 00:59:06,960
Speaker 1: it hard to understand how the technology endangers users privacies

1067
00:59:06,960 --> 00:59:10,600
Speaker 1: and freedom. Uh So, you know, like, does anybody actually

1068
00:59:10,640 --> 00:59:14,480
Speaker 1: read the end user license Agreement? Of course not No,

1069
00:59:14,800 --> 00:59:17,920
Speaker 1: I mean, like, even if we did, would some vague

1070
00:59:18,040 --> 00:59:21,600
Speaker 1: legal phrasing about ownership of face data cause us to

1071
00:59:21,720 --> 00:59:26,200
Speaker 1: actually forego participation in technological trends that everybody around us

1072
00:59:26,280 --> 00:59:29,360
Speaker 1: is adopting. I mean again, of course not, Like, even

1073
00:59:29,400 --> 00:59:33,160
Speaker 1: if the harms vastly outweigh the benefits. The benefits are

1074
00:59:33,200 --> 00:59:37,640
Speaker 1: immediate and concrete, and the harms are long term and abstract.

1075
00:59:38,000 --> 00:59:41,000
Speaker 1: Exactly the kinds of cases where we are so bad

1076
00:59:41,080 --> 00:59:44,000
Speaker 1: at like making informed decisions, like would you like a

1077
00:59:44,040 --> 00:59:46,640
Speaker 1: slice of pizza right now? Just be aware that you

1078
00:59:46,680 --> 00:59:49,560
Speaker 1: know it may compromise the integrity of your identity in

1079
00:59:49,600 --> 00:59:53,080
Speaker 1: some way that's difficult to picture. Now. The next counter

1080
00:59:53,200 --> 00:59:56,440
Speaker 1: argument they explore is that the idea that strong fears

1081
00:59:56,480 --> 01:00:01,439
Speaker 1: about new technologies are overreactions, and we've you know, we've

1082
01:00:01,440 --> 01:00:03,600
Speaker 1: looked at lots of ways that this can absolutely happen.

1083
01:00:03,680 --> 01:00:05,880
Speaker 1: We've discussed it on this show and also a lot

1084
01:00:05,920 --> 01:00:08,400
Speaker 1: on invention. Think about the panic about a racer. Just

1085
01:00:08,440 --> 01:00:11,280
Speaker 1: remember that, yes, yes, definitely, the idea that oh, racers

1086
01:00:11,360 --> 01:00:14,360
Speaker 1: are are exist. All my writings will be a race

1087
01:00:14,800 --> 01:00:17,680
Speaker 1: or remember some of the panics that came with the

1088
01:00:17,880 --> 01:00:21,040
Speaker 1: advent of photography, right, yeah, and just the general idea

1089
01:00:21,040 --> 01:00:23,680
Speaker 1: of a future shock, you know, the idea that that

1090
01:00:23,800 --> 01:00:27,920
Speaker 1: rapidly advancing technology is overwhelming and uh um yeah, that

1091
01:00:28,000 --> 01:00:30,919
Speaker 1: is a subject unto itself. Yeah, So this argument would

1092
01:00:30,960 --> 01:00:33,720
Speaker 1: be that things sometimes just seems scary and provoke shock

1093
01:00:34,040 --> 01:00:36,400
Speaker 1: because they're new, but once we get used to it,

1094
01:00:36,400 --> 01:00:39,480
Speaker 1: it's great. Uh. The authors disagree. They do not think

1095
01:00:39,520 --> 01:00:42,200
Speaker 1: this is the case with facial recognition. They argue that

1096
01:00:42,200 --> 01:00:46,200
Speaker 1: the backlash is not just hyperventilating about something that's unfamiliar.

1097
01:00:46,560 --> 01:00:50,280
Speaker 1: In very concrete ways, facial recognition does have a unique

1098
01:00:50,400 --> 01:00:54,600
Speaker 1: power to create a world with pervasive automated surveillance in

1099
01:00:54,640 --> 01:00:59,400
Speaker 1: a way that's disempowering to an individuals in almost uncountable ways,

1100
01:01:00,160 --> 01:01:03,480
Speaker 1: they write, quote, big companies, government agencies, and even your

1101
01:01:03,520 --> 01:01:06,360
Speaker 1: next door neighbors will seek to deploy it in more places.

1102
01:01:06,640 --> 01:01:08,920
Speaker 1: They'll want to identify and track you. They'll want to

1103
01:01:08,960 --> 01:01:12,640
Speaker 1: categorize your emotions and identity. They will want to infer

1104
01:01:12,720 --> 01:01:15,680
Speaker 1: where you might shop, protest, or work, and use that

1105
01:01:15,760 --> 01:01:19,520
Speaker 1: information to control and manipulate you, to deprive you of opportunities.

1106
01:01:19,880 --> 01:01:22,720
Speaker 1: It's likely that the technology will be used to police

1107
01:01:22,760 --> 01:01:26,800
Speaker 1: social norms. People who skip church or jaywalk will be

1108
01:01:26,880 --> 01:01:31,120
Speaker 1: noticed and potentially ostracized. And you'd better start practicing your

1109
01:01:31,120 --> 01:01:34,840
Speaker 1: most convincing facial expressions otherwise during your next job interview,

1110
01:01:35,040 --> 01:01:38,680
Speaker 1: a computer could code you as a liar or a malcontent. Now,

1111
01:01:38,720 --> 01:01:40,720
Speaker 1: remember in the last episode when we talked about like

1112
01:01:40,800 --> 01:01:44,040
Speaker 1: these services that are being sold as like identifying people's

1113
01:01:44,080 --> 01:01:47,480
Speaker 1: emotions through facial recognition, which there is a major study

1114
01:01:47,560 --> 01:01:49,960
Speaker 1: that really undercut that and said these things are not

1115
01:01:50,160 --> 01:01:53,000
Speaker 1: very accurate, but that doesn't mean they're not going to

1116
01:01:53,080 --> 01:01:57,000
Speaker 1: be used. Right, and in terms of policing social norms,

1117
01:01:57,280 --> 01:02:01,040
Speaker 1: again go back to the pajamas because it's it's slightly

1118
01:02:01,120 --> 01:02:03,800
Speaker 1: hilarious as the pajama cases. It is also frightening because

1119
01:02:03,800 --> 01:02:07,840
Speaker 1: it is a firm example of facial recognition technology being

1120
01:02:07,920 --> 01:02:11,080
Speaker 1: used to police social norms. Yes. Uh. And then the

1121
01:02:11,160 --> 01:02:14,240
Speaker 1: third counter argument they look at is basically the argument

1122
01:02:14,240 --> 01:02:18,240
Speaker 1: that Bruce Schneyer was making when we referenced his article earlier. Uh.

1123
01:02:18,280 --> 01:02:22,360
Speaker 1: It's that facial recognition technology is just one branch of

1124
01:02:22,400 --> 01:02:26,320
Speaker 1: a broader privacy and civil liberty debate. Um, and we

1125
01:02:26,320 --> 01:02:28,920
Speaker 1: we need to focus on all surveillance, not just this

1126
01:02:28,960 --> 01:02:32,600
Speaker 1: particular technology. So how about you know, things that identify

1127
01:02:32,680 --> 01:02:35,240
Speaker 1: you by your gate, the way you move, or about

1128
01:02:35,480 --> 01:02:38,760
Speaker 1: retinal scanning or brain scanning or anything like that. And

1129
01:02:38,800 --> 01:02:40,680
Speaker 1: I'd say about this one, well, you know, it's true

1130
01:02:40,800 --> 01:02:43,600
Speaker 1: like other technologies could represent the kind of threat to

1131
01:02:43,640 --> 01:02:46,920
Speaker 1: privacy and freedom that facial recognition presents. And some of

1132
01:02:46,920 --> 01:02:49,080
Speaker 1: those are scary and awful too. You know, we're not

1133
01:02:49,120 --> 01:02:53,439
Speaker 1: saying like don't, um, don't ban gate recognition scanning, or

1134
01:02:53,680 --> 01:02:58,160
Speaker 1: you know, don't scan my brain while I'm at a restaurant. Yeah, exactly.

1135
01:02:58,160 --> 01:03:00,480
Speaker 1: I mean I would say facial recognition is going special

1136
01:03:00,480 --> 01:03:05,240
Speaker 1: attention because it's already here, like people are selling these programs. Uh.

1137
01:03:05,240 --> 01:03:08,600
Speaker 1: And another thing is that, you know, different technologies are

1138
01:03:08,640 --> 01:03:12,520
Speaker 1: slightly different, and they require different regulatory schemes. You know,

1139
01:03:12,560 --> 01:03:17,120
Speaker 1: the authors here point out that the law singles out automobile, spyware,

1140
01:03:17,240 --> 01:03:19,760
Speaker 1: medical devices, and a bunch of other different kinds of

1141
01:03:19,760 --> 01:03:22,760
Speaker 1: technologies with their own laws and rules. They're not all

1142
01:03:22,800 --> 01:03:26,520
Speaker 1: covered under one type of law. We've got individual agencies

1143
01:03:26,520 --> 01:03:30,600
Speaker 1: for like airplanes and for phones and stuff. But then

1144
01:03:30,680 --> 01:03:34,919
Speaker 1: another thing is that faces are central to our identities.

1145
01:03:35,200 --> 01:03:37,280
Speaker 1: In the first episode, you know, we joked about going

1146
01:03:37,360 --> 01:03:42,200
Speaker 1: around constantly changing masks, but is this socially feasible? I mean,

1147
01:03:42,320 --> 01:03:45,400
Speaker 1: be realistic, Like we need to see each other's faces

1148
01:03:45,440 --> 01:03:48,760
Speaker 1: in order to see each other. Seeing faces is the

1149
01:03:48,800 --> 01:03:51,400
Speaker 1: soul of human life. Yeah. One of the one of

1150
01:03:51,440 --> 01:03:53,960
Speaker 1: the whole aspects of the whole bummer situation that we've

1151
01:03:54,000 --> 01:03:58,120
Speaker 1: discussed involving social media and just electronic culture in general,

1152
01:03:58,160 --> 01:04:00,600
Speaker 1: is that we don't see each other's faces, not not

1153
01:04:00,680 --> 01:04:04,280
Speaker 1: the living face, not the face that shows expression and humanity.

1154
01:04:04,800 --> 01:04:08,120
Speaker 1: We just see the manufactured faces. And uh, we don't

1155
01:04:08,120 --> 01:04:11,160
Speaker 1: want to live in a world of physically manufactured faces

1156
01:04:11,280 --> 01:04:13,960
Speaker 1: because we've could created a technology and allowed it to

1157
01:04:13,960 --> 01:04:16,360
Speaker 1: get out of hand. Yeah. Uh. And so I think

1158
01:04:16,400 --> 01:04:18,760
Speaker 1: often with issues like this and and something you run

1159
01:04:18,800 --> 01:04:21,120
Speaker 1: into with a lot of these counter arguments that these

1160
01:04:21,160 --> 01:04:24,840
Speaker 1: things that are in favor of facial recognition, or we

1161
01:04:24,960 --> 01:04:27,240
Speaker 1: argue in ways that fall into a trap of talking

1162
01:04:27,280 --> 01:04:31,439
Speaker 1: about just what's possible or what's technically true, rather than

1163
01:04:31,480 --> 01:04:35,240
Speaker 1: what's realistic and what's practical. Like could you protect people

1164
01:04:35,360 --> 01:04:37,520
Speaker 1: with opt in methods where they have to sign a

1165
01:04:37,680 --> 01:04:41,840
Speaker 1: ula disclosing that they've surrendered their privacy? I mean in theory, yes,

1166
01:04:41,880 --> 01:04:44,360
Speaker 1: but in practice we know we just know this doesn't work,

1167
01:04:44,400 --> 01:04:46,880
Speaker 1: you know, we all just click. I agree, And again,

1168
01:04:46,920 --> 01:04:49,800
Speaker 1: the benefits are immediate and concrete, and the harms are

1169
01:04:49,920 --> 01:04:53,720
Speaker 1: long term and abstract. Yeah, I want to see what

1170
01:04:53,760 --> 01:04:55,520
Speaker 1: I'll look like as an old man on my iPhone

1171
01:04:55,640 --> 01:04:58,480
Speaker 1: right now. I don't care where this software, where this

1172
01:04:58,520 --> 01:05:00,880
Speaker 1: app is coming from, and where of this facial data

1173
01:05:00,960 --> 01:05:03,600
Speaker 1: might or might not be going. Yeah, and I'm obviously

1174
01:05:03,640 --> 01:05:05,760
Speaker 1: i'd advise people not to do that. But if you

1175
01:05:05,840 --> 01:05:08,440
Speaker 1: react to that with the mentality of will you signed

1176
01:05:08,480 --> 01:05:11,360
Speaker 1: the contract, You've got nothing to complain about. That just

1177
01:05:11,480 --> 01:05:15,560
Speaker 1: doesn't take the experience of human life seriously, right, And

1178
01:05:15,640 --> 01:05:18,160
Speaker 1: it's nothing one of these cases which to to um

1179
01:05:18,200 --> 01:05:22,240
Speaker 1: to summon the words of William Gibson of technology making

1180
01:05:22,800 --> 01:05:25,640
Speaker 1: packs with the devil possible, things that were only the

1181
01:05:25,960 --> 01:05:30,040
Speaker 1: wild imaginings of of people in the past we bring

1182
01:05:30,080 --> 01:05:34,240
Speaker 1: to life with technology making AI demons, making it possible

1183
01:05:34,320 --> 01:05:38,240
Speaker 1: to to to sell your face to some faceless entity.

1184
01:05:38,480 --> 01:05:40,640
Speaker 1: I think that's exactly right. I mean, this is like,

1185
01:05:41,120 --> 01:05:43,760
Speaker 1: this is somehow a case of like a horrible fantasy

1186
01:05:43,840 --> 01:05:47,560
Speaker 1: becoming reality. Um. And you know, and this could be

1187
01:05:47,600 --> 01:05:48,840
Speaker 1: the case with a lot of things. So it's not

1188
01:05:48,880 --> 01:05:51,640
Speaker 1: just the u LA agreements about you know, about people

1189
01:05:51,880 --> 01:05:56,280
Speaker 1: arguing about like what's technically possible without just acknowledging what's realistic.

1190
01:05:56,400 --> 01:05:58,320
Speaker 1: What are people really going to do with their lives?

1191
01:05:58,600 --> 01:06:00,840
Speaker 1: Same thing is, you know, you could argue, well, if

1192
01:06:00,880 --> 01:06:02,800
Speaker 1: you don't like it, you could go around with masks

1193
01:06:02,840 --> 01:06:05,160
Speaker 1: on or something in order to opt out. I mean

1194
01:06:05,280 --> 01:06:08,880
Speaker 1: you could try that. But are people actually practically gonna

1195
01:06:08,920 --> 01:06:11,200
Speaker 1: want to live that way? Yeah? I mean not only

1196
01:06:11,240 --> 01:06:15,560
Speaker 1: living with a mask, but living among the masks, because God,

1197
01:06:15,600 --> 01:06:18,200
Speaker 1: we've talked about this before, just the psychology of of

1198
01:06:18,280 --> 01:06:20,440
Speaker 1: masks and what happens when you wear one, and the

1199
01:06:20,480 --> 01:06:23,360
Speaker 1: group think of mask wearing and and generally the idea

1200
01:06:23,400 --> 01:06:26,200
Speaker 1: of encountering a bunch of people wearing masks or the

1201
01:06:26,240 --> 01:06:29,400
Speaker 1: same mask is a frightening a proposition and they are

1202
01:06:29,560 --> 01:06:34,960
Speaker 1: countless historical examples of that. Yeah. So the authors here conclude, quote,

1203
01:06:35,160 --> 01:06:38,960
Speaker 1: we support a wide ranging ban on this powerful, powerful technology,

1204
01:06:39,000 --> 01:06:42,800
Speaker 1: but even limited prohibitions on its use in police body cams,

1205
01:06:42,920 --> 01:06:46,000
Speaker 1: d m V databases, public housing, and schools would be

1206
01:06:46,040 --> 01:06:49,320
Speaker 1: an important start. And they say the public is ready

1207
01:06:49,360 --> 01:06:52,520
Speaker 1: for this, and the actions by San Francisco, Somerville, Berkeley,

1208
01:06:52,520 --> 01:06:55,120
Speaker 1: and Oakland show it. Our society does not have to

1209
01:06:55,160 --> 01:06:59,040
Speaker 1: allow the spread of new technology that endangers our privacy.

1210
01:06:59,240 --> 01:07:01,480
Speaker 1: And I guess it, just speaking for myself, I I

1211
01:07:01,840 --> 01:07:04,600
Speaker 1: am highly convinced by this point of view. I mean, like,

1212
01:07:04,720 --> 01:07:08,200
Speaker 1: I think, if you ultimately in the future were really

1213
01:07:08,280 --> 01:07:10,880
Speaker 1: confident that you wanted to change your mind and move

1214
01:07:10,960 --> 01:07:14,200
Speaker 1: forward with facial recognition technology, you could lift a ban.

1215
01:07:14,800 --> 01:07:17,640
Speaker 1: But like, you can't undo it once the technology is

1216
01:07:17,680 --> 01:07:21,680
Speaker 1: out there and and it's coming really fast. Absolutely, yeah.

1217
01:07:21,680 --> 01:07:23,560
Speaker 1: I mean on the other side is yes, so you

1218
01:07:23,560 --> 01:07:26,040
Speaker 1: can think of various situations where you put bands in

1219
01:07:26,120 --> 01:07:29,480
Speaker 1: place and then something bad happened, something summone's up a

1220
01:07:29,480 --> 01:07:32,200
Speaker 1: great deal of fear in a nation and allows us

1221
01:07:32,200 --> 01:07:35,200
Speaker 1: to slide back into this situation and say, give up

1222
01:07:35,480 --> 01:07:38,000
Speaker 1: various privacies and rights in the name of feeling a

1223
01:07:38,040 --> 01:07:41,440
Speaker 1: little less afraid. But that doesn't mean it's not worth

1224
01:07:41,560 --> 01:07:45,280
Speaker 1: fighting for now, you know, before the fear, because imagine

1225
01:07:45,320 --> 01:07:49,000
Speaker 1: how much further we would sink into into the fear. Uh,

1226
01:07:49,040 --> 01:07:51,640
Speaker 1: you know if we already had all these technologies in place,

1227
01:07:51,720 --> 01:07:54,880
Speaker 1: eroting and taking aware of freedoms. This is a topic

1228
01:07:54,920 --> 01:07:57,000
Speaker 1: I think it's worth spreading the word about. I mean, like,

1229
01:07:57,080 --> 01:07:58,960
Speaker 1: this is something a lot of people just probably aren't

1230
01:07:58,960 --> 01:08:02,160
Speaker 1: thinking about at all, and it's coming really fast. So uh,

1231
01:08:03,040 --> 01:08:06,960
Speaker 1: it's worth raising the awareness. Yeah, yeah, worth uh worth

1232
01:08:07,640 --> 01:08:11,680
Speaker 1: recording three episodes that I must admit are not fun episodes.

1233
01:08:13,040 --> 01:08:14,800
Speaker 1: The middle one was a lot of fun. I thought, okay,

1234
01:08:14,800 --> 01:08:16,320
Speaker 1: the middle one. The middle one was more fun because

1235
01:08:16,320 --> 01:08:20,120
Speaker 1: we're just talking about the basic biological facial recognition and

1236
01:08:20,280 --> 01:08:23,720
Speaker 1: parts of this one we're at least entertaining that that

1237
01:08:23,760 --> 01:08:26,439
Speaker 1: Wired article was was a very fun but also you know,

1238
01:08:26,640 --> 01:08:30,840
Speaker 1: disturbing read. Uh, but it is. This is a troubling topic.

1239
01:08:30,920 --> 01:08:33,960
Speaker 1: I am. I am more troubled for having researched it

1240
01:08:34,000 --> 01:08:38,200
Speaker 1: and recorded it. But I think we should be troubled

1241
01:08:38,240 --> 01:08:40,880
Speaker 1: by things like this, and it gives us the It

1242
01:08:40,920 --> 01:08:44,160
Speaker 1: gives us the space from which to to act, uh,

1243
01:08:44,200 --> 01:08:47,000
Speaker 1: you know, hopefully just by spreading the word about it

1244
01:08:47,040 --> 01:08:48,960
Speaker 1: and uh and you know, getting to a place where

1245
01:08:49,000 --> 01:08:51,439
Speaker 1: we have some of these protections in place for our

1246
01:08:51,479 --> 01:08:55,439
Speaker 1: own faces. That's right, don't build the nightmare mask land right,

1247
01:08:55,560 --> 01:08:58,160
Speaker 1: save your face, Do not go gentle into that good

1248
01:08:58,240 --> 01:09:02,320
Speaker 1: night right rage, rage against the scanning of the face. Yes,

1249
01:09:04,320 --> 01:09:06,800
Speaker 1: I will stress though. This was the third of a

1250
01:09:06,880 --> 01:09:08,600
Speaker 1: three part series, So if you didn't listen to the

1251
01:09:08,600 --> 01:09:10,640
Speaker 1: other two, go back and listen to them because it

1252
01:09:10,640 --> 01:09:12,839
Speaker 1: will help fill in all the pieces for you totally.

1253
01:09:12,960 --> 01:09:14,920
Speaker 1: But also it's worth pointing out that this is such

1254
01:09:14,960 --> 01:09:17,360
Speaker 1: a bleeding edge issue that even though you'll be listening

1255
01:09:17,400 --> 01:09:20,719
Speaker 1: to this episode mere days after we recorded it, uh,

1256
01:09:20,760 --> 01:09:22,839
Speaker 1: there's just there's just gonna be more and more coverage.

1257
01:09:22,880 --> 01:09:27,000
Speaker 1: There's gonna be you know, new movements happening, new new studies,

1258
01:09:27,479 --> 01:09:30,680
Speaker 1: sadly new rollouts that are going to meet with controversy.

1259
01:09:30,800 --> 01:09:33,640
Speaker 1: So um, just keep that in mind, especially if you

1260
01:09:33,720 --> 01:09:36,559
Speaker 1: end up coming back to this episode same months from now.

1261
01:09:36,680 --> 01:09:39,040
Speaker 1: We're in the frenzy period right now, all right in

1262
01:09:39,040 --> 01:09:40,719
Speaker 1: the meantime, if you want to check out other episodes

1263
01:09:40,720 --> 01:09:42,400
Speaker 1: of Stuff to Blow Your Mind, you can find us

1264
01:09:42,400 --> 01:09:44,840
Speaker 1: wherever you get your podcasts. If you want to just

1265
01:09:45,080 --> 01:09:47,040
Speaker 1: jump on over to the I Heart listing for the show,

1266
01:09:47,040 --> 01:09:48,519
Speaker 1: go to stuff to Blow your Mind dot com and

1267
01:09:48,520 --> 01:09:51,120
Speaker 1: you will be redirected. But wherever you get the show,

1268
01:09:51,280 --> 01:09:54,320
Speaker 1: make sure that you rate, review, and spread the word,

1269
01:09:54,400 --> 01:09:56,960
Speaker 1: because that's how we keep the show alive. Hughes, thanks

1270
01:09:57,000 --> 01:09:59,960
Speaker 1: as always to our excellent audio producer Seth Nicholas Johnson.

1271
01:10:00,200 --> 01:10:01,640
Speaker 1: If you'd like to get in touch with us with

1272
01:10:01,680 --> 01:10:04,160
Speaker 1: feedback on this episode or any other, to suggest topic

1273
01:10:04,240 --> 01:10:06,320
Speaker 1: for the future, or just to say hello, you can

1274
01:10:06,360 --> 01:10:09,200
Speaker 1: email us at contact at stuff to Blow your Mind

1275
01:10:09,360 --> 01:10:18,719
Speaker 1: dot com. Stuff to Blow Your Mind is a production

1276
01:10:18,760 --> 01:10:21,240
Speaker 1: of iHeart Radio's How Stuff Works. For more podcasts from

1277
01:10:21,280 --> 01:10:24,120
Speaker 1: my Heart Radio, visit the iHeart Radio app, Apple Podcasts,

1278
01:10:24,200 --> 01:10:31,880
Speaker 1: or wherever you listen to your favorite shows. Bo