1
00:00:02,040 --> 00:00:05,000
Speaker 1: Too weekend Breakfast the Future of.

2
00:00:05,280 --> 00:00:08,200
Speaker 2: It is eight minutes after nine o'clock time for us

3
00:00:08,480 --> 00:00:11,360
Speaker 2: to talk about the future of And this week is

4
00:00:11,360 --> 00:00:14,000
Speaker 2: our week where we talk about AI and we'll be

5
00:00:14,040 --> 00:00:17,480
Speaker 2: looking at two stories this week. The one Google employees

6
00:00:17,520 --> 00:00:21,200
Speaker 2: as well as Open AI employees are asking for clearer

7
00:00:21,239 --> 00:00:24,680
Speaker 2: limits on how their employers work with the military. This

8
00:00:24,800 --> 00:00:28,880
Speaker 2: is after the US carried out strikes on Iran last

9
00:00:28,880 --> 00:00:33,479
Speaker 2: weekend and they blacklisted the AI models from Anthropic. We'll

10
00:00:33,479 --> 00:00:37,520
Speaker 2: also speak about the Meta ray Band AI glasses, those

11
00:00:37,560 --> 00:00:41,159
Speaker 2: smart glasses that they released a short while ago. What

12
00:00:41,200 --> 00:00:44,200
Speaker 2: do we know about these glasses and how they work?

13
00:00:44,240 --> 00:00:46,479
Speaker 2: We'll talk about that as well. Joined us always in

14
00:00:46,520 --> 00:00:51,000
Speaker 2: studio by our resident expert is chief Data and Analytics

15
00:00:51,000 --> 00:00:54,080
Speaker 2: offsidet F and b Risk. He's also the author of

16
00:00:54,160 --> 00:00:57,640
Speaker 2: two must read books on AI. Doctor Mark Nacilla joins

17
00:00:57,720 --> 00:01:01,000
Speaker 2: us in studio. Doctor Nasilla, Good mone.

18
00:01:00,720 --> 00:01:04,200
Speaker 3: Good morning gooks, and to the listeners who are joining us.

19
00:01:04,280 --> 00:01:05,120
Speaker 3: Great to be here again.

20
00:01:05,240 --> 00:01:07,200
Speaker 2: Oh is a pleasure to see you or is a

21
00:01:07,200 --> 00:01:11,000
Speaker 2: pleasure to be in studios? So no shortage of developments

22
00:01:11,040 --> 00:01:14,080
Speaker 2: in the AI space. Lot's happening, but I thought it

23
00:01:14,200 --> 00:01:17,280
Speaker 2: was quite interesting that you now have employees or some

24
00:01:17,319 --> 00:01:20,759
Speaker 2: of these big companies that are making massive investments in AI,

25
00:01:21,319 --> 00:01:25,440
Speaker 2: they are coming out and asking their employers to sit

26
00:01:25,880 --> 00:01:29,840
Speaker 2: very clear or clearer limits on how the employers are

27
00:01:29,840 --> 00:01:32,880
Speaker 2: going to be working with the military. So let's just

28
00:01:32,959 --> 00:01:35,959
Speaker 2: take a step back before we get into this. Just

29
00:01:36,000 --> 00:01:39,920
Speaker 2: remind us what happened with Anthropic and the Pentagon just

30
00:01:39,920 --> 00:01:42,280
Speaker 2: a few weeks ago, which is kind of the backdrop

31
00:01:42,480 --> 00:01:45,080
Speaker 2: for what we're seeing now with the employees of the

32
00:01:45,080 --> 00:01:47,280
Speaker 2: other tech companies gooks.

33
00:01:48,240 --> 00:01:53,400
Speaker 3: A few weeks ago Anthropic and the Pentagon, Donald Trump

34
00:01:53,480 --> 00:02:00,080
Speaker 3: had a bit of a controversy around leveraging their and

35
00:02:00,160 --> 00:02:05,840
Speaker 3: Tropic platform cloud for mass surveillance as well as autonomous weapons,

36
00:02:05,880 --> 00:02:12,120
Speaker 3: and that led to the United States government blacklisting Anthropy.

37
00:02:12,000 --> 00:02:13,760
Speaker 1: And calling them a supply chain risk.

38
00:02:13,960 --> 00:02:19,560
Speaker 3: Supply chain risk basically being blacklisted for inability to deliver

39
00:02:20,360 --> 00:02:21,320
Speaker 3: based on their promise.

40
00:02:21,400 --> 00:02:22,279
Speaker 1: But also.

41
00:02:23,960 --> 00:02:30,040
Speaker 3: Some of the controversy has been noted under the Defense forts,

42
00:02:31,639 --> 00:02:35,960
Speaker 3: you know, strategies around being able not to protect the

43
00:02:35,960 --> 00:02:36,839
Speaker 3: government right.

44
00:02:37,080 --> 00:02:40,639
Speaker 2: Yes, And so this week, after kind of that fallout,

45
00:02:41,440 --> 00:02:44,840
Speaker 2: we now hear that tech workers at Google Open AI.

46
00:02:45,160 --> 00:02:48,639
Speaker 2: Two of the companies that we know are making pretty

47
00:02:48,639 --> 00:02:53,240
Speaker 2: big investments in AI. But was it who was in

48
00:02:53,240 --> 00:02:56,160
Speaker 2: the news this week for apparently they're now laying of

49
00:02:56,280 --> 00:02:58,960
Speaker 2: people to try recoup the amount of investment that's been

50
00:02:58,960 --> 00:03:01,960
Speaker 2: made in AI. But we now see Google, Open Ai

51
00:03:02,320 --> 00:03:05,040
Speaker 2: and some of their peers, the tech workers asking for

52
00:03:05,200 --> 00:03:08,920
Speaker 2: clearer limits and how employers their employers are going to

53
00:03:09,000 --> 00:03:12,840
Speaker 2: work with the military. And this reminds me of a conversation.

54
00:03:13,000 --> 00:03:14,800
Speaker 2: I mean, it's a conversation we have all the time,

55
00:03:15,120 --> 00:03:18,680
Speaker 2: but this idea of what are the guardrails, what are

56
00:03:18,680 --> 00:03:21,840
Speaker 2: the limits, what are the things we are I guess

57
00:03:21,919 --> 00:03:25,919
Speaker 2: okay with AI doing. Andthropic didn't want Claude being used

58
00:03:26,160 --> 00:03:28,240
Speaker 2: in this particular war, so that was kind of a

59
00:03:28,280 --> 00:03:31,800
Speaker 2: line for them. And now we're seeing from employees there

60
00:03:31,840 --> 00:03:34,280
Speaker 2: seems to be a line for them too, or they're

61
00:03:34,320 --> 00:03:38,720
Speaker 2: asking employers to be clear about what the line is exactly.

62
00:03:38,800 --> 00:03:44,680
Speaker 3: And so a few employees from Google and open ai

63
00:03:45,320 --> 00:03:49,360
Speaker 3: have come out arging the organizations to have limits on

64
00:03:49,480 --> 00:03:56,280
Speaker 3: how they leverage their platforms for military AI contracts and

65
00:03:56,440 --> 00:04:00,680
Speaker 3: also just how they engage the United States Department of Defense.

66
00:04:01,520 --> 00:04:06,440
Speaker 3: But beyond just the blacklisting of Anthropic, an Iranian linked

67
00:04:06,480 --> 00:04:10,920
Speaker 3: media came out and stated that Iran is planning to

68
00:04:11,000 --> 00:04:16,360
Speaker 3: hit Google offices. It's now planning to expand its view

69
00:04:16,560 --> 00:04:21,800
Speaker 3: of conflict based on the fact that it feels that

70
00:04:22,320 --> 00:04:26,760
Speaker 3: there is an infrastructure warfare involved because now the United

71
00:04:26,760 --> 00:04:33,560
Speaker 3: States is leveraging some of the tech enabled technologies from

72
00:04:33,560 --> 00:04:37,200
Speaker 3: these companies and among its targets, it's planning to hit

73
00:04:37,240 --> 00:04:41,960
Speaker 3: the likes of Google, Microsoft, Palanter, IBM, and Video and Oracle.

74
00:04:43,120 --> 00:04:46,560
Speaker 1: And this is why more than nine hundred.

75
00:04:46,200 --> 00:04:51,760
Speaker 3: Employees from Google and OpenAI have through a letter titled

76
00:04:51,760 --> 00:04:54,080
Speaker 3: will not be Divided. I've signed a letter, you know,

77
00:04:54,240 --> 00:04:58,960
Speaker 3: urging the employees to actually revisit the contracts. Over eight

78
00:04:59,040 --> 00:05:03,839
Speaker 3: hundred of the signatory risk came from Google. We're also

79
00:05:03,920 --> 00:05:09,280
Speaker 3: told that, you know, companies like Data Breaks. Employees from

80
00:05:09,320 --> 00:05:12,440
Speaker 3: companies like Data Breaks, IBM, and Cells have also signed

81
00:05:12,440 --> 00:05:16,600
Speaker 3: another letter because of the implication of now them being

82
00:05:16,600 --> 00:05:21,680
Speaker 3: at risk their lives because of the US's involvement and

83
00:05:21,800 --> 00:05:23,039
Speaker 3: leveraging their technologies.

84
00:05:23,200 --> 00:05:26,279
Speaker 2: Yeah, and it's interesting that part of the letter reads

85
00:05:26,360 --> 00:05:29,839
Speaker 2: they're trying to divide each company with fear that the

86
00:05:29,920 --> 00:05:32,440
Speaker 2: other will give in. And it was interesting the week

87
00:05:32,560 --> 00:05:36,760
Speaker 2: that Anthropic said they didn't want their technology being used

88
00:05:36,880 --> 00:05:39,480
Speaker 2: in combat or military operations. I think it was open

89
00:05:39,520 --> 00:05:43,359
Speaker 2: ai who said, well, we're willing to speak, and so

90
00:05:43,400 --> 00:05:45,960
Speaker 2: it's interesting that the letter kind of references that, it

91
00:05:46,040 --> 00:05:48,560
Speaker 2: says the strategy only works if none of us know

92
00:05:48,600 --> 00:05:51,360
Speaker 2: where the other stands. The letter serves to create shed

93
00:05:51,440 --> 00:05:55,360
Speaker 2: understanding and solidarity in the face of this pressure from

94
00:05:55,360 --> 00:05:58,560
Speaker 2: the Department of War. So this kind of idea that

95
00:05:58,880 --> 00:06:03,839
Speaker 2: currently each company, each kind of this giant are working

96
00:06:03,880 --> 00:06:07,239
Speaker 2: in secrets. So Google doesn't know what open ai is doing.

97
00:06:07,360 --> 00:06:10,240
Speaker 2: We kind of don't know where Anthropic stands. But for

98
00:06:10,279 --> 00:06:14,400
Speaker 2: the most part it's quite murky, which works beautifully for

99
00:06:14,480 --> 00:06:17,200
Speaker 2: the US Department of War because they can kind of

100
00:06:17,640 --> 00:06:19,840
Speaker 2: exploit that you don't know what the other one is doing,

101
00:06:19,839 --> 00:06:21,920
Speaker 2: you don't know what your competitors doing exactly.

102
00:06:21,960 --> 00:06:25,520
Speaker 3: And I think one of the factors a lot of

103
00:06:25,600 --> 00:06:30,239
Speaker 3: technology companies and even startups must now be cautious against

104
00:06:30,400 --> 00:06:33,800
Speaker 3: is the divide and conquer strategy which the United States

105
00:06:33,960 --> 00:06:37,920
Speaker 3: government is using, and there's an element of the US

106
00:06:38,000 --> 00:06:39,440
Speaker 3: playing labs against.

107
00:06:39,160 --> 00:06:40,440
Speaker 1: Each other AI labs.

108
00:06:40,960 --> 00:06:45,760
Speaker 3: Obviously, with Anthropic being blacklisted, we saw open ai jump

109
00:06:45,800 --> 00:06:50,440
Speaker 3: to the opportunity. We're now told Google's also you know,

110
00:06:50,720 --> 00:06:54,600
Speaker 3: planning to use its Gemini AI within classified government systems.

111
00:06:55,400 --> 00:07:00,000
Speaker 3: And it's not the first time Google has actually followed

112
00:07:00,120 --> 00:07:04,240
Speaker 3: into controversy with these employees being against these types of initiatives.

113
00:07:04,240 --> 00:07:09,320
Speaker 3: In twenty eighteen, employees came out against using some of

114
00:07:09,360 --> 00:07:14,760
Speaker 3: the dronts developed by AI against you know, mass surveillance

115
00:07:15,800 --> 00:07:19,720
Speaker 3: in what was called Project Marvin. And this is concerning

116
00:07:19,760 --> 00:07:23,720
Speaker 3: because beyond just them being involved in this conflict, there's

117
00:07:23,760 --> 00:07:28,400
Speaker 3: also social and ethical aspects of it that employees want

118
00:07:28,520 --> 00:07:29,880
Speaker 3: organizations to consider.

119
00:07:30,160 --> 00:07:34,080
Speaker 2: Yeah, and so it's interesting here that workers now are

120
00:07:34,120 --> 00:07:39,680
Speaker 2: asking for greater clarity from their bosses on what kind

121
00:07:39,720 --> 00:07:42,880
Speaker 2: of work are we doing with the government, and especially

122
00:07:42,960 --> 00:07:48,560
Speaker 2: with regards to cloud and artificial intelligence contract So I

123
00:07:48,600 --> 00:07:51,840
Speaker 2: guess in a way, workers are saying, what are we

124
00:07:52,000 --> 00:07:55,960
Speaker 2: working on because they don't know at the moment what's

125
00:07:55,960 --> 00:07:59,040
Speaker 2: been agreed on a kind of a senior higher level

126
00:07:59,320 --> 00:08:02,640
Speaker 2: with government contrac But now they're asking for that that

127
00:08:02,800 --> 00:08:07,440
Speaker 2: transparency that well, what have you agree to that we essentially.

128
00:08:07,040 --> 00:08:09,760
Speaker 1: Are all now responsible for exactly.

129
00:08:09,840 --> 00:08:14,320
Speaker 3: And there's a couple of things here. One is employees

130
00:08:14,400 --> 00:08:19,400
Speaker 3: want non negotiable red lines around. For example, they should

131
00:08:19,400 --> 00:08:23,480
Speaker 3: not be involved in developing tools that will allow autonomous

132
00:08:23,480 --> 00:08:29,880
Speaker 3: weapons or mass surveillance. They do feel that introduces risks

133
00:08:30,120 --> 00:08:36,720
Speaker 3: or non ethical practices society. But also we're seeing a

134
00:08:36,760 --> 00:08:40,440
Speaker 3: new trend well and someone called it the anthropic catalyst,

135
00:08:41,200 --> 00:08:44,480
Speaker 3: where employees are now quitting the high paying jobs at

136
00:08:44,520 --> 00:08:48,120
Speaker 3: tech companies because of this conflict.

137
00:08:48,160 --> 00:08:48,959
Speaker 1: We're also.

138
00:08:50,720 --> 00:08:54,000
Speaker 3: They just don't want to be involved in creating tools

139
00:08:54,000 --> 00:08:57,160
Speaker 3: that I'm making the world a worse place to be in.

140
00:08:57,320 --> 00:09:02,760
Speaker 3: We saw after anthropics controversy and open AI taking this

141
00:09:03,360 --> 00:09:07,960
Speaker 3: new opportunity, the open head of Hardware, her name is Caitlin,

142
00:09:08,040 --> 00:09:11,960
Speaker 3: actually resigned because she did not want to be involved.

143
00:09:12,520 --> 00:09:17,520
Speaker 3: And I think there's a major discrepancy on what employees feel.

144
00:09:17,600 --> 00:09:21,000
Speaker 3: There is a lack of the balance between leadership and

145
00:09:21,160 --> 00:09:25,560
Speaker 3: workforce values. Values determine the direction and the beliefs of

146
00:09:25,600 --> 00:09:30,120
Speaker 3: a company, and if companies are going to jump into

147
00:09:30,160 --> 00:09:34,040
Speaker 3: opportunities which employees feel are introducing dangers to society with

148
00:09:34,080 --> 00:09:37,520
Speaker 3: the work they do, they do feel it's beyond what

149
00:09:37,640 --> 00:09:41,840
Speaker 3: they believe in. But lastly, it's just the issue of

150
00:09:41,880 --> 00:09:46,760
Speaker 3: transparency and accountability of them being involved in developing capabilities

151
00:09:47,160 --> 00:09:50,360
Speaker 3: are now being used in warfare other than being used

152
00:09:50,400 --> 00:09:53,720
Speaker 3: for developing or creating a better world. It's a concern

153
00:09:53,760 --> 00:09:56,800
Speaker 3: to them, and that's why we're in this position today.

154
00:09:57,040 --> 00:10:00,400
Speaker 2: It also seems that in addition to wanting their or

155
00:10:00,440 --> 00:10:03,880
Speaker 2: their employers to be transparent about what the redlines are

156
00:10:03,960 --> 00:10:06,640
Speaker 2: and to clarify those, there seems to also be an

157
00:10:06,640 --> 00:10:11,280
Speaker 2: appeal to Congress, so to lawmakers to examine whether the

158
00:10:11,440 --> 00:10:16,160
Speaker 2: use of these extraordinary authorities against an American technology company

159
00:10:16,679 --> 00:10:20,480
Speaker 2: is appropriate. Of course, so the use of the Department

160
00:10:20,559 --> 00:10:24,960
Speaker 2: of Defense saying anthropical as a supply chain risk. Now

161
00:10:25,000 --> 00:10:31,199
Speaker 2: there's a question about how does how are lawmakers going

162
00:10:31,240 --> 00:10:34,600
Speaker 2: to deal with this and should companies face retaliation from

163
00:10:34,600 --> 00:10:36,960
Speaker 2: the state if they say, well, we don't want our

164
00:10:36,960 --> 00:10:40,080
Speaker 2: technology to be used in war? And so here there's

165
00:10:40,160 --> 00:10:43,920
Speaker 2: now kind of regulators being brought in as well, and

166
00:10:43,960 --> 00:10:49,200
Speaker 2: how they manage the situation and how we treat governments

167
00:10:49,920 --> 00:10:52,760
Speaker 2: departments are wanting to use AI and if the company

168
00:10:52,800 --> 00:10:55,920
Speaker 2: says no, well should they be blacklisted for it?

169
00:10:56,400 --> 00:10:56,840
Speaker 1: Exactly?

170
00:10:57,160 --> 00:11:00,400
Speaker 3: But you see in our previous conversation we talked about

171
00:11:00,559 --> 00:11:04,760
Speaker 3: engineering safety, right, but a bigger part is these are

172
00:11:04,800 --> 00:11:11,160
Speaker 3: things organizations technology companies should think about before developing capabilities.

173
00:11:12,280 --> 00:11:16,840
Speaker 3: Authorities like the United States Department of Defense. Governments have

174
00:11:17,520 --> 00:11:22,560
Speaker 3: come up with ways of working around these requirements the regulations,

175
00:11:23,160 --> 00:11:25,959
Speaker 3: and in this instance, we're learning that even with the

176
00:11:26,000 --> 00:11:29,200
Speaker 3: regulations and rules, they might not mean anything when it

177
00:11:29,240 --> 00:11:32,240
Speaker 3: comes to laws and those who control their law. And

178
00:11:33,880 --> 00:11:38,719
Speaker 3: maybe it reminds me when those petitions signed by some

179
00:11:38,800 --> 00:11:43,600
Speaker 3: of you know, the subject matter experts and leading minds

180
00:11:43,720 --> 00:11:46,480
Speaker 3: like you know, Jeoff Hinton, is like that technology was

181
00:11:46,520 --> 00:11:49,160
Speaker 3: actually moving so much faster than we could what we

182
00:11:49,320 --> 00:11:53,719
Speaker 3: prepared for. Because the fact that the capability exists, it

183
00:11:53,760 --> 00:11:58,000
Speaker 3: can be used for anything, and therefore we maybe we

184
00:11:58,040 --> 00:12:01,120
Speaker 3: shouldn't develop and mature it so much that it can

185
00:12:01,200 --> 00:12:04,400
Speaker 3: be used for anything, right, And that is the only

186
00:12:04,480 --> 00:12:06,280
Speaker 3: way to guarantee safety.

187
00:12:06,840 --> 00:12:09,400
Speaker 2: And I mean even with that where we talk about

188
00:12:09,440 --> 00:12:11,719
Speaker 2: sort of how we kind of plan.

189
00:12:11,800 --> 00:12:13,199
Speaker 1: Or engineer safety.

190
00:12:14,520 --> 00:12:18,200
Speaker 2: I was reading something about from Google's chief scientists and

191
00:12:18,200 --> 00:12:21,760
Speaker 2: he was talking about surveillance systems are just so prone

192
00:12:21,800 --> 00:12:26,720
Speaker 2: for misuse for political reasons. So what And there is

193
00:12:26,760 --> 00:12:29,599
Speaker 2: this idea that technologies are often developed elsewhere in the

194
00:12:29,640 --> 00:12:31,840
Speaker 2: world and it is brought home. So if it is

195
00:12:31,880 --> 00:12:34,520
Speaker 2: being used for surveillance in Iraq, or Iran or whatever,

196
00:12:34,760 --> 00:12:38,400
Speaker 2: it will invariably be brought home to the US, and

197
00:12:38,440 --> 00:12:40,680
Speaker 2: so he was raising that as kind of as a

198
00:12:40,760 --> 00:12:46,680
Speaker 2: concern that these technologies have this ability to limit freedom

199
00:12:46,720 --> 00:12:50,960
Speaker 2: of speech, crush political discourse, and we should be a

200
00:12:50,960 --> 00:12:54,439
Speaker 2: little bit concerned about how these technologies are being used,

201
00:12:54,520 --> 00:12:55,599
Speaker 2: especially in conflicts.

202
00:12:56,120 --> 00:13:01,359
Speaker 3: And that's a big concern because the pioneers of technology,

203
00:13:01,440 --> 00:13:06,400
Speaker 3: especially what's happening with AI today, are developing capabilities while

204
00:13:06,480 --> 00:13:11,200
Speaker 3: speculating where they'll be used, and governments control the laws.

205
00:13:11,280 --> 00:13:13,959
Speaker 3: Some of these governments have got subject matter experts who

206
00:13:14,040 --> 00:13:16,680
Speaker 3: understand what else it could be used for to make

207
00:13:17,080 --> 00:13:21,080
Speaker 3: government's decisioning much easier, or even make governments more powerful,

208
00:13:21,160 --> 00:13:24,959
Speaker 3: like what we're seeing today the war that is happening

209
00:13:25,000 --> 00:13:30,720
Speaker 3: in the Middle East. And that's why it's important that

210
00:13:31,120 --> 00:13:34,200
Speaker 3: before we mature these capabilities, let's think about what could

211
00:13:34,240 --> 00:13:38,720
Speaker 3: go wrong, the dangers, and hold back until there's clear

212
00:13:39,160 --> 00:13:45,960
Speaker 3: clarity across various sectors or even leadership spectrum, or else

213
00:13:46,000 --> 00:13:48,400
Speaker 3: we're going to find ourselves where there's a conflict now.

214
00:13:48,440 --> 00:13:53,920
Speaker 3: For example, America is using its laws to force organizations

215
00:13:54,120 --> 00:13:56,679
Speaker 3: to ensure that the technology is being used for whatever

216
00:13:56,720 --> 00:14:00,000
Speaker 3: purpose they want to and that is a very very

217
00:14:00,160 --> 00:14:02,840
Speaker 3: big issue we're going to see going forward, and it

218
00:14:02,880 --> 00:14:05,280
Speaker 3: will start spreading across to other countries.

219
00:14:05,400 --> 00:14:05,920
Speaker 1: Yeah.

220
00:14:06,000 --> 00:14:10,479
Speaker 3: So while AI startups are very proud that they're developing technology,

221
00:14:10,520 --> 00:14:14,400
Speaker 3: we might see those technologies being usedful other reasons than

222
00:14:14,480 --> 00:14:15,320
Speaker 3: what they were meant for.

223
00:14:15,520 --> 00:14:16,360
Speaker 1: Yeah. Yes.

224
00:14:16,920 --> 00:14:21,680
Speaker 2: So the other story this week is well certainly started

225
00:14:21,760 --> 00:14:24,520
Speaker 2: last week is a we're getting a better sense of

226
00:14:24,560 --> 00:14:30,040
Speaker 2: how the Meta ray ban AI glasses are working. And

227
00:14:30,440 --> 00:14:34,800
Speaker 2: actually the UK Data Watchdog has written to Meta excuse me,

228
00:14:35,960 --> 00:14:39,440
Speaker 2: saying they are concerned that outsourced workers were able to

229
00:14:39,600 --> 00:14:45,080
Speaker 2: view sensitive content filmed by the company's AI smart classes,

230
00:14:45,680 --> 00:14:49,520
Speaker 2: and they're also concerned about what these actual humans because

231
00:14:49,560 --> 00:14:53,440
Speaker 2: they are people reviewing this content. They are concerns at

232
00:14:53,440 --> 00:14:57,920
Speaker 2: what these people are actually seeing when they review these

233
00:14:57,960 --> 00:14:59,320
Speaker 2: glasses thoughts.

234
00:15:01,120 --> 00:15:07,440
Speaker 3: So, for the listener's purpose, Meta collects data and one

235
00:15:07,520 --> 00:15:10,640
Speaker 3: channel it chooses to collect data is its meta RABE

236
00:15:10,720 --> 00:15:14,840
Speaker 3: and glasses, right, and it uses that data to train

237
00:15:14,920 --> 00:15:19,120
Speaker 3: AI models whose output is used to improve it customer

238
00:15:19,200 --> 00:15:26,920
Speaker 3: experience or even their ecosystem. But to Swedish newspapers broke

239
00:15:27,000 --> 00:15:30,440
Speaker 3: up the story and reported that you know, in the

240
00:15:30,520 --> 00:15:35,440
Speaker 3: process of actually developing AI systems, Meta has to ensure

241
00:15:35,440 --> 00:15:40,040
Speaker 3: that the data or the videos that are collected have

242
00:15:40,240 --> 00:15:43,120
Speaker 3: a representation of the ground truth because that data is

243
00:15:43,120 --> 00:15:49,920
Speaker 3: not taged properly, and this process of adding descriptive data

244
00:15:50,040 --> 00:15:53,160
Speaker 3: or tags or visual markers is actually done by human beings.

245
00:15:53,720 --> 00:15:57,680
Speaker 3: And Meta has outsourced its outsourced this to a company

246
00:15:57,680 --> 00:16:02,160
Speaker 3: in Kenya that you mentioned, called Summer. But what came

247
00:16:02,200 --> 00:16:06,080
Speaker 3: out is that it came out some of the data

248
00:16:06,120 --> 00:16:10,000
Speaker 3: that is being collected is you know, confidential data. Later

249
00:16:10,080 --> 00:16:14,760
Speaker 3: in the bathrooms, nudity, intimate data, date off bank accounts

250
00:16:15,400 --> 00:16:20,080
Speaker 3: which were collected without users being aware of. And when

251
00:16:20,200 --> 00:16:23,560
Speaker 3: Metal was asked, it gave two reasons that are not

252
00:16:23,680 --> 00:16:29,320
Speaker 3: very convincing. One that it reserves the right to send users,

253
00:16:30,640 --> 00:16:35,800
Speaker 3: you know, interaction with its AI platforms or capabilities.

254
00:16:36,320 --> 00:16:37,400
Speaker 1: But also.

255
00:16:38,640 --> 00:16:42,640
Speaker 3: It Meta promised that it blored or masked some of

256
00:16:42,680 --> 00:16:47,400
Speaker 3: the data though the people who are doing the annotation

257
00:16:47,520 --> 00:16:51,080
Speaker 3: of data were actually able to see this confidential data.

258
00:16:50,800 --> 00:16:55,000
Speaker 2: Right, And so two things are happening now. One users

259
00:16:55,040 --> 00:16:58,960
Speaker 2: of the classes of finding out that the stuff that

260
00:16:59,000 --> 00:17:03,680
Speaker 2: they were seeing was not confidential, so it hadn't been blurred,

261
00:17:03,720 --> 00:17:06,680
Speaker 2: it hadn't so your bank card, you on the toilet,

262
00:17:07,040 --> 00:17:09,640
Speaker 2: you being intimate with someone, all of those things were

263
00:17:09,680 --> 00:17:11,919
Speaker 2: being kind of captured as is.

264
00:17:12,240 --> 00:17:13,560
Speaker 1: That's the one issue.

265
00:17:13,720 --> 00:17:17,359
Speaker 2: The other is now the group of workers who have

266
00:17:17,520 --> 00:17:20,199
Speaker 2: to what did you call it, data labeling? Who do

267
00:17:20,280 --> 00:17:25,200
Speaker 2: this data labeling, they're now apparently looking to put together

268
00:17:25,240 --> 00:17:29,240
Speaker 2: a class action lawsuit against the company because they feel

269
00:17:29,240 --> 00:17:32,119
Speaker 2: as though they've been exploited, they've been forced to review

270
00:17:32,480 --> 00:17:36,919
Speaker 2: traumatic content without proper working conditions. One person told the

271
00:17:36,920 --> 00:17:40,439
Speaker 2: Swedish publications. You understand that it's someone's private life that

272
00:17:40,480 --> 00:17:42,480
Speaker 2: you are looking at, but at the same time you

273
00:17:42,520 --> 00:17:45,080
Speaker 2: were just expected to carry out the work. You're not

274
00:17:45,119 --> 00:17:48,439
Speaker 2: supposed to question it. If you start asking questions, you

275
00:17:48,480 --> 00:17:51,840
Speaker 2: are gone so interesting that we're seeing kind of two

276
00:17:51,920 --> 00:17:55,800
Speaker 2: issues there, one for the user of the technology and

277
00:17:55,960 --> 00:17:58,879
Speaker 2: be for the person that has to I guess moderate

278
00:17:58,920 --> 00:18:04,760
Speaker 2: it label it could make sense. We're seeing two groups

279
00:18:05,359 --> 00:18:09,080
Speaker 2: having an issue with the use of this particular technology exactly.

280
00:18:09,119 --> 00:18:11,679
Speaker 3: And for those of us who use a lot of

281
00:18:12,320 --> 00:18:15,720
Speaker 3: interact on platforms and use them for various reasons, this

282
00:18:16,000 --> 00:18:20,040
Speaker 3: provides a lot of lessons. The first lesson is that

283
00:18:20,480 --> 00:18:26,440
Speaker 3: we normally assume that the data that is collected on

284
00:18:26,520 --> 00:18:32,359
Speaker 3: these platforms, it's just processed directly by AI systems. There's

285
00:18:32,440 --> 00:18:35,120
Speaker 3: always a human on their loop to ensure that AI

286
00:18:35,240 --> 00:18:38,359
Speaker 3: models are trained, and this is through this process of

287
00:18:38,440 --> 00:18:43,679
Speaker 3: data leveling. People are involved, so they're likely to be

288
00:18:43,800 --> 00:18:47,560
Speaker 3: seeing these data, whether it's confidential or private or you know.

289
00:18:48,000 --> 00:18:48,960
Speaker 1: The second part.

290
00:18:48,840 --> 00:18:54,680
Speaker 3: Is we're seeing a failure in technical safeguards. And then

291
00:18:54,720 --> 00:18:57,080
Speaker 3: the first one is well met to say is that

292
00:18:57,240 --> 00:19:00,760
Speaker 3: it's blurring faces or confidential eye speaks of data. It's

293
00:19:00,760 --> 00:19:04,120
Speaker 3: actually not doing that. The second part is we must

294
00:19:04,119 --> 00:19:09,000
Speaker 3: store that confidential data is always at risk because this

295
00:19:09,240 --> 00:19:13,400
Speaker 3: data once is collected through the rayburn glasses or other

296
00:19:13,560 --> 00:19:15,919
Speaker 3: channels or metal platform, it has to be moved to

297
00:19:15,960 --> 00:19:18,879
Speaker 3: the cloud and then from the cloud sent to Kenya

298
00:19:18,920 --> 00:19:19,439
Speaker 3: for labeling.

299
00:19:20,400 --> 00:19:22,000
Speaker 1: That's already a risk by itself.

300
00:19:22,920 --> 00:19:27,159
Speaker 3: Then the one that is more concerning is there's a

301
00:19:27,240 --> 00:19:31,880
Speaker 3: consent gap in the way technology companies collect data and

302
00:19:32,119 --> 00:19:36,600
Speaker 3: drive interaction, in the sense that there is so much

303
00:19:36,680 --> 00:19:39,120
Speaker 3: he didn't find print that people just accept he didn't

304
00:19:39,160 --> 00:19:42,920
Speaker 3: find print, and they don't think about it. This he

305
00:19:42,960 --> 00:19:46,840
Speaker 3: didn't find print is buried in legal jargon that people

306
00:19:46,920 --> 00:19:50,439
Speaker 3: don't understand and they just accept it.

307
00:19:50,440 --> 00:19:51,800
Speaker 1: Because they don't want to spend on it.

308
00:19:52,400 --> 00:19:55,320
Speaker 3: And then there's an element of bystandard privacy in the

309
00:19:55,359 --> 00:19:59,199
Speaker 3: sense that, so think about Rayburn glasses, a lot of

310
00:19:59,240 --> 00:20:02,560
Speaker 3: people who are not even owning their gadgets will be

311
00:20:02,680 --> 00:20:06,200
Speaker 3: part of the content without actually providing permission.

312
00:20:05,960 --> 00:20:10,040
Speaker 1: Right, so someone is wearing them, someone wearing them them.

313
00:20:11,240 --> 00:20:14,000
Speaker 2: And so I could end up in the content exactly,

314
00:20:14,040 --> 00:20:17,280
Speaker 2: and that content that you say is recorded by the glasses,

315
00:20:17,560 --> 00:20:20,159
Speaker 2: it's stored on a cloud. It is then sent to

316
00:20:20,720 --> 00:20:24,040
Speaker 2: I guess the third party, the workers in Kenya who

317
00:20:24,040 --> 00:20:27,520
Speaker 2: are reviewing this content exactly, And so then there are

318
00:20:27,560 --> 00:20:32,199
Speaker 2: new issues as well of privacy, not just for the

319
00:20:32,280 --> 00:20:33,160
Speaker 2: person wearing the.

320
00:20:33,119 --> 00:20:36,840
Speaker 1: Glasses, for people all of us exactly.

321
00:20:36,400 --> 00:20:39,640
Speaker 2: Because I haven't contented to being recorded by the glasses.

322
00:20:39,680 --> 00:20:42,880
Speaker 2: But there was something about how the newer ones are

323
00:20:43,000 --> 00:20:46,840
Speaker 2: almost always on, they're recording all the time, and so

324
00:20:47,240 --> 00:20:50,560
Speaker 2: what does that mean for I guess things like you know,

325
00:20:50,880 --> 00:20:52,800
Speaker 2: people who are in public, You're going about your day,

326
00:20:52,840 --> 00:20:55,280
Speaker 2: you're doing your groceries, you're at the gym, you're at

327
00:20:55,280 --> 00:20:58,480
Speaker 2: a party, whatever it is, and now you are part

328
00:20:58,600 --> 00:21:01,000
Speaker 2: of this because this content and also being used to

329
00:21:01,080 --> 00:21:03,240
Speaker 2: train exactly matters AI.

330
00:21:03,400 --> 00:21:06,640
Speaker 1: Exactly so, and Meta wants to have more data.

331
00:21:07,320 --> 00:21:09,600
Speaker 3: The more data they have, the more their AI models

332
00:21:09,640 --> 00:21:16,080
Speaker 3: are powerful. And this is we're now seeing gaps in regulation.

333
00:21:16,840 --> 00:21:19,359
Speaker 3: Do we introduce the right of individuals who are not

334
00:21:19,560 --> 00:21:21,720
Speaker 3: using their devices and around them? And how do we

335
00:21:21,800 --> 00:21:24,200
Speaker 3: go about it to ensure that this is embedded in

336
00:21:24,320 --> 00:21:28,879
Speaker 3: the way technologies are developed. But also you see it

337
00:21:29,320 --> 00:21:33,080
Speaker 3: shows you how difficult it is to embed regulation and

338
00:21:33,119 --> 00:21:37,760
Speaker 3: to hold technology companies accountable for this. And maybe the

339
00:21:37,840 --> 00:21:41,679
Speaker 3: advice for users is that you need to limit and

340
00:21:41,800 --> 00:21:46,200
Speaker 3: monitor your involvement on these platforms. Don't just get excited.

341
00:21:46,320 --> 00:21:50,359
Speaker 3: Don't just you know, apploy your images or take things

342
00:21:50,400 --> 00:21:54,080
Speaker 3: around you with a sense of excitement. The other party

343
00:21:54,160 --> 00:21:57,320
Speaker 3: is try to make sure you don't forget the manual

344
00:21:57,400 --> 00:22:02,639
Speaker 3: control of this technology, switch off the camera. And maybe

345
00:22:02,720 --> 00:22:08,280
Speaker 3: regulators should hold these technology companies accountable for introducing minor

346
00:22:08,400 --> 00:22:11,200
Speaker 3: controls in every aspect of AI capability.

347
00:22:11,480 --> 00:22:15,480
Speaker 2: Right, And is it that also that regulators just don't

348
00:22:15,600 --> 00:22:22,119
Speaker 2: move as quickly as developers of tech. Lawmaking is a

349
00:22:22,640 --> 00:22:29,280
Speaker 2: slow process whereas technology is moving so quickly. Is that

350
00:22:29,320 --> 00:22:32,040
Speaker 2: also one of the challenges. It's just like the regulators

351
00:22:32,080 --> 00:22:34,200
Speaker 2: who are meant to be managing this process.

352
00:22:34,800 --> 00:22:38,400
Speaker 1: Are just they can't keep up exactly. And I think.

353
00:22:40,520 --> 00:22:43,840
Speaker 3: Lawmakers today regulators can change the way they do things.

354
00:22:45,119 --> 00:22:48,520
Speaker 3: Some of the regulators are well equipped with skills, but

355
00:22:48,560 --> 00:22:52,000
Speaker 3: I think they need to instead of playing just an

356
00:22:52,040 --> 00:22:55,760
Speaker 3: oversight role, they should start call creating these technologies with

357
00:22:55,840 --> 00:23:02,200
Speaker 3: technology companies, not just wait and technologies advanced and by

358
00:23:02,240 --> 00:23:04,760
Speaker 3: the time something goes wrong, it's too late, like we're

359
00:23:04,760 --> 00:23:08,560
Speaker 3: finding out that you know about this. You know, the

360
00:23:08,600 --> 00:23:12,439
Speaker 3: meta raban was launched in twenty twenty three, only finding

361
00:23:12,480 --> 00:23:16,120
Speaker 3: out three years later, almost three years later, of some

362
00:23:16,200 --> 00:23:22,320
Speaker 3: of these gaps in the in being ethical and therefore

363
00:23:22,359 --> 00:23:27,679
Speaker 3: core creating testing for embedment of ethical aspects of creation

364
00:23:28,560 --> 00:23:31,639
Speaker 3: is better than you know, having SEF guards. Self guards

365
00:23:31,680 --> 00:23:35,680
Speaker 3: do not actually prevent dangers of AI. They actually catch

366
00:23:35,720 --> 00:23:38,919
Speaker 3: the dangers if they actually catch them. But some of

367
00:23:38,960 --> 00:23:42,720
Speaker 3: the dangers, like privacy, destroy people's lives. Like we said,

368
00:23:42,760 --> 00:23:45,640
Speaker 3: you know, the type of data we're talking about is

369
00:23:45,680 --> 00:23:50,080
Speaker 3: not just there for AI, it's actually private life. It's

370
00:23:50,119 --> 00:23:55,000
Speaker 3: people bank accounts, it's people's private intimate moments. Now it's

371
00:23:55,040 --> 00:23:58,800
Speaker 3: being used to train algorithms for personal experiences. The whole

372
00:23:58,800 --> 00:24:01,320
Speaker 3: world is seeing them, and we don't know who else

373
00:24:01,400 --> 00:24:02,880
Speaker 3: you have access to that information.

374
00:24:03,119 --> 00:24:07,000
Speaker 2: Yeah, it does seem though that the idea of privacy

375
00:24:07,560 --> 00:24:10,600
Speaker 2: and what's left of it is kind of the big

376
00:24:10,960 --> 00:24:13,800
Speaker 2: question with regards. Initially, it was a big question with

377
00:24:13,880 --> 00:24:16,760
Speaker 2: social media people. There are people who are convinced their

378
00:24:16,800 --> 00:24:21,200
Speaker 2: phones are listening to them. Right, Why is my algorithm

379
00:24:21,320 --> 00:24:23,679
Speaker 2: or my timeline so accurate? I was talking about these

380
00:24:23,680 --> 00:24:26,399
Speaker 2: issues and now they're on my timeline. It was a

381
00:24:26,440 --> 00:24:29,119
Speaker 2: bit of a concern with social media. It looks as

382
00:24:29,119 --> 00:24:32,280
Speaker 2: though it will be an even bigger concern with AI.

383
00:24:32,600 --> 00:24:37,119
Speaker 2: This idea of what is private, what happens to that information,

384
00:24:37,560 --> 00:24:40,480
Speaker 2: the idea that you shouldn't be asking AI legal medical

385
00:24:40,560 --> 00:24:45,680
Speaker 2: questions because you know that's a privacy issue, and also

386
00:24:45,720 --> 00:24:48,959
Speaker 2: if you can admit to a crime or wrongdoing. Some

387
00:24:49,040 --> 00:24:51,159
Speaker 2: of these companies have told you, well, if the cops

388
00:24:51,200 --> 00:24:54,000
Speaker 2: ask for the information, will give it to them.

389
00:24:54,160 --> 00:24:58,200
Speaker 1: Right. So, this idea of what of yours, your.

390
00:24:57,960 --> 00:25:01,199
Speaker 2: Information, your data is private, it seems to be like

391
00:25:01,280 --> 00:25:05,200
Speaker 2: one of the big questions that will continue to be

392
00:25:06,080 --> 00:25:07,720
Speaker 2: a big issue as we developed the.

393
00:25:07,680 --> 00:25:13,200
Speaker 3: Tech exactly, and I think if you think about it, Unfortunately,

394
00:25:13,640 --> 00:25:17,879
Speaker 3: technology companies like Meta, Google Apple need this data for

395
00:25:17,920 --> 00:25:21,199
Speaker 3: them to be relevant to their value proposition, but we

396
00:25:21,440 --> 00:25:23,960
Speaker 3: enable them as well. I've always said, you know, you

397
00:25:24,000 --> 00:25:26,600
Speaker 3: need to clean your closet watch what you give to them.

398
00:25:27,200 --> 00:25:31,000
Speaker 3: But a more concerning point is that this type of

399
00:25:31,080 --> 00:25:34,840
Speaker 3: data can actually change the meaning of life. We remember

400
00:25:34,840 --> 00:25:38,239
Speaker 3: there was an example a few years ago where there

401
00:25:38,280 --> 00:25:42,160
Speaker 3: was a company called Cambridge Analytica that used Facebook data

402
00:25:43,160 --> 00:25:48,440
Speaker 3: to emotionally swaar people's decisions on elections. And this very

403
00:25:48,520 --> 00:25:52,120
Speaker 3: evidence that you know that is actually how breaks it happened,

404
00:25:52,760 --> 00:25:55,520
Speaker 3: where you know, emotional data was you know, things you

405
00:25:55,800 --> 00:25:58,480
Speaker 3: like on the platform was used to target you the

406
00:25:58,560 --> 00:26:01,320
Speaker 3: right time to swear your dish on who you'd vote for.

407
00:26:02,119 --> 00:26:04,159
Speaker 1: And this is just a build up on that.

408
00:26:05,440 --> 00:26:09,520
Speaker 3: This has been many years since the Cambridge Analytica incidents,

409
00:26:09,760 --> 00:26:12,120
Speaker 3: and it tells you that it's been a growing concern

410
00:26:12,200 --> 00:26:15,680
Speaker 3: even though Cambridge analytic It does not exist anymore as

411
00:26:15,680 --> 00:26:20,600
Speaker 3: a company. Because of those reasons, we need better ways

412
00:26:20,920 --> 00:26:24,080
Speaker 3: of driving this accountability and making sure that we don't

413
00:26:24,119 --> 00:26:25,720
Speaker 3: introduce dangers to society.

414
00:26:26,359 --> 00:26:29,040
Speaker 2: Yes, doctor Nacilla Ohway, is a great pleasure having you

415
00:26:29,080 --> 00:26:31,480
Speaker 2: on the show. Thank you so much for your time

416
00:26:31,560 --> 00:26:32,040
Speaker 2: this morning.

417
00:26:32,240 --> 00:26:35,800
Speaker 3: Thank you so much. Let's continue the conversation. We don't

418
00:26:35,800 --> 00:26:38,160
Speaker 3: know what will come up next to the AI, but yes,

419
00:26:38,240 --> 00:26:40,600
Speaker 3: that's why we're here. Yeah, thank you. Doctor.

420
00:26:40,720 --> 00:26:43,920
Speaker 2: That's chief Data and Analytics Officer at F and B.

421
00:26:44,400 --> 00:26:48,520
Speaker 2: RISK is also the author of African Artificial Intelligence as

422
00:26:48,560 --> 00:26:49,920
Speaker 2: well as Sovereign AI.

423
00:26:50,280 --> 00:26:51,560
Speaker 1: Doctor Mark Nacilla joining