1
00:00:03,120 --> 00:00:06,600
Speaker 1: Hello, Hello, Welcome to Smart Talks with IBM, a podcast

2
00:00:06,680 --> 00:00:13,000
Speaker 1: from Pushkin Industries, iHeartRadio and IBM. I'm Malcolm Gladwell. This season,

3
00:00:13,000 --> 00:00:16,159
Speaker 1: we're diving back into the world of artificial intelligence, but

4
00:00:16,239 --> 00:00:22,240
Speaker 1: with a focus on the powerful concept of open its possibilities, implications,

5
00:00:22,280 --> 00:00:26,320
Speaker 1: and misconceptions. We'll look at openness from a variety of

6
00:00:26,360 --> 00:00:30,200
Speaker 1: angles and explore how the concept is already reshaping industries,

7
00:00:30,640 --> 00:00:34,199
Speaker 1: ways of doing business and our very notion of what's possible.

8
00:00:35,320 --> 00:00:39,239
Speaker 1: In today's episode, Jacob Goldstein sits down with Rebecca Finley,

9
00:00:39,520 --> 00:00:43,720
Speaker 1: the CEO of the Partnership on Ai, a nonprofit group

10
00:00:44,000 --> 00:00:48,040
Speaker 1: grappling with important questions around the future of AI. Their

11
00:00:48,080 --> 00:00:52,120
Speaker 1: conversation focuses on Rebecca's work bringing together a community of

12
00:00:52,240 --> 00:00:58,000
Speaker 1: diverse stakeholders to help shape the conversation around accountable AI governance.

13
00:00:58,880 --> 00:01:02,960
Speaker 1: Rebecca explains why transparency is so crucial for scaling the

14
00:01:03,000 --> 00:01:07,280
Speaker 1: technology responsibly, and she highlights how working with groups like

15
00:01:07,319 --> 00:01:10,960
Speaker 1: the AI Alliance can provide valuable insights in order to

16
00:01:10,959 --> 00:01:17,560
Speaker 1: build the resources, infrastructure, and community around releasing open source models. So,

17
00:01:17,760 --> 00:01:21,000
Speaker 1: without further ado, let's get to that conversation.

18
00:01:28,120 --> 00:01:29,800
Speaker 2: Can you just say your name? And your job.

19
00:01:30,280 --> 00:01:33,480
Speaker 3: My name is Rebecca Finley. I am the CEO of

20
00:01:33,520 --> 00:01:37,240
Speaker 3: the Partnership on AI to Benefit People and Society, often

21
00:01:37,280 --> 00:01:39,160
Speaker 3: referred to as PAI.

22
00:01:40,080 --> 00:01:42,640
Speaker 2: How did you get here? What was your job before

23
00:01:42,680 --> 00:01:44,559
Speaker 2: you have the job that you have now?

24
00:01:45,480 --> 00:01:50,400
Speaker 3: I came to PAI about three years ago, having had

25
00:01:50,440 --> 00:01:55,640
Speaker 3: the opportunity to work for the Canadian Institute for Advance Research,

26
00:01:56,200 --> 00:02:00,920
Speaker 3: developing and deploying all of their programs related to the

27
00:02:00,960 --> 00:02:06,480
Speaker 3: intersection of technology and society. And one of the areas

28
00:02:06,720 --> 00:02:11,040
Speaker 3: that the Canadian Institute had been funding since nineteen eighty

29
00:02:11,080 --> 00:02:14,919
Speaker 3: two was research into artificial intelligence.

30
00:02:15,160 --> 00:02:17,160
Speaker 2: Wow early, they were early.

31
00:02:18,240 --> 00:02:22,480
Speaker 3: It was a very early commitment and an ongoing commitment

32
00:02:22,680 --> 00:02:28,040
Speaker 3: at the Institute to fund long term fundamental questions of

33
00:02:28,120 --> 00:02:37,040
Speaker 3: scientific importance in interdisciplinary research programs that were often committed

34
00:02:37,080 --> 00:02:40,600
Speaker 3: and funded to for well over a decade. The AI

35
00:02:40,800 --> 00:02:43,840
Speaker 3: Robotics and Society program that kicked off the work at

36
00:02:43,919 --> 00:02:49,640
Speaker 3: the Institute eventually became a program very much focused on

37
00:02:50,240 --> 00:02:55,360
Speaker 3: deep learning and reinforcement learning, neural networks. All of the

38
00:02:56,040 --> 00:03:00,560
Speaker 3: current iteration of AI, or certainly the pregenerative of AI

39
00:03:00,800 --> 00:03:05,640
Speaker 3: iteration of AI that led to this transformation that we've

40
00:03:05,680 --> 00:03:09,040
Speaker 3: seen in terms of online search and all sorts of

41
00:03:09,040 --> 00:03:12,040
Speaker 3: ways in which predictive AI has been deployed. So I

42
00:03:12,080 --> 00:03:15,400
Speaker 3: had the opportunity to see the very early days of

43
00:03:15,480 --> 00:03:19,919
Speaker 3: that research coming together, and when in the early sort

44
00:03:19,960 --> 00:03:25,360
Speaker 3: of two thousand, twenty and tens, when compute capability came

45
00:03:25,400 --> 00:03:29,960
Speaker 3: together with data capability through some of the Internet companies

46
00:03:29,960 --> 00:03:33,960
Speaker 3: and otherwise, and we really saw this technology start to

47
00:03:34,000 --> 00:03:37,000
Speaker 3: take off. I had the opportunity to start up a

48
00:03:37,040 --> 00:03:42,160
Speaker 3: program specifically focused on the impacts of AI in society.

49
00:03:42,840 --> 00:03:45,600
Speaker 3: There was, as you know, at that time, some concerns

50
00:03:45,640 --> 00:03:50,120
Speaker 3: both about the potential for the technology, but also in

51
00:03:50,240 --> 00:03:52,800
Speaker 3: terms of what we were seeing around data sets and

52
00:03:53,080 --> 00:03:57,360
Speaker 3: bias and discrimination and potential impact on future jobs. And

53
00:03:57,440 --> 00:04:01,600
Speaker 3: so bringing a whole group of experts, whether they were

54
00:04:01,640 --> 00:04:07,400
Speaker 3: ethicists or lawyers or economists sociologists into the discussion about

55
00:04:07,440 --> 00:04:10,880
Speaker 3: AI was core to that new program and continues to

56
00:04:10,920 --> 00:04:14,600
Speaker 3: be core to my commitment to bringing diverse perspectives together

57
00:04:14,680 --> 00:04:18,520
Speaker 3: to solve the challenges and opportunities that AI offers today.

58
00:04:19,600 --> 00:04:22,159
Speaker 2: So specifically, what is your job now? What is the

59
00:04:22,200 --> 00:04:24,240
Speaker 2: work you do? What is the work that PAI does?

60
00:04:25,320 --> 00:04:29,160
Speaker 3: I like to answer that question by asking two questions,

61
00:04:29,640 --> 00:04:33,000
Speaker 3: First and foremost, do you believe that the world is

62
00:04:33,240 --> 00:04:37,159
Speaker 3: more divided today than it ever has been in recent history?

63
00:04:38,040 --> 00:04:41,839
Speaker 3: And do you believe that if we don't create spaces

64
00:04:42,440 --> 00:04:45,760
Speaker 3: for very different perspectives to come together, we won't be

65
00:04:45,839 --> 00:04:48,599
Speaker 3: able to solve the challenges that are in front of

66
00:04:48,600 --> 00:04:52,560
Speaker 3: the world today. My answer to both of those questions is, yes,

67
00:04:53,040 --> 00:04:56,599
Speaker 3: we're more divided, and two, we need to seek out

68
00:04:56,760 --> 00:05:01,479
Speaker 3: those spaces where those very different perspectives can come together

69
00:05:02,000 --> 00:05:05,200
Speaker 3: to solve those great challenges. And that's what I get

70
00:05:05,240 --> 00:05:08,760
Speaker 3: to do as CEO of the Partnership on AI. We

71
00:05:08,760 --> 00:05:13,479
Speaker 3: were begun in twenty sixteen with a fundamental commitment to

72
00:05:13,720 --> 00:05:20,280
Speaker 3: bringing together experts, whether they were in industry, academia, civil society,

73
00:05:20,360 --> 00:05:24,600
Speaker 3: or philanthropy, coming together to identify what are the most

74
00:05:24,640 --> 00:05:28,359
Speaker 3: important questions when we think about developing AI centered on

75
00:05:28,440 --> 00:05:31,800
Speaker 3: people and communities, and then how do we begin to

76
00:05:31,920 --> 00:05:34,760
Speaker 3: develop the solutions to make sure we benefit appropriately.

77
00:05:35,760 --> 00:05:41,039
Speaker 2: So that's a very big picture set of ideas. I'm

78
00:05:41,080 --> 00:05:43,400
Speaker 2: curious on a sort of more day to day level.

79
00:05:43,400 --> 00:05:45,719
Speaker 2: I mean, you talk about collaborating with all these different

80
00:05:45,839 --> 00:05:48,160
Speaker 2: kinds of people, all these different groups, what does that

81
00:05:48,200 --> 00:05:50,920
Speaker 2: actually look like, what are some specific examples of how

82
00:05:50,960 --> 00:05:51,839
Speaker 2: you do this work?

83
00:05:52,360 --> 00:05:55,920
Speaker 3: So right now we have about one hundred and twenty

84
00:05:56,040 --> 00:06:01,640
Speaker 3: partners in sixteen countries. They come together through working groups

85
00:06:01,720 --> 00:06:04,640
Speaker 3: that we look at through a variety of different perspectives.

86
00:06:04,680 --> 00:06:08,520
Speaker 3: It could be AI, labor and the economy. It could

87
00:06:08,600 --> 00:06:13,280
Speaker 3: be how do you build a healthy information ecosystem. It

88
00:06:13,320 --> 00:06:16,520
Speaker 3: could be how do you bring more diverse perspectives into

89
00:06:16,640 --> 00:06:21,039
Speaker 3: the inclusive and equitable development of AI. It could be

90
00:06:21,200 --> 00:06:25,480
Speaker 3: what are the emerging opportunities with these very very large

91
00:06:25,520 --> 00:06:29,440
Speaker 3: foundation model applications and how do you deploy those safely?

92
00:06:29,920 --> 00:06:33,479
Speaker 3: And these groups come together most importantly to say what

93
00:06:33,600 --> 00:06:37,040
Speaker 3: are the questions we need to answer collectively, So they

94
00:06:37,040 --> 00:06:39,840
Speaker 3: come together in working groups. I have an amazing staff

95
00:06:39,920 --> 00:06:43,440
Speaker 3: team who hold the pen on synthesizing research and data

96
00:06:43,800 --> 00:06:49,360
Speaker 3: and evidence, developing frameworks, best practices, resources, all sorts of

97
00:06:49,400 --> 00:06:51,839
Speaker 3: things that we can offer up to the community, be

98
00:06:51,960 --> 00:06:55,599
Speaker 3: they in industry or in policy, to say this is

99
00:06:55,680 --> 00:06:58,360
Speaker 3: how we can well, this is what good looks like,

100
00:06:58,440 --> 00:06:59,720
Speaker 3: and this is how we can do it on a

101
00:06:59,800 --> 00:07:01,880
Speaker 3: day to day basis. So that's what we do, and

102
00:07:01,920 --> 00:07:05,240
Speaker 3: then we publish our materials. It's all open. We make

103
00:07:05,320 --> 00:07:07,520
Speaker 3: sure that we get them into the hands of those

104
00:07:07,560 --> 00:07:10,240
Speaker 3: communities that can use them, and then we drive and

105
00:07:10,320 --> 00:07:12,840
Speaker 3: work with those communities to put them into practice.

106
00:07:13,480 --> 00:07:16,480
Speaker 2: You used the word to open there and describing your publications.

107
00:07:17,400 --> 00:07:20,120
Speaker 2: I know, in the world of AI, on the sort

108
00:07:20,160 --> 00:07:24,880
Speaker 2: of technical side, there's a debate, say, or discussion about

109
00:07:24,920 --> 00:07:29,640
Speaker 2: kind of open versus closed AI, And I'm curious how

110
00:07:29,760 --> 00:07:33,560
Speaker 2: you kind of encounter that particular discussion. What is your

111
00:07:33,640 --> 00:07:35,360
Speaker 2: view on open versus closed AI.

112
00:07:36,440 --> 00:07:42,240
Speaker 3: So the current discussion between open and closed release of

113
00:07:42,400 --> 00:07:47,640
Speaker 3: AI models came once we saw chat, GPT and other

114
00:07:47,920 --> 00:07:52,280
Speaker 3: very large generative AI systems being deployed out into the

115
00:07:52,360 --> 00:07:58,040
Speaker 3: hands of consumers around the world, and there emerged some

116
00:07:58,200 --> 00:08:02,960
Speaker 3: fear about the potential of these models to act in

117
00:08:03,040 --> 00:08:06,560
Speaker 3: all sorts of catastrophic ways. So there were concerns that

118
00:08:06,600 --> 00:08:11,640
Speaker 3: the models could be deployed with regard to different development

119
00:08:11,680 --> 00:08:17,000
Speaker 3: of viruses or biomedical weapons or even nuclear weapons, or

120
00:08:17,040 --> 00:08:21,120
Speaker 3: through manipulation or otherwise. So this are emerged about over

121
00:08:21,160 --> 00:08:26,960
Speaker 3: the last eighteen months, this real concern that these models,

122
00:08:27,080 --> 00:08:31,400
Speaker 3: if deployed openly, could lead to some level of truly

123
00:08:31,520 --> 00:08:37,160
Speaker 3: catastrophic risk. And what emerged is actually that we discovered

124
00:08:37,720 --> 00:08:39,559
Speaker 3: that through a whole bunch of work that's been done

125
00:08:39,600 --> 00:08:42,960
Speaker 3: over the last little while, that releasing them openly has

126
00:08:43,000 --> 00:08:45,360
Speaker 3: not led and doesn't appear to be leading in any

127
00:08:45,400 --> 00:08:50,640
Speaker 3: way to catastrophic risk. In facts, releasing them openly allows

128
00:08:50,679 --> 00:08:55,439
Speaker 3: for much more greater scrutiny and understanding of the safety

129
00:08:55,480 --> 00:08:58,080
Speaker 3: measures that have been put into place, And so what

130
00:08:58,240 --> 00:09:01,880
Speaker 3: happened was sort of the pendulum swamp very much towards

131
00:09:01,920 --> 00:09:05,480
Speaker 3: concerned about really catastrophic risk and safety over the last year,

132
00:09:05,520 --> 00:09:08,120
Speaker 3: and over the last year we've seen it swing back

133
00:09:08,200 --> 00:09:10,880
Speaker 3: as we learn more and more about how these models

134
00:09:11,200 --> 00:09:13,760
Speaker 3: are being used and how they are being deployed into

135
00:09:13,800 --> 00:09:19,760
Speaker 3: the world. My feeling is we must approach this work openly,

136
00:09:20,160 --> 00:09:23,439
Speaker 3: and it's not just open release of models or what

137
00:09:23,480 --> 00:09:27,400
Speaker 3: we think of as traditional open source forms of model

138
00:09:27,480 --> 00:09:30,600
Speaker 3: development or otherwise, but we really need to think about

139
00:09:30,679 --> 00:09:35,400
Speaker 3: how do we build an open innovation ecosystem that fundamentally

140
00:09:35,440 --> 00:09:39,520
Speaker 3: allows both for the innovation to be shared with many people,

141
00:09:39,600 --> 00:09:43,320
Speaker 3: but also for safety and security to be rigorously upheld.

142
00:09:43,760 --> 00:09:47,199
Speaker 2: So when you talk about this kind of broader idea

143
00:09:47,240 --> 00:09:51,840
Speaker 2: of open innovation beyond open source or you know, transparency

144
00:09:51,840 --> 00:09:55,119
Speaker 2: in models like what do you mean sort of specifically,

145
00:09:55,160 --> 00:09:56,680
Speaker 2: how does that look in the world.

146
00:09:57,040 --> 00:10:00,959
Speaker 3: So I have three particular points view when it comes

147
00:10:01,000 --> 00:10:03,520
Speaker 3: to open innovation, because I think we need to think

148
00:10:03,720 --> 00:10:07,400
Speaker 3: both upstream around the research that is driving these models,

149
00:10:07,679 --> 00:10:10,559
Speaker 3: and downstream in terms of the benefits of these models

150
00:10:10,559 --> 00:10:14,200
Speaker 3: to others. So first and foremost, what we have known

151
00:10:14,400 --> 00:10:16,720
Speaker 3: in terms of how AI has been developed, and yes,

152
00:10:16,800 --> 00:10:18,680
Speaker 3: I had an opportunity to see it when I was

153
00:10:18,720 --> 00:10:22,080
Speaker 3: at the Canadian Institute for Advanced Research is a very

154
00:10:22,200 --> 00:10:28,120
Speaker 3: open form of scientific publication and rigorous peer review. And

155
00:10:28,240 --> 00:10:31,240
Speaker 3: what happens when we release openly is you have an

156
00:10:31,240 --> 00:10:35,440
Speaker 3: opportunity for the research to be interrogated to determine the

157
00:10:35,559 --> 00:10:38,840
Speaker 3: quality and significance of that, but then also for it

158
00:10:38,880 --> 00:10:41,559
Speaker 3: to be picked up by many others. And then secondly,

159
00:10:42,040 --> 00:10:46,120
Speaker 3: openness for me is about transparency. We released a set

160
00:10:46,160 --> 00:10:49,720
Speaker 3: of very strong recommendations last year around the way in

161
00:10:49,760 --> 00:10:54,040
Speaker 3: which these very large foundation models could be deployed safely.

162
00:10:54,679 --> 00:10:58,760
Speaker 3: They're all about disclosure. They're all about disclosure and documentation

163
00:10:58,920 --> 00:11:02,199
Speaker 3: right from the early days pre R and D development

164
00:11:02,240 --> 00:11:04,720
Speaker 3: of these systems, right in terms of thinking about what's

165
00:11:04,760 --> 00:11:07,319
Speaker 3: in the training data and how is it being used

166
00:11:07,480 --> 00:11:11,480
Speaker 3: all the way through to post deployment monitoring and disclosure.

167
00:11:12,040 --> 00:11:15,280
Speaker 3: So I really think that this is important transparency through it.

168
00:11:15,320 --> 00:11:18,240
Speaker 3: And then the third piece is openness in terms of

169
00:11:18,280 --> 00:11:21,640
Speaker 3: who is around the table to benefit from this technology.

170
00:11:22,080 --> 00:11:24,200
Speaker 3: We know that if we're really going to see these

171
00:11:24,240 --> 00:11:29,280
Speaker 3: new models having being successful deployed into education or healthcare

172
00:11:29,440 --> 00:11:33,319
Speaker 3: or climate and sustainability, we need to have those experts

173
00:11:33,320 --> 00:11:36,480
Speaker 3: in those communities at the table charting this and making

174
00:11:36,520 --> 00:11:39,079
Speaker 3: sure that the technology is working for them. So those

175
00:11:39,080 --> 00:11:40,880
Speaker 3: are the three ways I think about openness.

176
00:11:41,960 --> 00:11:45,400
Speaker 2: Is there like a particular project that you've worked on

177
00:11:45,480 --> 00:11:49,440
Speaker 2: that you feel like you know reflects your approach to

178
00:11:49,600 --> 00:11:50,600
Speaker 2: responsible AI.

179
00:11:51,640 --> 00:11:54,600
Speaker 3: So there's a really interesting project that we have underway

180
00:11:54,600 --> 00:11:58,800
Speaker 3: at PAI that is looking at responsible practices squarely when

181
00:11:58,840 --> 00:12:02,360
Speaker 3: it comes to the use of synthetic media. And what

182
00:12:02,400 --> 00:12:06,000
Speaker 3: we heard from our community was that they were looking

183
00:12:06,040 --> 00:12:09,600
Speaker 3: for a clear code of conduct about what does it

184
00:12:09,679 --> 00:12:12,600
Speaker 3: mean to be responsible in this space. And so what

185
00:12:12,800 --> 00:12:15,800
Speaker 3: happened is we pulled together a number of working groups

186
00:12:15,840 --> 00:12:19,679
Speaker 3: to come together. They included industry representatives. They also included

187
00:12:20,000 --> 00:12:25,760
Speaker 3: civil society organizations like WITNESS, a number of academic institutions

188
00:12:25,760 --> 00:12:28,840
Speaker 3: and otherwise, And what we heard was that there were

189
00:12:29,120 --> 00:12:35,000
Speaker 3: clear requirements that creators could take, that developers of the

190
00:12:35,080 --> 00:12:38,160
Speaker 3: technology could take, and then also distributors. So when we

191
00:12:38,200 --> 00:12:42,719
Speaker 3: think about those generative AI systems being deployed across platforms

192
00:12:42,720 --> 00:12:45,920
Speaker 3: and otherwise, and we came up with a framework for

193
00:12:45,960 --> 00:12:49,360
Speaker 3: what responsibility looks like. What does it mean to have consent,

194
00:12:49,520 --> 00:12:53,120
Speaker 3: what does it mean to disclose responsibly, what does it

195
00:12:53,200 --> 00:12:57,440
Speaker 3: mean to embed technology into it? So, for example, we've

196
00:12:57,440 --> 00:13:00,520
Speaker 3: heard many people talk about the importance of water marking

197
00:13:00,600 --> 00:13:02,920
Speaker 3: systems right and making sure that we have a way

198
00:13:03,000 --> 00:13:05,240
Speaker 3: to water mark them. But what we know from the

199
00:13:05,320 --> 00:13:09,560
Speaker 3: technology is that is a very very complex and complicated problem,

200
00:13:09,800 --> 00:13:13,040
Speaker 3: and what might work on a technical level certainly hits

201
00:13:13,040 --> 00:13:16,040
Speaker 3: a whole new set of complications when we start labeling

202
00:13:16,080 --> 00:13:19,200
Speaker 3: and disclosing out to the public about what that technology

203
00:13:19,240 --> 00:13:22,840
Speaker 3: actually means. All of these, I believe are solvable problems,

204
00:13:22,840 --> 00:13:26,480
Speaker 3: but they all needed to have a clear code underneath

205
00:13:26,520 --> 00:13:28,720
Speaker 3: them that was saying this is what we will commit to.

206
00:13:29,040 --> 00:13:32,280
Speaker 3: And we now have a number of organizations, many many

207
00:13:32,360 --> 00:13:35,600
Speaker 3: of the large technology companies, but also many of the

208
00:13:36,080 --> 00:13:39,200
Speaker 3: small startups who are operating in this based civil society

209
00:13:39,200 --> 00:13:43,040
Speaker 3: and media organizations like the BBC and the CBC who's

210
00:13:43,080 --> 00:13:46,559
Speaker 3: have signed on. And one of the really exciting pieces

211
00:13:46,679 --> 00:13:50,559
Speaker 3: of that is that we're now seeing how it's changing practice.

212
00:13:50,800 --> 00:13:53,559
Speaker 3: So a year in we asked each of our partners

213
00:13:53,600 --> 00:13:56,800
Speaker 3: to come up with a clear case study about how

214
00:13:56,840 --> 00:14:00,000
Speaker 3: that work has changed the way they are making decisions,

215
00:14:00,480 --> 00:14:05,079
Speaker 3: deploying technology and ensuring that they're being responsible in their use.

216
00:14:05,120 --> 00:14:08,040
Speaker 3: And that is creating now a whole resource online that

217
00:14:08,080 --> 00:14:10,440
Speaker 3: we're able to share with others about what does it

218
00:14:10,520 --> 00:14:13,840
Speaker 3: mean to be responsible in this place. There's so much

219
00:14:13,880 --> 00:14:16,040
Speaker 3: more work to be done, and the exciting thing is

220
00:14:16,080 --> 00:14:18,240
Speaker 3: once you have a foundation like this in place, we

221
00:14:18,320 --> 00:14:21,880
Speaker 3: can continue to build on it. So much interest now

222
00:14:21,920 --> 00:14:24,880
Speaker 3: in the policy space, for example, about this work as well.

223
00:14:25,960 --> 00:14:29,760
Speaker 2: Are there any specific examples of those sort of case

224
00:14:29,800 --> 00:14:34,520
Speaker 2: studies or the real world experiences that say media organizations

225
00:14:34,560 --> 00:14:37,520
Speaker 2: had that are interesting that are illuminating. Yes.

226
00:14:37,840 --> 00:14:43,040
Speaker 3: So, for example, what we saw with the BBC is

227
00:14:43,080 --> 00:14:47,000
Speaker 3: that they're developing a lot of content as a public broadcaster,

228
00:14:47,160 --> 00:14:50,240
Speaker 3: both in terms of their news coverage but also in

229
00:14:50,320 --> 00:14:52,960
Speaker 3: terms of some of the resources that they are developing

230
00:14:53,720 --> 00:14:56,480
Speaker 3: for the British public as well. And what they talked

231
00:14:56,480 --> 00:14:59,720
Speaker 3: about was the way in which they had used synthetic

232
00:14:59,760 --> 00:15:04,600
Speaker 3: meat in a very very sensitive environment where they were

233
00:15:04,840 --> 00:15:09,440
Speaker 3: hearing from individuals talk about personal experiences, but wanted to

234
00:15:09,480 --> 00:15:13,040
Speaker 3: have some way to change the face entirely in terms

235
00:15:13,080 --> 00:15:16,240
Speaker 3: of the individuals who were speaking. So that's a very

236
00:15:16,280 --> 00:15:19,920
Speaker 3: complicated ethical question, right, how do you do that responsibily

237
00:15:20,080 --> 00:15:23,160
Speaker 3: and what is the way in which you use that technology,

238
00:15:23,520 --> 00:15:26,640
Speaker 3: and most importantly, how do you disclose it? So their

239
00:15:26,720 --> 00:15:29,800
Speaker 3: case study looked at that in some real detail about

240
00:15:29,840 --> 00:15:33,800
Speaker 3: the process they went through to make the decision responsibly

241
00:15:33,920 --> 00:15:36,840
Speaker 3: to do what they chose, how they intended to use

242
00:15:36,880 --> 00:15:38,240
Speaker 3: the technology in that space.

243
00:15:39,000 --> 00:15:41,960
Speaker 2: As you describe your work in some of these studies,

244
00:15:42,160 --> 00:15:47,160
Speaker 2: the idea of transparency seems to be a theme. Talk

245
00:15:47,200 --> 00:15:49,560
Speaker 2: about the importance of transparency in this kind of work.

246
00:15:50,680 --> 00:15:55,640
Speaker 3: Yeah, transparency is fundamental to responsibility. I always like to

247
00:15:55,680 --> 00:15:59,480
Speaker 3: say it's not accountability in a complete sense, but it

248
00:15:59,560 --> 00:16:03,520
Speaker 3: is a first step to driving accountability more fully, so,

249
00:16:04,040 --> 00:16:07,320
Speaker 3: when we think about how these systems are developed, they're

250
00:16:07,320 --> 00:16:12,560
Speaker 3: often developed behind closed doors inside companies who are making

251
00:16:12,640 --> 00:16:16,680
Speaker 3: decisions about what and how these products will work from

252
00:16:16,680 --> 00:16:21,920
Speaker 3: a business perspective, and what disclosure and transparency can provide

253
00:16:22,000 --> 00:16:25,360
Speaker 3: is some sense of the decisions that were made leading

254
00:16:25,440 --> 00:16:28,239
Speaker 3: up to the way in which those models were deployed.

255
00:16:28,320 --> 00:16:33,720
Speaker 3: So this could be ensuring that individual's private information was

256
00:16:33,760 --> 00:16:38,080
Speaker 3: protected through the process and won't be inadvertently disclosed, or otherwise,

257
00:16:38,520 --> 00:16:41,560
Speaker 3: it could be providing some sense of how well the

258
00:16:41,600 --> 00:16:45,320
Speaker 3: system performs against a whole level of quality measures. So

259
00:16:45,400 --> 00:16:48,160
Speaker 3: we have all of these different types of evaluations and

260
00:16:48,240 --> 00:16:51,520
Speaker 3: a measures that are emerging about the quality of these

261
00:16:51,560 --> 00:16:55,400
Speaker 3: systems as they're deployed. Being transparent about how they perform

262
00:16:55,480 --> 00:16:58,560
Speaker 3: against these systems is really crucial to that as well.

263
00:16:58,840 --> 00:17:01,920
Speaker 3: We have a whole ecosis that's starting to emerge around

264
00:17:02,000 --> 00:17:04,960
Speaker 3: auditing of these systems. So what does that look like

265
00:17:05,080 --> 00:17:07,480
Speaker 3: we think about auditors and all sorts of other sectors

266
00:17:07,480 --> 00:17:10,000
Speaker 3: of the economy. What does it look like to be

267
00:17:10,080 --> 00:17:13,240
Speaker 3: auditing these systems to ensure that they're meeting all of

268
00:17:13,280 --> 00:17:16,960
Speaker 3: those both legal but additional ethical requirements that we want

269
00:17:17,040 --> 00:17:18,200
Speaker 3: to make sure that are in place.

270
00:17:19,520 --> 00:17:24,040
Speaker 2: What are some of the hardest ethical dilemmas you've come

271
00:17:24,119 --> 00:17:26,240
Speaker 2: up against in AI policy.

272
00:17:27,480 --> 00:17:30,840
Speaker 3: Well, the interesting thing about AI policy right is what

273
00:17:30,880 --> 00:17:35,119
Speaker 3: it works very simply in one setting can be highly

274
00:17:35,200 --> 00:17:38,520
Speaker 3: complicated in another setting. And so, for example, I have

275
00:17:38,600 --> 00:17:41,439
Speaker 3: an app that I adore. It's an app on my

276
00:17:41,600 --> 00:17:44,199
Speaker 3: phone that allows me to take a photo of a

277
00:17:44,240 --> 00:17:47,560
Speaker 3: bird and it will help me to better understand what

278
00:17:47,640 --> 00:17:50,280
Speaker 3: that bird is and give me all sorts of information

279
00:17:50,359 --> 00:17:54,520
Speaker 3: about that bird. Now, it's probably right most of the time,

280
00:17:54,600 --> 00:17:56,880
Speaker 3: and it's certainly right enough of the time to give

281
00:17:56,920 --> 00:18:00,159
Speaker 3: me great pleasure and delight when I'm out walking. You

282
00:18:00,200 --> 00:18:04,520
Speaker 3: could think about that exact same technology applied. So for example,

283
00:18:04,600 --> 00:18:07,840
Speaker 3: now you're a security guard and you're working in a

284
00:18:07,960 --> 00:18:12,000
Speaker 3: shopping plaza, and you're able to take photos of individuals

285
00:18:12,040 --> 00:18:15,040
Speaker 3: who you may think are acting suspiciously in some way

286
00:18:15,119 --> 00:18:17,840
Speaker 3: and match that photo up with some sort of a

287
00:18:18,040 --> 00:18:21,679
Speaker 3: database of individuals that may have been found, you know,

288
00:18:21,800 --> 00:18:25,119
Speaker 3: to have some sort of connection to other criminal behavior

289
00:18:25,119 --> 00:18:27,639
Speaker 3: in the past. Right, So what goes from being a

290
00:18:27,680 --> 00:18:30,840
Speaker 3: delightful Oh, isn't this an interesting bird? To a very

291
00:18:31,000 --> 00:18:35,600
Speaker 3: very creepy What does this say about surveillance and privacy

292
00:18:35,720 --> 00:18:39,000
Speaker 3: and access to public spaces? And that is the nature

293
00:18:39,160 --> 00:18:42,520
Speaker 3: of AI. So much of the concern about the ethical

294
00:18:42,720 --> 00:18:48,520
Speaker 3: use and deployment of AI is how an organization is

295
00:18:48,600 --> 00:18:53,960
Speaker 3: making the choices within the social and systemic structure they sit.

296
00:18:54,200 --> 00:18:57,960
Speaker 3: So so much about the ethics of AI is understanding

297
00:18:58,000 --> 00:19:00,959
Speaker 3: what is the use case, how is it being used,

298
00:19:01,080 --> 00:19:04,560
Speaker 3: how is it being constrained? How does it start to

299
00:19:04,760 --> 00:19:08,159
Speaker 3: infringe upon what we think of as the human rights

300
00:19:08,160 --> 00:19:12,240
Speaker 3: of an individual to privacy? And so you have to

301
00:19:12,359 --> 00:19:15,640
Speaker 3: constantly be thinking about ethics. What could work very well

302
00:19:15,640 --> 00:19:19,240
Speaker 3: in one situation absolutely doesn't work in another. We often

303
00:19:19,280 --> 00:19:23,200
Speaker 3: talk about these as socio technical questions. Right, just because

304
00:19:23,240 --> 00:19:26,640
Speaker 3: the technology works doesn't actually mean that it should be

305
00:19:27,040 --> 00:19:28,080
Speaker 3: used and deployed.

306
00:19:28,840 --> 00:19:33,840
Speaker 2: What's an example of where the partnership on AI influence

307
00:19:34,080 --> 00:19:38,399
Speaker 2: changes either in policy or in industry practice.

308
00:19:38,720 --> 00:19:41,399
Speaker 3: We talked a little bit about the Framework for Synthetic

309
00:19:41,480 --> 00:19:45,880
Speaker 3: Media and how that has allowed companies and media organizations

310
00:19:45,880 --> 00:19:49,080
Speaker 3: and civil society organizations to really think deeply about the

311
00:19:49,080 --> 00:19:51,960
Speaker 3: way in which they're using this. Another area that we

312
00:19:52,119 --> 00:19:58,239
Speaker 3: focused on has been around responsible deployment of foundation on

313
00:19:58,320 --> 00:20:01,479
Speaker 3: large scale models. I said, we issued a set of

314
00:20:01,560 --> 00:20:06,359
Speaker 3: recommendations last year that really laid out for these very

315
00:20:06,480 --> 00:20:11,000
Speaker 3: large developers and deployers of foundation and frontier models were

316
00:20:11,520 --> 00:20:14,760
Speaker 3: what does good look like? Right from R and D

317
00:20:15,000 --> 00:20:18,880
Speaker 3: through to deployment monitoring, and it has been very encouraging

318
00:20:19,040 --> 00:20:21,840
Speaker 3: to see that that work has been picked up by

319
00:20:22,240 --> 00:20:26,359
Speaker 3: companies and really articulated as part of the fabric of

320
00:20:26,400 --> 00:20:31,200
Speaker 3: the deployment of their foundation models and systems moving forward.

321
00:20:31,600 --> 00:20:34,680
Speaker 3: So much of this work is around creating clear definitions

322
00:20:34,680 --> 00:20:37,960
Speaker 3: of what we're meaning as the technology evolves and clear

323
00:20:38,000 --> 00:20:40,520
Speaker 3: sets of responsibilities. So it's great to see that work

324
00:20:40,520 --> 00:20:44,000
Speaker 3: getting picked up. The NTIA in the United States just

325
00:20:44,080 --> 00:20:48,439
Speaker 3: released a report on open models and the release of

326
00:20:48,480 --> 00:20:51,159
Speaker 3: open models. Great to see our work cited there as

327
00:20:51,240 --> 00:20:54,560
Speaker 3: contributing to that analysis. Great to see some of our

328
00:20:54,560 --> 00:20:58,520
Speaker 3: definitions and synthetic media getting picked up by legislators in

329
00:20:58,560 --> 00:21:03,159
Speaker 3: different countries. It's important, i think, for us to build capacity,

330
00:21:03,200 --> 00:21:06,199
Speaker 3: knowledge and understanding and our policy makers in this moment

331
00:21:06,680 --> 00:21:10,960
Speaker 3: as the technology is evolving and accelerating in its development.

332
00:21:12,080 --> 00:21:15,680
Speaker 2: What's the AI Alliance and why did Partnership on AI

333
00:21:15,760 --> 00:21:16,520
Speaker 2: decide to join?

334
00:21:17,080 --> 00:21:20,600
Speaker 3: So you had asked about the debate between open versus

335
00:21:20,720 --> 00:21:25,200
Speaker 3: closed models and how that has evolved over the last year,

336
00:21:25,560 --> 00:21:30,040
Speaker 3: and the AI Alliance was a community of organizations that

337
00:21:30,200 --> 00:21:34,480
Speaker 3: came together to really think about, okay, if we support

338
00:21:34,960 --> 00:21:38,479
Speaker 3: open release of models what does that look like and

339
00:21:38,520 --> 00:21:41,000
Speaker 3: what does the community need? And so that's about one

340
00:21:41,119 --> 00:21:45,680
Speaker 3: hundred organizations. IBM, one of our founding partners, is also

341
00:21:45,760 --> 00:21:48,960
Speaker 3: one of the founding partners of the AI Alliance. It's

342
00:21:49,000 --> 00:21:53,000
Speaker 3: a community that brings together a number of academic institutions

343
00:21:53,400 --> 00:21:56,800
Speaker 3: many countries around the world, and they're really focused on

344
00:21:57,400 --> 00:22:01,800
Speaker 3: how do you build the resource is an infrastructure and

345
00:22:01,920 --> 00:22:06,000
Speaker 3: community around what open source in these large scale models

346
00:22:06,040 --> 00:22:09,600
Speaker 3: really mean. So that could be open data sets, that

347
00:22:09,720 --> 00:22:14,520
Speaker 3: could be open technology development. Really building on that understanding

348
00:22:14,560 --> 00:22:17,639
Speaker 3: that we need an infrastructure in place and a community

349
00:22:17,680 --> 00:22:22,959
Speaker 3: engaged in thinking about safety and innovation through the open lens.

350
00:22:23,720 --> 00:22:27,879
Speaker 1: This approach brings together organizations and experts from around the

351
00:22:27,880 --> 00:22:33,840
Speaker 1: globe with different backgrounds, experiences, and perspectives to transparently and

352
00:22:34,000 --> 00:22:38,800
Speaker 1: openly address the challenges and opportunities today. I poses the

353
00:22:38,840 --> 00:22:43,639
Speaker 1: collaborative nature of the AI Alliance encourages discussion, debate, and innovation.

354
00:22:44,440 --> 00:22:47,520
Speaker 1: Through these efforts, IBM is helping to build a community

355
00:22:47,840 --> 00:22:51,480
Speaker 1: around transparent open technology.

356
00:22:52,160 --> 00:22:55,240
Speaker 2: So I want to talk about the future for a minute.

357
00:22:55,480 --> 00:22:58,720
Speaker 2: I'm true, is what you see as the biggest obstacles

358
00:22:58,800 --> 00:23:03,240
Speaker 2: to why spread adoption of responsible AI practices.

359
00:23:03,960 --> 00:23:09,280
Speaker 3: One of the biggest obstacles today is an inability and

360
00:23:09,480 --> 00:23:13,560
Speaker 3: really a lack of understanding about how to use these

361
00:23:13,680 --> 00:23:18,159
Speaker 3: models and how they can most effectively drive forward a

362
00:23:18,240 --> 00:23:22,840
Speaker 3: company's commitment to whatever products and services it might be deploying.

363
00:23:23,240 --> 00:23:26,520
Speaker 3: So I always recommend a couple of things for companies

364
00:23:26,760 --> 00:23:30,080
Speaker 3: really to think about this and to get started. One

365
00:23:30,280 --> 00:23:34,639
Speaker 3: is think about how you are already using AI across

366
00:23:34,680 --> 00:23:38,439
Speaker 3: all of your business products and services, because already AI

367
00:23:38,760 --> 00:23:42,680
Speaker 3: is integrated into our workforces and into our workstreams, and

368
00:23:42,720 --> 00:23:45,680
Speaker 3: into the way in which companies are communicating with their

369
00:23:45,760 --> 00:23:49,280
Speaker 3: clients every day. So understand how you are already using

370
00:23:49,320 --> 00:23:53,800
Speaker 3: it and understand how you are integrating oversight and monitoring

371
00:23:53,840 --> 00:23:56,600
Speaker 3: into those One of the best and clearest ways in

372
00:23:56,640 --> 00:23:59,560
Speaker 3: which a company can really understand how to use this

373
00:23:59,640 --> 00:24:02,960
Speaker 3: response is through documentation. It's one of the areas where

374
00:24:02,960 --> 00:24:06,199
Speaker 3: there's a clear consensus in the community. So how do

375
00:24:06,240 --> 00:24:09,120
Speaker 3: you document the models that you are using, making sure

376
00:24:09,160 --> 00:24:11,240
Speaker 3: that you've got a registry in place. How do you

377
00:24:11,359 --> 00:24:13,879
Speaker 3: document the data that you are using and where that

378
00:24:14,000 --> 00:24:16,439
Speaker 3: data comes from. This is sort of the first system,

379
00:24:16,560 --> 00:24:19,879
Speaker 3: first line of defense in terms of understanding both what

380
00:24:20,080 --> 00:24:22,119
Speaker 3: is in place and what you need to do in

381
00:24:22,200 --> 00:24:25,600
Speaker 3: order to monitor it moving forward. And then secondly, once

382
00:24:25,600 --> 00:24:28,480
Speaker 3: you've got an understanding of how you're already using the system,

383
00:24:28,880 --> 00:24:31,120
Speaker 3: look at ways in which you could begin to pilot

384
00:24:31,280 --> 00:24:34,000
Speaker 3: or iterate in a low risk way using these systems

385
00:24:34,040 --> 00:24:36,960
Speaker 3: to really begin to see how and what structures you

386
00:24:37,040 --> 00:24:39,560
Speaker 3: need to have in place to use it moving forward.

387
00:24:39,920 --> 00:24:43,840
Speaker 3: And then thirdly, make sure that you structure a team

388
00:24:44,000 --> 00:24:47,040
Speaker 3: in place internally that's able to do some of this

389
00:24:47,200 --> 00:24:52,520
Speaker 3: cross departmental monitoring, knowledge sharing and learning boards are very

390
00:24:52,680 --> 00:24:55,679
Speaker 3: very interested in this technology, So thinking about how you

391
00:24:55,720 --> 00:24:58,080
Speaker 3: can have a system or a team in place internally

392
00:24:58,119 --> 00:25:01,000
Speaker 3: that's reporting to your board, giving them a sense of

393
00:25:01,040 --> 00:25:04,920
Speaker 3: both the opportunities that it identifies for you and the

394
00:25:04,960 --> 00:25:08,240
Speaker 3: additional risk mitigation and management you might be putting into place.

395
00:25:08,600 --> 00:25:11,439
Speaker 3: And then you know, once you have those things into place,

396
00:25:11,800 --> 00:25:15,399
Speaker 3: you're really going to need to understand how you work

397
00:25:15,480 --> 00:25:18,760
Speaker 3: with the most valuable asset you have, which is your people.

398
00:25:19,400 --> 00:25:22,400
Speaker 3: How do you make sure that AI systems are working

399
00:25:22,760 --> 00:25:25,240
Speaker 3: for the workers, making sure that they're going into place.

400
00:25:25,280 --> 00:25:28,960
Speaker 3: The most important and impressive implementations we see are those

401
00:25:29,000 --> 00:25:31,439
Speaker 3: where you have the workers who are going to be

402
00:25:31,480 --> 00:25:35,480
Speaker 3: engaged in this process central to figuring out how to

403
00:25:35,560 --> 00:25:39,400
Speaker 3: develop and deploy it in order to really enhance their work.

404
00:25:39,440 --> 00:25:42,159
Speaker 3: It's a core part of a set of Shared Prosperity

405
00:25:42,160 --> 00:25:44,400
Speaker 3: guidelines that we issued last year.

406
00:25:45,200 --> 00:25:50,960
Speaker 2: And then, from the side of policy makers, how should

407
00:25:51,000 --> 00:25:56,560
Speaker 2: policy makers think about the balance between innovation and regulation.

408
00:25:57,280 --> 00:25:59,840
Speaker 3: Yeah, it's so interesting, isn't it that we always think of,

409
00:26:00,040 --> 00:26:04,520
Speaker 3: you know, innovation and regulation as being two sides of

410
00:26:04,560 --> 00:26:08,840
Speaker 3: a coin, when in fact, so much innovation comes from

411
00:26:09,640 --> 00:26:13,639
Speaker 3: having a clear set of guardrails and regulation in place.

412
00:26:14,000 --> 00:26:16,679
Speaker 3: We think about all of the innovation that's happened in

413
00:26:16,920 --> 00:26:22,359
Speaker 3: the automotive industry, right we can drive faster because we

414
00:26:22,600 --> 00:26:25,840
Speaker 3: have breaks, we can drive faster because we have seat

415
00:26:25,840 --> 00:26:28,879
Speaker 3: belts in place. So I think it's often interesting to

416
00:26:28,880 --> 00:26:30,639
Speaker 3: me that we think about the two as being on

417
00:26:30,760 --> 00:26:33,600
Speaker 3: either side of the coin, but an actual fact, you

418
00:26:33,760 --> 00:26:39,479
Speaker 3: can't be innovative without being responsible as well. And so

419
00:26:40,680 --> 00:26:43,159
Speaker 3: I think from a policy maker perspective, what we have

420
00:26:43,240 --> 00:26:46,800
Speaker 3: been really encouraging them to do is to understand that

421
00:26:46,920 --> 00:26:51,400
Speaker 3: you've got foundational regulation in place that works for you. Nationally,

422
00:26:51,440 --> 00:26:55,360
Speaker 3: this could be ensuring that you have strong privacy protections

423
00:26:55,400 --> 00:26:59,640
Speaker 3: in place. It could be ensuring that you are understanding

424
00:26:59,680 --> 00:27:04,120
Speaker 3: pential online harms, particularly to vulnerable communities, and then look

425
00:27:04,119 --> 00:27:07,399
Speaker 3: at what you need to be doing internationally to being

426
00:27:07,440 --> 00:27:11,520
Speaker 3: both competitive and sustainable. There's all sorts of mechanisms that

427
00:27:11,560 --> 00:27:13,919
Speaker 3: are in place right now at the international level to

428
00:27:13,920 --> 00:27:17,359
Speaker 3: think about how do we build an interoperable space for

429
00:27:17,440 --> 00:27:19,159
Speaker 3: these technologies moving forward.

430
00:27:19,760 --> 00:27:23,359
Speaker 2: We've been talking in various ways about what it means

431
00:27:23,520 --> 00:27:29,159
Speaker 2: to responsibly develop AI, and if you're going to boil

432
00:27:29,240 --> 00:27:33,000
Speaker 2: that down, you know the essential concerns that people should

433
00:27:33,040 --> 00:27:35,840
Speaker 2: be thinking about, like what are the key things to

434
00:27:35,920 --> 00:27:38,760
Speaker 2: think about in responsible AI?

435
00:27:39,560 --> 00:27:43,800
Speaker 3: So if you are a company, if we're talking specifically

436
00:27:43,880 --> 00:27:47,680
Speaker 3: through the company lens, when we're thinking about responsible use

437
00:27:47,840 --> 00:27:52,720
Speaker 3: of AI, the most important difference between this form of

438
00:27:52,800 --> 00:27:56,639
Speaker 3: AI technologies and other forms of technologies that we have

439
00:27:56,800 --> 00:28:01,720
Speaker 3: used previously is the integration of data and the training

440
00:28:02,000 --> 00:28:04,600
Speaker 3: models that go on top of that data. So when

441
00:28:04,640 --> 00:28:08,240
Speaker 3: we think about responsibility, first and foremost, you need to

442
00:28:08,280 --> 00:28:11,920
Speaker 3: think about your data. Where did it come from, What

443
00:28:12,080 --> 00:28:15,800
Speaker 3: consent and disclosure requirements do you have on it? Are

444
00:28:15,840 --> 00:28:20,040
Speaker 3: you privacy protecting? You can't be thinking about AI within

445
00:28:20,080 --> 00:28:22,840
Speaker 3: your company without thinking about data, and that's both your

446
00:28:22,880 --> 00:28:26,919
Speaker 3: training data. But then once you're using your systems and

447
00:28:27,040 --> 00:28:30,200
Speaker 3: integrating and interacting with your consumers, how are you protecting

448
00:28:30,240 --> 00:28:33,520
Speaker 3: the data that's coming out of those systems as well?

449
00:28:33,960 --> 00:28:38,000
Speaker 3: And then secondly is when you're thinking about how to

450
00:28:38,160 --> 00:28:42,120
Speaker 3: deploy that AI system, the most important thing you want

451
00:28:42,160 --> 00:28:46,000
Speaker 3: to think about is are we being transparent about how

452
00:28:46,040 --> 00:28:49,400
Speaker 3: it's being used with our clients and our partners. So

453
00:28:49,960 --> 00:28:52,920
Speaker 3: you know the idea that if I'm a customer, I

454
00:28:52,960 --> 00:28:57,040
Speaker 3: should know when I'm interacting with an AI system, I

455
00:28:57,080 --> 00:29:00,160
Speaker 3: should know when I'm interacting with a human. So I

456
00:29:00,160 --> 00:29:03,680
Speaker 3: think those two pieces are the fundamentals. And then of

457
00:29:03,720 --> 00:29:07,520
Speaker 3: course you want to be thinking carefully about making sure

458
00:29:07,560 --> 00:29:11,920
Speaker 3: that whatever jurisdiction you're operating in, you're meeting all of

459
00:29:11,960 --> 00:29:15,400
Speaker 3: the legal requirements with regard to the services and products

460
00:29:15,440 --> 00:29:16,080
Speaker 3: that you're offering.

461
00:29:16,600 --> 00:29:21,719
Speaker 2: Let's finish with the speed round, complete the sentence. In

462
00:29:21,800 --> 00:29:25,160
Speaker 2: five years, AI will will.

463
00:29:25,000 --> 00:29:30,800
Speaker 3: Drive equity, justice, and shared prosperity if we choose to

464
00:29:30,920 --> 00:29:33,600
Speaker 3: set that future trajectory for this technology.

465
00:29:34,560 --> 00:29:38,360
Speaker 2: What is the number one thing that people misunderstand about AI.

466
00:29:39,480 --> 00:29:43,360
Speaker 3: AI is not good, and AI is not bad, but

467
00:29:43,480 --> 00:29:48,560
Speaker 3: AI is also not neutral. It is a product of

468
00:29:48,640 --> 00:29:52,880
Speaker 3: the choices we make as humans about how we deploy

469
00:29:52,960 --> 00:29:53,680
Speaker 3: it in the world.

470
00:29:55,120 --> 00:29:58,360
Speaker 2: What advice would you give yourself ten years ago to

471
00:29:58,600 --> 00:30:03,920
Speaker 2: better prepare yourself for today?

472
00:30:04,560 --> 00:30:09,640
Speaker 3: Ten years ago, I wish that I had known just

473
00:30:10,040 --> 00:30:17,800
Speaker 3: how fundamental the enduring questions of ethics and responsibility would

474
00:30:17,880 --> 00:30:23,160
Speaker 3: be as we developed this technology moving forward, So many

475
00:30:23,240 --> 00:30:26,880
Speaker 3: of the questions that we ask about AI are questions

476
00:30:26,880 --> 00:30:31,640
Speaker 3: about ourselves and the way in which we use technology,

477
00:30:32,200 --> 00:30:34,840
Speaker 3: and the way in which technology can advance the work

478
00:30:34,880 --> 00:30:35,440
Speaker 3: we're doing.

479
00:30:36,600 --> 00:30:38,960
Speaker 2: How do you use AI in your day to day

480
00:30:39,000 --> 00:30:39,680
Speaker 2: life today?

481
00:30:40,280 --> 00:30:43,640
Speaker 3: I use AI all day every day. So whether it's

482
00:30:43,720 --> 00:30:47,600
Speaker 3: my bird app when I go out for my morning walk,

483
00:30:47,880 --> 00:30:51,000
Speaker 3: helping me to better identify birds that I see, or

484
00:30:51,080 --> 00:30:54,080
Speaker 3: whether it is my mapping app that's helping me to

485
00:30:54,120 --> 00:30:57,880
Speaker 3: get more speedily through traffic to whatever meeting I need

486
00:30:57,920 --> 00:31:01,320
Speaker 3: to go to, I use AI all the time. I

487
00:31:01,480 --> 00:31:05,840
Speaker 3: really enjoy using some of the generative AI chatbots more

488
00:31:05,920 --> 00:31:09,040
Speaker 3: for fun than for anything else. As a creative partner

489
00:31:09,080 --> 00:31:13,000
Speaker 3: in thinking through ideas and integrating it into all aspects

490
00:31:13,000 --> 00:31:15,760
Speaker 3: of our lives. Is just so much about the way

491
00:31:15,760 --> 00:31:16,880
Speaker 3: in which we live today.

492
00:31:18,160 --> 00:31:22,400
Speaker 2: So people use the word open to mean different things,

493
00:31:23,000 --> 00:31:26,200
Speaker 2: even just in the context of technology. How do you

494
00:31:26,280 --> 00:31:28,640
Speaker 2: define open in the context of your work.

495
00:31:29,280 --> 00:31:31,440
Speaker 3: So there is the question of open as it is

496
00:31:31,560 --> 00:31:35,360
Speaker 3: deployed to technology, which we've talked a lot about. But

497
00:31:35,480 --> 00:31:39,800
Speaker 3: I do think a big piece of PAI is open minded.

498
00:31:40,680 --> 00:31:44,120
Speaker 3: We need to be open minded truly to listen to,

499
00:31:44,680 --> 00:31:48,840
Speaker 3: for example, what a civil society advocate might say about

500
00:31:48,840 --> 00:31:51,080
Speaker 3: what they're seeing in terms of the way in which

501
00:31:51,160 --> 00:31:55,040
Speaker 3: AI is interacting in a particular community. Or we need

502
00:31:55,080 --> 00:31:57,880
Speaker 3: to be open minded to hear from a technologist about

503
00:31:57,880 --> 00:32:00,360
Speaker 3: their hopes and dreams of where this technology you might

504
00:32:00,400 --> 00:32:04,520
Speaker 3: go moving forward. And we need to have those conversations

505
00:32:04,520 --> 00:32:08,080
Speaker 3: listening to each other to really identify how we're going

506
00:32:08,120 --> 00:32:11,760
Speaker 3: to meet the challenge and opportunity of AI today. So

507
00:32:12,000 --> 00:32:18,479
Speaker 3: open is just fundamental to the partnership on AI. I

508
00:32:18,520 --> 00:32:21,920
Speaker 3: often call it an experiment in open innovation.

509
00:32:23,560 --> 00:32:25,280
Speaker 2: Rebecca, thank you so much for your time.

510
00:32:26,120 --> 00:32:28,040
Speaker 3: It is my pleasure. Thank you for having me.

511
00:32:30,560 --> 00:32:33,320
Speaker 1: Thank you to Rebecca and Jacob for that engaging discussion

512
00:32:33,640 --> 00:32:36,520
Speaker 1: about some of the most pressing issues facing the future

513
00:32:36,600 --> 00:32:40,720
Speaker 1: of AI. As Rebecca emphasized, whether you're thinking about data

514
00:32:40,720 --> 00:32:45,400
Speaker 1: privacy or disclosure, transparency and openness are key to solving

515
00:32:45,520 --> 00:32:52,440
Speaker 1: challenges and capitalizing on new opportunities by developing best practices

516
00:32:52,480 --> 00:32:56,840
Speaker 1: and resources. Partnership on AI is building out the guardrails

517
00:32:57,160 --> 00:33:00,680
Speaker 1: to support the release of open source models the practice

518
00:33:00,720 --> 00:33:05,200
Speaker 1: of post deployment monitoring. By sharing their work with the

519
00:33:05,240 --> 00:33:10,520
Speaker 1: broader community, Rebecca and Pai are demonstrating how working responsibly,

520
00:33:10,920 --> 00:33:17,480
Speaker 1: ethically and openly can help drive innovation. Smart Talks with

521
00:33:17,560 --> 00:33:22,040
Speaker 1: IBM is produced by Matt Ramano, Joey Fishground, Amy Gaines McQuaid,

522
00:33:22,480 --> 00:33:26,640
Speaker 1: and Jacob Goldstein. We're edited by Lydia jen Kott. Our

523
00:33:26,680 --> 00:33:31,280
Speaker 1: engineers are Sarah Brugaer and Ben Tolliday. Theme song by Gramoscope.

524
00:33:31,440 --> 00:33:34,640
Speaker 1: Special thanks to the eight Bar and IBM teams, as

525
00:33:34,720 --> 00:33:38,240
Speaker 1: well as the Pushkin marketing team. Smart Talks with IBM

526
00:33:38,320 --> 00:33:42,160
Speaker 1: is a production of Pushkin Industries and Ruby Studio at iHeartMedia.

527
00:33:42,840 --> 00:33:46,280
Speaker 1: To find more Pushkin podcasts, listen on the iHeartRadio app,

528
00:33:46,520 --> 00:33:51,520
Speaker 1: Apple Podcasts, or wherever you listen to podcasts. I'm Malcolm Glabo.

529
00:33:58,640 --> 00:34:02,200
Speaker 1: This is a paid advertised span from IBM. The conversations

530
00:34:02,240 --> 00:34:08,920
Speaker 1: on this podcast don't necessarily represent IBM's positions, strategies or opinions,