1
00:00:00,120 --> 00:00:02,880
Speaker 1: Hey everyone, it's Robert and Joe here. Today we've got

2
00:00:02,880 --> 00:00:05,120
Speaker 1: something a little different to share with you. It's a

3
00:00:05,160 --> 00:00:08,760
Speaker 1: new season of the Smart Talks with IBM podcast series.

4
00:00:09,240 --> 00:00:12,039
Speaker 2: This season, on smart Talks, Malcolm Gladwell and team are

5
00:00:12,080 --> 00:00:15,280
Speaker 2: diving into the transformative world of artificial intelligence with a

6
00:00:15,320 --> 00:00:18,640
Speaker 2: fresh perspective on the concept of open What does open

7
00:00:18,720 --> 00:00:21,919
Speaker 2: really mean in the context of AI. It can mean

8
00:00:22,040 --> 00:00:25,639
Speaker 2: open source code or open data, but it also encompasses

9
00:00:25,720 --> 00:00:30,800
Speaker 2: fostering an ecosystem of ideas, ensuring diverse perspectives are heard,

10
00:00:31,160 --> 00:00:33,559
Speaker 2: and enabling new levels of transparency.

11
00:00:33,880 --> 00:00:37,120
Speaker 1: Join hosts from your favorite pushkin podcasts as they explore

12
00:00:37,120 --> 00:00:40,960
Speaker 1: how openness and AI is reshaping industries, driving innovation, and

13
00:00:41,000 --> 00:00:44,880
Speaker 1: redefining what's possible. You'll hear from industry experts and leaders

14
00:00:44,880 --> 00:00:48,360
Speaker 1: about the implication and possibilities of open AI, and of course,

15
00:00:48,720 --> 00:00:50,920
Speaker 1: Malcolm Gladwell will be there to guide you through the

16
00:00:50,960 --> 00:00:52,720
Speaker 1: season with his unique insights.

17
00:00:53,000 --> 00:00:55,560
Speaker 2: Look out for new episodes of Smart Talks every other

18
00:00:55,640 --> 00:00:59,200
Speaker 2: week on the iHeartRadio app, Apple Podcasts, or wherever you

19
00:00:59,240 --> 00:01:02,520
Speaker 2: get your podcast and learn more at IBM dot com,

20
00:01:02,560 --> 00:01:14,160
Speaker 2: Slash smart Talks.

21
00:01:11,200 --> 00:01:19,240
Speaker 3: Pushkin Hello, Hello, Welcome to Smart Talks with IBM, a

22
00:01:19,280 --> 00:01:24,600
Speaker 3: podcast from Pushkin Industries, iHeartRadio and IBM. I'm Malcolm Gladwell.

23
00:01:25,600 --> 00:01:28,840
Speaker 3: This season, we're diving back into the world of artificial intelligence,

24
00:01:29,200 --> 00:01:32,200
Speaker 3: but with a focus on the powerful concept of open

25
00:01:32,800 --> 00:01:38,880
Speaker 3: its possibilities, implications, and misconceptions. We'll look at openness from

26
00:01:38,880 --> 00:01:41,679
Speaker 3: a variety of angles and explore how the concept is

27
00:01:41,720 --> 00:01:45,759
Speaker 3: already reshaping industries, ways of doing business and our very

28
00:01:45,840 --> 00:01:51,080
Speaker 3: notion of what's possible. In today's episode, Jacob Goldstein sits

29
00:01:51,120 --> 00:01:55,080
Speaker 3: down with Rebecca Finley, the CEO of the Partnership on Ai,

30
00:01:55,800 --> 00:01:59,720
Speaker 3: a nonprofit group grappling with important questions around the future

31
00:01:59,720 --> 00:02:04,040
Speaker 3: of AI. Their conversation focuses on Rebecca's work bringing together

32
00:02:04,360 --> 00:02:08,280
Speaker 3: a community of diverse stakeholders to help shape the conversation

33
00:02:08,800 --> 00:02:14,280
Speaker 3: around accountable AI governance. Rebecca explains why transparency is so

34
00:02:14,400 --> 00:02:19,440
Speaker 3: crucial for scaling the technology responsibly, and she highlights how

35
00:02:19,480 --> 00:02:22,960
Speaker 3: working with groups like the AI Alliance can provide valuable

36
00:02:23,000 --> 00:02:27,559
Speaker 3: insights in order to build the resources, infrastructure, and community

37
00:02:27,919 --> 00:02:32,800
Speaker 3: around releasing open source models. So, without further ado, let's

38
00:02:32,880 --> 00:02:34,120
Speaker 3: get to that conversation.

39
00:02:41,240 --> 00:02:42,919
Speaker 4: Can you just say your name and your job?

40
00:02:43,400 --> 00:02:46,600
Speaker 5: My name is Rebecca Finley. I am the CEO of

41
00:02:46,639 --> 00:02:50,320
Speaker 5: the Partnership on AI to Benefit People and Society, often

42
00:02:50,400 --> 00:02:52,280
Speaker 5: referred to as PAI.

43
00:02:53,200 --> 00:02:55,760
Speaker 4: How did you get here? What was your job before

44
00:02:55,800 --> 00:02:57,639
Speaker 4: you have the job that you have now?

45
00:02:58,600 --> 00:03:03,640
Speaker 5: I came to about three years ago having had the

46
00:03:03,680 --> 00:03:08,760
Speaker 5: opportunity to work for the Canadian Institute for Advance Research,

47
00:03:09,320 --> 00:03:14,040
Speaker 5: developing and deploying all of their programs related to the

48
00:03:14,080 --> 00:03:19,639
Speaker 5: intersection of technology and society, and one of the areas

49
00:03:19,840 --> 00:03:24,160
Speaker 5: that the Canadian Institute had been funding since nineteen eighty

50
00:03:24,200 --> 00:03:28,040
Speaker 5: two was research into artificial intelligence.

51
00:03:28,280 --> 00:03:30,280
Speaker 4: Wow, early, they were early.

52
00:03:31,360 --> 00:03:35,680
Speaker 5: It was a very early commitment and an ongoing commitment

53
00:03:35,800 --> 00:03:41,160
Speaker 5: at the Institute to fund long term fundamental questions of

54
00:03:41,240 --> 00:03:50,160
Speaker 5: scientific importance in interdisciplinary research programs that were often committed

55
00:03:50,200 --> 00:03:53,720
Speaker 5: and funded to for well over a decade. The AI

56
00:03:53,920 --> 00:03:57,000
Speaker 5: Robotics and Society program that kicked off the work at

57
00:03:57,040 --> 00:04:02,880
Speaker 5: the Institute eventually became a program very much focused on

58
00:04:03,360 --> 00:04:08,480
Speaker 5: deep learning and reinforcement learning, neural networks. All of the

59
00:04:09,160 --> 00:04:14,400
Speaker 5: current iteration of AI, or certainly the pregenerative AI iteration

60
00:04:14,560 --> 00:04:19,080
Speaker 5: of AI that led to this transformation that we've seen

61
00:04:19,160 --> 00:04:22,440
Speaker 5: in terms of online search and all sorts of ways

62
00:04:22,440 --> 00:04:25,360
Speaker 5: in which predictive AI has been deployed. So I had

63
00:04:25,360 --> 00:04:28,760
Speaker 5: the opportunity to see the very early days of that

64
00:04:28,880 --> 00:04:33,159
Speaker 5: research coming together, and when in the early sort of

65
00:04:33,200 --> 00:04:38,480
Speaker 5: two thousand and twenty and tens, when compute capability came

66
00:04:38,560 --> 00:04:43,080
Speaker 5: together with data capability through some of the Internet companies

67
00:04:43,080 --> 00:04:47,080
Speaker 5: and otherwise, and we really saw this technology start to

68
00:04:47,120 --> 00:04:50,120
Speaker 5: take off, I had the opportunity to start up a

69
00:04:50,160 --> 00:04:55,280
Speaker 5: program specifically focused on the impacts of AI in society.

70
00:04:55,960 --> 00:04:58,720
Speaker 5: There was, as you know, at that time, some concerns

71
00:04:58,760 --> 00:05:03,240
Speaker 5: both about the potential for the technology, but also in

72
00:05:03,360 --> 00:05:05,920
Speaker 5: terms of what we were seeing around data sets and

73
00:05:06,200 --> 00:05:10,440
Speaker 5: bias and discrimination and potential impact on future jobs. And

74
00:05:10,560 --> 00:05:14,760
Speaker 5: so bringing a whole group of experts, whether they were

75
00:05:14,760 --> 00:05:20,520
Speaker 5: ethicists or lawyers or economists sociologists into the discussion about

76
00:05:20,560 --> 00:05:24,000
Speaker 5: AI was core to that new program and continues to

77
00:05:24,040 --> 00:05:27,720
Speaker 5: be core to my commitment to bringing diverse perspectives together

78
00:05:27,839 --> 00:05:31,640
Speaker 5: to solve the challenges and opportunities that AI offers today.

79
00:05:32,720 --> 00:05:35,279
Speaker 4: So specifically, what is your job now? What is the

80
00:05:35,320 --> 00:05:37,360
Speaker 4: work you do? What is the work that PAI does.

81
00:05:38,440 --> 00:05:42,320
Speaker 5: I like to answer that question by asking two questions.

82
00:05:42,760 --> 00:05:46,120
Speaker 5: First and foremost, do you believe that the world is

83
00:05:46,360 --> 00:05:50,279
Speaker 5: more divided today than it ever has been in recent history.

84
00:05:51,160 --> 00:05:55,000
Speaker 5: And do you believe that if we don't create spaces

85
00:05:55,560 --> 00:05:58,880
Speaker 5: for very different perspectives to come together, we won't be

86
00:05:58,960 --> 00:06:01,719
Speaker 5: able to solve the challenges that are in front of

87
00:06:01,720 --> 00:06:05,680
Speaker 5: the world today. My answer to both of those questions is, yes,

88
00:06:06,200 --> 00:06:09,760
Speaker 5: we're more divided, and two, we need to seek out

89
00:06:09,880 --> 00:06:14,600
Speaker 5: those spaces where those very different perspectives can come together

90
00:06:15,120 --> 00:06:18,320
Speaker 5: to solve those great challenges. And that's what I get

91
00:06:18,360 --> 00:06:21,880
Speaker 5: to do as CEO of the Partnership on AI. We

92
00:06:21,880 --> 00:06:26,599
Speaker 5: were begun in twenty sixteen with a fundamental commitment to

93
00:06:26,839 --> 00:06:33,320
Speaker 5: bringing together experts, whether they were in industry, academia, civil society,

94
00:06:33,480 --> 00:06:37,720
Speaker 5: or philanthropy, coming together to identify what are the most

95
00:06:37,760 --> 00:06:41,479
Speaker 5: important questions when we think about developing AI centered on

96
00:06:41,560 --> 00:06:44,920
Speaker 5: people and communities, and then how do we begin to

97
00:06:45,040 --> 00:06:47,880
Speaker 5: develop the solutions to make sure we benefit appropriately.

98
00:06:48,880 --> 00:06:54,159
Speaker 4: So that's a very big picture set of ideas. I'm

99
00:06:54,200 --> 00:06:56,520
Speaker 4: curious on a sort of more day to day level.

100
00:06:56,520 --> 00:06:58,839
Speaker 4: I mean, you talk about collaborating with all these different

101
00:06:58,960 --> 00:07:01,280
Speaker 4: kinds of people, all these different groups, what does that

102
00:07:01,320 --> 00:07:04,040
Speaker 4: actually look like. What are some specific examples of how

103
00:07:04,080 --> 00:07:04,960
Speaker 4: you do this work.

104
00:07:05,520 --> 00:07:09,040
Speaker 5: So right now we have about one hundred and twenty

105
00:07:09,160 --> 00:07:14,760
Speaker 5: partners in sixteen countries. They come together through working groups

106
00:07:14,840 --> 00:07:17,760
Speaker 5: that we look at through a variety of different perspectives.

107
00:07:17,800 --> 00:07:21,640
Speaker 5: It could be AI, labor and the economy. It could

108
00:07:21,720 --> 00:07:26,400
Speaker 5: be how do you build a healthy information ecosystem. It

109
00:07:26,440 --> 00:07:29,640
Speaker 5: could be how do you bring more diverse perspectives into

110
00:07:29,760 --> 00:07:34,200
Speaker 5: the inclusive and equitable development of AI. It could be

111
00:07:34,320 --> 00:07:38,600
Speaker 5: what are the emerging opportunities with these very very large

112
00:07:38,640 --> 00:07:42,560
Speaker 5: foundation model applications and how do you deploy those safely?

113
00:07:43,040 --> 00:07:46,600
Speaker 5: And these groups come together most importantly to say what

114
00:07:46,720 --> 00:07:50,160
Speaker 5: are the questions we need to answer collectively, So they

115
00:07:50,160 --> 00:07:52,960
Speaker 5: come together in working groups. I have an amazing staff

116
00:07:53,040 --> 00:07:56,560
Speaker 5: team who hold the pen on synthesizing research and data

117
00:07:56,920 --> 00:08:02,480
Speaker 5: and evidence, developing frameworks, best practice resources, all sorts of

118
00:08:02,520 --> 00:08:04,960
Speaker 5: things that we can offer up to the community, be

119
00:08:05,080 --> 00:08:08,720
Speaker 5: they in industry or in policy, to say this is

120
00:08:08,800 --> 00:08:11,480
Speaker 5: how we can well, this is what good looks like,

121
00:08:11,560 --> 00:08:12,960
Speaker 5: and this is how we can do it on a

122
00:08:13,040 --> 00:08:15,000
Speaker 5: day to day basis. So that's what we do, and

123
00:08:15,040 --> 00:08:18,360
Speaker 5: then we publish our materials. It's all open. We make

124
00:08:18,440 --> 00:08:20,680
Speaker 5: sure that we get them into the hands of those

125
00:08:20,680 --> 00:08:23,360
Speaker 5: communities that can use them, and then we drive and

126
00:08:23,440 --> 00:08:25,960
Speaker 5: work with those communities to put them into practice.

127
00:08:26,600 --> 00:08:29,600
Speaker 4: You use the word open there and describing your publications.

128
00:08:30,520 --> 00:08:33,280
Speaker 4: I know, in the world of AI on the sort

129
00:08:33,280 --> 00:08:37,079
Speaker 4: of technical side, there's a lot of debate, say, or

130
00:08:37,200 --> 00:08:42,160
Speaker 4: discussion about kind of open versus closed AI, And I'm

131
00:08:42,200 --> 00:08:46,320
Speaker 4: curious how you kind of encounter that particular discussion. What

132
00:08:46,480 --> 00:08:48,480
Speaker 4: is your view on open versus closed AI?

133
00:08:49,559 --> 00:08:55,360
Speaker 5: So the current discussion between open and closed release of

134
00:08:55,559 --> 00:09:00,400
Speaker 5: AI models came once we saw a CHAT, GPT and

135
00:09:00,520 --> 00:09:05,319
Speaker 5: other very large generative AI systems being deployed out into

136
00:09:05,320 --> 00:09:10,679
Speaker 5: the hands of consumers around the world, and there emerged

137
00:09:11,000 --> 00:09:15,960
Speaker 5: some fear about the potential of these models to act

138
00:09:16,000 --> 00:09:19,479
Speaker 5: in all sorts of catastrophic ways. So there were concerns

139
00:09:19,520 --> 00:09:23,679
Speaker 5: that the models could be deployed with regard to different

140
00:09:24,240 --> 00:09:29,880
Speaker 5: development of viruses or biomedical weapons or even nuclear weapons,

141
00:09:30,000 --> 00:09:33,800
Speaker 5: or through manipulation or otherwise. So this are emerged about

142
00:09:34,040 --> 00:09:40,080
Speaker 5: over the last eighteen months, this real concern that these models,

143
00:09:40,200 --> 00:09:44,520
Speaker 5: if deployed openly, could lead to some level of truly

144
00:09:44,640 --> 00:09:50,280
Speaker 5: catastrophic risk. And what emerged is actually that we discovered

145
00:09:50,840 --> 00:09:52,720
Speaker 5: that through a whole bunch of work that's been done

146
00:09:52,720 --> 00:09:56,120
Speaker 5: over the last little while that releasing them openly has

147
00:09:56,120 --> 00:09:58,480
Speaker 5: not led and doesn't appear to be leading in any

148
00:09:58,520 --> 00:10:03,760
Speaker 5: way to catastrophic risk. In facts, releasing them openly allows

149
00:10:03,800 --> 00:10:08,559
Speaker 5: for much more greater scrutiny and understanding of the safety

150
00:10:08,600 --> 00:10:11,200
Speaker 5: measures that have been put into place, And so what

151
00:10:11,360 --> 00:10:15,000
Speaker 5: happened was sort of the pendulum swung very much towards

152
00:10:15,040 --> 00:10:18,600
Speaker 5: concerned about really catastrophic risk and safety over the last year,

153
00:10:18,640 --> 00:10:21,240
Speaker 5: and over the last year we've seen it swing back

154
00:10:21,320 --> 00:10:24,000
Speaker 5: as we learn more and more about how these models

155
00:10:24,320 --> 00:10:26,880
Speaker 5: are being used and how they are being deployed into

156
00:10:26,920 --> 00:10:32,880
Speaker 5: the world. My feeling is we must approach this work openly,

157
00:10:33,280 --> 00:10:36,520
Speaker 5: and it's not just open release of models or what

158
00:10:36,600 --> 00:10:40,520
Speaker 5: we think of as traditional open source forms of model

159
00:10:40,600 --> 00:10:43,760
Speaker 5: development or otherwise, but we really need to think about

160
00:10:43,800 --> 00:10:48,520
Speaker 5: how do we build an open innovation ecosystem that fundamentally

161
00:10:48,559 --> 00:10:52,640
Speaker 5: allows both for the innovation to be shared with many people,

162
00:10:52,720 --> 00:10:56,440
Speaker 5: but also for safety and security to be rigorously upheld.

163
00:10:56,880 --> 00:11:00,319
Speaker 4: So when you talk about this kind of broader idea

164
00:11:00,400 --> 00:11:04,960
Speaker 4: of open innovation beyond open source or you know, transparency

165
00:11:04,960 --> 00:11:08,239
Speaker 4: and models, like, what do you mean sort of specifically

166
00:11:08,280 --> 00:11:09,800
Speaker 4: how does that look in the world.

167
00:11:10,160 --> 00:11:13,760
Speaker 5: So I have three particular points of view when it

168
00:11:13,800 --> 00:11:16,400
Speaker 5: comes to open innovation, because I think we need to

169
00:11:16,440 --> 00:11:19,880
Speaker 5: think both both upstream around the research that is driving

170
00:11:19,880 --> 00:11:23,000
Speaker 5: these models and downstream in terms of the benefits of

171
00:11:23,040 --> 00:11:26,760
Speaker 5: these models to others. So, first and foremost, what we

172
00:11:26,840 --> 00:11:29,440
Speaker 5: have known in terms of how AI has been developed,

173
00:11:29,440 --> 00:11:31,480
Speaker 5: and yes, I had an opportunity to see it when

174
00:11:31,520 --> 00:11:34,559
Speaker 5: I was at the Canadian Institute for Advanced Research is

175
00:11:34,640 --> 00:11:40,800
Speaker 5: a very open form of scientific publication and rigorous peer review.

176
00:11:41,120 --> 00:11:44,200
Speaker 5: And what happens when we release openly is you have

177
00:11:44,240 --> 00:11:48,440
Speaker 5: an opportunity for the research to be interrogated to determine

178
00:11:48,480 --> 00:11:51,760
Speaker 5: the quality and significance of that, but then also for

179
00:11:51,880 --> 00:11:54,679
Speaker 5: it to be picked up by many others. And then secondly,

180
00:11:55,160 --> 00:11:59,240
Speaker 5: openness for me is about transparency. We released a set

181
00:11:59,280 --> 00:12:02,840
Speaker 5: of very strong recommendations last year around the way in

182
00:12:02,880 --> 00:12:07,160
Speaker 5: which these very large foundation models could be deployed safely.

183
00:12:07,800 --> 00:12:11,880
Speaker 5: They're all about disclosure. They're all about disclosure and documentation

184
00:12:12,040 --> 00:12:15,359
Speaker 5: right from the early days pre R and D development

185
00:12:15,360 --> 00:12:17,840
Speaker 5: of these systems, right in terms of thinking about what's

186
00:12:17,880 --> 00:12:20,439
Speaker 5: in the training data and how is it being used?

187
00:12:20,600 --> 00:12:24,640
Speaker 5: All the way through to post deployment monitoring and disclosure.

188
00:12:25,160 --> 00:12:28,400
Speaker 5: So I really think that this is important transparency through it.

189
00:12:28,440 --> 00:12:31,360
Speaker 5: And then the third piece is openness in terms of

190
00:12:31,400 --> 00:12:34,760
Speaker 5: who was around the table to benefit from this technology.

191
00:12:35,200 --> 00:12:37,320
Speaker 5: We know that if we're really going to see these

192
00:12:37,360 --> 00:12:42,679
Speaker 5: new models being successful deployed into education or healthcare or

193
00:12:42,760 --> 00:12:46,520
Speaker 5: climate and sustainability, we need to have those experts in

194
00:12:46,600 --> 00:12:49,920
Speaker 5: those communities at the table charting this and making sure

195
00:12:50,000 --> 00:12:52,320
Speaker 5: that the technology is working for them. So those are

196
00:12:52,320 --> 00:12:54,000
Speaker 5: the three ways I think about openness.

197
00:12:55,080 --> 00:12:58,480
Speaker 4: Is there like a particular project that you've worked on

198
00:12:58,640 --> 00:13:02,560
Speaker 4: that you feel like you know reflects your approach to

199
00:13:02,720 --> 00:13:03,720
Speaker 4: responsible AI.

200
00:13:04,760 --> 00:13:07,720
Speaker 5: So there's a really interesting project that we have underway

201
00:13:07,720 --> 00:13:11,880
Speaker 5: at PAI that is looking at responsible practices squarely when

202
00:13:11,960 --> 00:13:15,480
Speaker 5: it comes to the use of synthetic media. And what

203
00:13:15,520 --> 00:13:19,120
Speaker 5: we heard from our community was that they were looking

204
00:13:19,160 --> 00:13:22,680
Speaker 5: for a clear code of conduct about what does it

205
00:13:22,800 --> 00:13:25,720
Speaker 5: mean to be responsible in this space? And so what

206
00:13:25,920 --> 00:13:28,920
Speaker 5: happened is we pulled together a number of working groups

207
00:13:28,960 --> 00:13:32,800
Speaker 5: to come together. They included industry representatives, They also included

208
00:13:33,120 --> 00:13:38,880
Speaker 5: civil society organizations like WITNESS, a number of academic institutions

209
00:13:38,920 --> 00:13:41,960
Speaker 5: and otherwise, and what we heard was that there were

210
00:13:42,240 --> 00:13:48,160
Speaker 5: clear requirements that creators could take, that developers of the

211
00:13:48,200 --> 00:13:51,280
Speaker 5: technology could take, and then also distributors. So when we

212
00:13:51,320 --> 00:13:55,839
Speaker 5: think about those generative AI systems being deployed across platforms

213
00:13:55,840 --> 00:13:59,040
Speaker 5: and otherwise, and we came up with a framework for

214
00:13:59,120 --> 00:14:02,480
Speaker 5: what responsibility looks like. What does it mean to have consent,

215
00:14:02,640 --> 00:14:06,240
Speaker 5: what does it mean to disclose responsibly, what does it

216
00:14:06,320 --> 00:14:10,560
Speaker 5: mean to embed technology into it? So, for example, we've

217
00:14:10,559 --> 00:14:13,600
Speaker 5: heard many people talk about the importance of water marking

218
00:14:13,679 --> 00:14:16,079
Speaker 5: systems right and making sure that we have a way

219
00:14:16,120 --> 00:14:18,360
Speaker 5: to water mark them. But what we know from the

220
00:14:18,440 --> 00:14:22,680
Speaker 5: technology is that is a very very complex and complicated problem,

221
00:14:22,920 --> 00:14:26,160
Speaker 5: and what might work on a technical level certainly hits

222
00:14:26,160 --> 00:14:29,160
Speaker 5: a whole new set of complications when we start labeling

223
00:14:29,240 --> 00:14:32,320
Speaker 5: and disclosing out to the public about what that technology

224
00:14:32,360 --> 00:14:35,960
Speaker 5: actually means. All of these, I believe are solvable problems,

225
00:14:35,960 --> 00:14:39,560
Speaker 5: but they all needed to have a clear code underneath

226
00:14:39,640 --> 00:14:41,840
Speaker 5: them that was saying this is what we will commit to.

227
00:14:42,160 --> 00:14:45,440
Speaker 5: And we now have a number of organizations, many many

228
00:14:45,480 --> 00:14:48,720
Speaker 5: of the large technology companies but also many of the

229
00:14:49,200 --> 00:14:52,320
Speaker 5: small startups who are operating in this based civil society

230
00:14:52,360 --> 00:14:56,160
Speaker 5: and media organizations like the BBC and the CBC who's

231
00:14:56,200 --> 00:14:59,640
Speaker 5: have signed on. And one of the really exciting pieces

232
00:14:59,720 --> 00:15:03,640
Speaker 5: of that is that we're now seeing how it's changing practice.

233
00:15:03,920 --> 00:15:06,640
Speaker 5: So a year in we asked each of our partners

234
00:15:06,720 --> 00:15:09,920
Speaker 5: to come up with a clear case study about how

235
00:15:09,960 --> 00:15:13,160
Speaker 5: that work has changed the way they are making decisions,

236
00:15:13,600 --> 00:15:18,200
Speaker 5: deploying technology and ensuring that they're being responsible in their use.

237
00:15:18,240 --> 00:15:21,160
Speaker 5: And that is creating now a whole resource online that

238
00:15:21,200 --> 00:15:23,560
Speaker 5: we're able to share with others about what does it

239
00:15:23,640 --> 00:15:26,960
Speaker 5: mean to be responsible in this place. There's so much

240
00:15:27,000 --> 00:15:29,160
Speaker 5: more work to be done, and the exciting thing is

241
00:15:29,200 --> 00:15:31,360
Speaker 5: once you have a foundation like this in place, we

242
00:15:31,440 --> 00:15:35,000
Speaker 5: can continue to build on it. So much interest now

243
00:15:35,040 --> 00:15:38,000
Speaker 5: in the policy space, for example, about this work as well.

244
00:15:39,080 --> 00:15:42,880
Speaker 4: Are there any specific examples of those sort of case

245
00:15:42,920 --> 00:15:47,640
Speaker 4: studies or the real world experiences that say media organizations

246
00:15:47,680 --> 00:15:50,000
Speaker 4: had that are interesting that are illuminating.

247
00:15:50,400 --> 00:15:55,400
Speaker 5: Yes. So, for example, what we saw with the BBC

248
00:15:56,000 --> 00:15:59,120
Speaker 5: is that they're developing a lot of content as a

249
00:15:59,120 --> 00:16:02,560
Speaker 5: public broadcast are both in terms of their news coverage

250
00:16:02,560 --> 00:16:05,200
Speaker 5: but also in terms of some of the resources that

251
00:16:05,200 --> 00:16:08,840
Speaker 5: they are developing for the British public as well. And

252
00:16:08,880 --> 00:16:11,320
Speaker 5: what they talked about was the way in which they

253
00:16:11,360 --> 00:16:16,960
Speaker 5: had used synthetic media in a very very sensitive environment

254
00:16:17,080 --> 00:16:21,520
Speaker 5: where they were hearing from individuals talk about personal experiences,

255
00:16:21,880 --> 00:16:25,000
Speaker 5: but wanted to have some way to change the face

256
00:16:25,240 --> 00:16:28,440
Speaker 5: entirely in terms of the individuals who were speaking. So

257
00:16:28,560 --> 00:16:31,720
Speaker 5: that's a very complicated ethical question, right, how do you

258
00:16:31,840 --> 00:16:34,760
Speaker 5: do that responsibly? And what is the way in which

259
00:16:34,800 --> 00:16:38,280
Speaker 5: you use that technology, and most importantly, how do you

260
00:16:38,400 --> 00:16:41,240
Speaker 5: disclose it? So their case study looked at that in

261
00:16:41,320 --> 00:16:45,200
Speaker 5: some real detail about the process they went through to

262
00:16:45,280 --> 00:16:49,000
Speaker 5: make the decision responsibly to do what they chose, how

263
00:16:49,000 --> 00:16:51,360
Speaker 5: they intended to use the technology in that space.

264
00:16:52,120 --> 00:16:55,119
Speaker 4: As you describe your work and some of these studies,

265
00:16:55,280 --> 00:17:00,320
Speaker 4: the idea of transparency seems to be a theme. Talk

266
00:17:00,360 --> 00:17:02,680
Speaker 4: about the importance of transparency in this kind of work.

267
00:17:03,840 --> 00:17:08,760
Speaker 5: Yeah, transparency is fundamental to responsibility. I always like to

268
00:17:08,800 --> 00:17:12,600
Speaker 5: say it's not accountability in a complete sense, but it

269
00:17:12,680 --> 00:17:16,639
Speaker 5: is a first step to driving accountability more fully, so,

270
00:17:17,160 --> 00:17:20,440
Speaker 5: when we think about how these systems are developed, they're

271
00:17:20,440 --> 00:17:25,679
Speaker 5: often developed behind closed doors inside companies who are making

272
00:17:25,760 --> 00:17:29,800
Speaker 5: decisions about what and how these products will work from

273
00:17:29,800 --> 00:17:35,040
Speaker 5: a business perspective, and what disclosure and transparency can provide

274
00:17:35,119 --> 00:17:38,480
Speaker 5: is some sense of the decisions that were made leading

275
00:17:38,560 --> 00:17:41,359
Speaker 5: up to the way in which those models were deployed.

276
00:17:41,440 --> 00:17:46,840
Speaker 5: So this could be ensuring that individual's private information was

277
00:17:46,880 --> 00:17:51,199
Speaker 5: protected through the process and won't be inadvertently disclosed or otherwise.

278
00:17:51,640 --> 00:17:54,679
Speaker 5: It could be providing some sense of how well the

279
00:17:54,720 --> 00:17:58,439
Speaker 5: system performs against a whole level of quality measures. So

280
00:17:58,520 --> 00:18:01,320
Speaker 5: we have all of these different types of evaluations and

281
00:18:01,359 --> 00:18:04,640
Speaker 5: a measures that are emerging about the quality of these

282
00:18:04,680 --> 00:18:08,520
Speaker 5: systems as they're deployed. Being transparent about how they perform

283
00:18:08,600 --> 00:18:11,720
Speaker 5: against these systems is really crucial to that as well.

284
00:18:11,960 --> 00:18:15,040
Speaker 5: We have a whole ecosystem that's starting to emerge around

285
00:18:15,119 --> 00:18:18,119
Speaker 5: auditing of these systems. So what does that look like

286
00:18:18,200 --> 00:18:20,600
Speaker 5: we think about auditors and all sorts of other sectors

287
00:18:20,600 --> 00:18:23,119
Speaker 5: of the economy. What does it look like to be

288
00:18:23,200 --> 00:18:26,359
Speaker 5: auditing these systems to ensure that they're meeting all of

289
00:18:26,400 --> 00:18:30,120
Speaker 5: those both legal but additional ethical requirements that we want

290
00:18:30,160 --> 00:18:31,320
Speaker 5: to make sure that are in place.

291
00:18:32,640 --> 00:18:37,160
Speaker 4: What are some of the hardest ethical dilemmas you've come

292
00:18:37,280 --> 00:18:39,359
Speaker 4: up against in AI policy.

293
00:18:40,600 --> 00:18:43,959
Speaker 5: Well, the interesting thing about AI policy right is what

294
00:18:44,000 --> 00:18:48,280
Speaker 5: it works very simply in one setting, can be highly

295
00:18:48,320 --> 00:18:51,679
Speaker 5: complicated in another setting. And so, for example, I have

296
00:18:51,720 --> 00:18:54,560
Speaker 5: an app that I adore. It's an app on my

297
00:18:54,720 --> 00:18:57,719
Speaker 5: phone that allows me to take a photo of a bird,

298
00:18:58,280 --> 00:19:00,520
Speaker 5: and it will help me to better understand and you know,

299
00:19:00,520 --> 00:19:02,879
Speaker 5: what that bird is, and give me all sorts of

300
00:19:02,920 --> 00:19:07,080
Speaker 5: information about that bird. Now, it's probably right most of

301
00:19:07,119 --> 00:19:09,639
Speaker 5: the time, and it's certainly right enough of the time

302
00:19:09,680 --> 00:19:12,640
Speaker 5: to give me great pleasure and delight when I'm out walking.

303
00:19:13,200 --> 00:19:16,679
Speaker 5: You could think about that exact same technology applied. So

304
00:19:16,880 --> 00:19:20,120
Speaker 5: for example, now you're a security guard and you're working

305
00:19:20,680 --> 00:19:24,280
Speaker 5: in a shopping plaza, and you're able to take photos

306
00:19:24,320 --> 00:19:27,719
Speaker 5: of individuals who you may think are acting suspiciously in

307
00:19:27,760 --> 00:19:30,320
Speaker 5: some way and match that photo up with some sort

308
00:19:30,359 --> 00:19:34,560
Speaker 5: of a database of individuals that may have been found,

309
00:19:34,600 --> 00:19:37,400
Speaker 5: you know, to have some sort of connection to other

310
00:19:37,440 --> 00:19:39,880
Speaker 5: criminal behavior in the past. Right, So what goes from

311
00:19:39,920 --> 00:19:43,359
Speaker 5: being a delightful Oh, isn't this an interesting bird? To

312
00:19:43,480 --> 00:19:47,879
Speaker 5: a very very creepy What is this say about surveillance

313
00:19:47,920 --> 00:19:51,560
Speaker 5: and privacy and access to public spaces? And that is

314
00:19:51,600 --> 00:19:54,800
Speaker 5: the nature of AI. So much of the concern about

315
00:19:54,840 --> 00:20:00,840
Speaker 5: the ethical use and deployment of AI is how organization

316
00:20:01,520 --> 00:20:06,080
Speaker 5: is making the choices within the social and systemic structure

317
00:20:06,600 --> 00:20:09,760
Speaker 5: they sit. So so much about the ethics of AI

318
00:20:10,040 --> 00:20:13,280
Speaker 5: is understanding what is the use case, how is it

319
00:20:13,320 --> 00:20:17,080
Speaker 5: being used, how is it being constrained? How does it

320
00:20:17,160 --> 00:20:20,439
Speaker 5: start to infringe upon what we think of as the

321
00:20:20,520 --> 00:20:24,920
Speaker 5: human rights of an individual to privacy? And so you

322
00:20:25,080 --> 00:20:28,280
Speaker 5: have to constantly be thinking about ethics. What could work

323
00:20:28,400 --> 00:20:31,600
Speaker 5: very well in one situation absolutely doesn't work in another.

324
00:20:31,920 --> 00:20:35,440
Speaker 5: We often talk about these as socio technical questions. Right,

325
00:20:35,840 --> 00:20:39,199
Speaker 5: just because the technology works doesn't actually mean that it

326
00:20:39,240 --> 00:20:41,200
Speaker 5: should be used and deployed.

327
00:20:41,960 --> 00:20:46,959
Speaker 4: What's an example of where the partnership on AI influence

328
00:20:47,200 --> 00:20:50,640
Speaker 4: changes either in policy or in industry practice.

329
00:20:51,840 --> 00:20:54,520
Speaker 5: We talked a little bit about the framework for Synthetic

330
00:20:54,600 --> 00:20:59,000
Speaker 5: Media and how that has allowed companies and media organizations

331
00:20:59,000 --> 00:21:02,120
Speaker 5: and civil society or organizations to really think deeply about

332
00:21:02,119 --> 00:21:04,960
Speaker 5: the way in which they're using this. Another area that

333
00:21:05,000 --> 00:21:10,960
Speaker 5: we focused on has been around responsible deployment of foundation

334
00:21:11,240 --> 00:21:14,040
Speaker 5: and large scale models. So, as I said, we issued

335
00:21:14,080 --> 00:21:18,240
Speaker 5: a set of recommendations last year that really laid out

336
00:21:18,480 --> 00:21:22,399
Speaker 5: for these very large developers and deployers of foundation and

337
00:21:22,480 --> 00:21:27,639
Speaker 5: frontier models, what does good look like right from R

338
00:21:27,680 --> 00:21:30,720
Speaker 5: and D through to deployment monitoring, and it has been

339
00:21:30,880 --> 00:21:34,480
Speaker 5: very encouraging to see that that work has been picked

340
00:21:34,560 --> 00:21:38,840
Speaker 5: up by companies and really articulated as part of the

341
00:21:38,840 --> 00:21:43,520
Speaker 5: fabric of the deployment of their foundation models and systems

342
00:21:43,520 --> 00:21:46,640
Speaker 5: moving forward. So much of this work is around creating

343
00:21:46,720 --> 00:21:50,280
Speaker 5: clear definitions of what we're meaning as the technology evolves

344
00:21:50,640 --> 00:21:53,280
Speaker 5: and clear sets of responsibility. So it's great to see

345
00:21:53,280 --> 00:21:56,600
Speaker 5: that work getting picked up. The NTIA in the United

346
00:21:56,600 --> 00:22:01,119
Speaker 5: States just released a report on open models and the

347
00:22:01,160 --> 00:22:03,879
Speaker 5: release of open models. Great to see our work sited

348
00:22:03,920 --> 00:22:07,320
Speaker 5: there as contributing to that analysis. Great to see some

349
00:22:07,400 --> 00:22:10,720
Speaker 5: of our definitions and synthetic media getting picked up by

350
00:22:10,800 --> 00:22:14,960
Speaker 5: legislators in different countries. Really just it's important, I think,

351
00:22:15,000 --> 00:22:17,679
Speaker 5: for us to build capacity, knowledge and understanding and our

352
00:22:17,720 --> 00:22:22,320
Speaker 5: policy makers in this moment as the technology is evolving

353
00:22:22,400 --> 00:22:24,080
Speaker 5: and accelerating in its development.

354
00:22:25,200 --> 00:22:28,800
Speaker 4: What's the AI Alliance and why did Partnership on AI

355
00:22:28,880 --> 00:22:29,640
Speaker 4: decide to join?

356
00:22:30,200 --> 00:22:33,720
Speaker 5: So you had asked about the debate between open versus

357
00:22:33,840 --> 00:22:38,320
Speaker 5: closed models and how that has evolved over the last year,

358
00:22:38,680 --> 00:22:43,159
Speaker 5: and the AI Alliance was a community of organizations that

359
00:22:43,320 --> 00:22:47,600
Speaker 5: came together to really think about, okay, if we support

360
00:22:48,080 --> 00:22:51,600
Speaker 5: open release of models, what does that look like and

361
00:22:51,640 --> 00:22:54,119
Speaker 5: what does the community need? And so that's about one

362
00:22:54,240 --> 00:22:58,800
Speaker 5: hundred organizations. IBM, one of our founding partners, is also

363
00:22:58,880 --> 00:23:02,080
Speaker 5: one of the founding partner of the AI Alliance. It's

364
00:23:02,119 --> 00:23:06,120
Speaker 5: a community that brings together a number of academic institutions

365
00:23:06,520 --> 00:23:09,959
Speaker 5: many countries around the world, and they're really focused on

366
00:23:10,520 --> 00:23:15,639
Speaker 5: how do you build the resources and infrastructure and community

367
00:23:15,800 --> 00:23:19,600
Speaker 5: around what open source in these large scale models really mean.

368
00:23:19,720 --> 00:23:23,199
Speaker 5: So that could be open data sets, that could be

369
00:23:23,320 --> 00:23:27,919
Speaker 5: open technology development. Really building on that understanding that we

370
00:23:28,000 --> 00:23:31,520
Speaker 5: need an infrastructure in place and a community engaged in

371
00:23:31,600 --> 00:23:36,040
Speaker 5: thinking about safety and innovation through the open lens.

372
00:23:36,880 --> 00:23:41,000
Speaker 3: This approach brings together organizations and experts from around the

373
00:23:41,000 --> 00:23:46,960
Speaker 3: globe with different backgrounds, experiences, and perspectives to transparently and

374
00:23:47,119 --> 00:23:51,919
Speaker 3: openly address the challenges and opportunities today. I poses the

375
00:23:51,960 --> 00:23:56,760
Speaker 3: collaborative nature of the AI Alliance encourages discussion, debate, and innovation.

376
00:23:57,560 --> 00:24:00,639
Speaker 3: Through these efforts, IBM is helping to build the community

377
00:24:00,960 --> 00:24:04,640
Speaker 3: around transparent open technology.

378
00:24:05,280 --> 00:24:08,359
Speaker 4: So I want to talk about the future for a minute.

379
00:24:08,560 --> 00:24:11,840
Speaker 4: I'm sure is what you see as the biggest obstacles

380
00:24:11,920 --> 00:24:16,359
Speaker 4: to widespread adoption of responsible AI practices.

381
00:24:17,080 --> 00:24:22,399
Speaker 5: One of the biggest obstacles today is an inability and

382
00:24:22,600 --> 00:24:26,720
Speaker 5: really a lack of understanding about how to use these

383
00:24:26,800 --> 00:24:31,280
Speaker 5: models and how they can most effectively drive forward a

384
00:24:31,359 --> 00:24:35,959
Speaker 5: company's commitment to whatever products and services it might be deploying.

385
00:24:36,359 --> 00:24:39,720
Speaker 5: So I always recommend a couple of things for companies

386
00:24:39,880 --> 00:24:43,119
Speaker 5: really to think about this and to get started. One

387
00:24:43,440 --> 00:24:47,760
Speaker 5: is think about how you are already using AI across

388
00:24:47,800 --> 00:24:51,560
Speaker 5: all of your business products and services, Because already AI

389
00:24:51,880 --> 00:24:55,840
Speaker 5: is integrated into our workforces and into our workstreams, and

390
00:24:55,840 --> 00:24:58,840
Speaker 5: into the way in which companies are communicating with their

391
00:24:58,880 --> 00:25:02,360
Speaker 5: clients every day. So understand how you are already using

392
00:25:02,440 --> 00:25:06,919
Speaker 5: it and understand how you are integrating oversight and monitoring

393
00:25:06,960 --> 00:25:09,720
Speaker 5: into those One of the best and clearest ways in

394
00:25:09,760 --> 00:25:12,720
Speaker 5: which a company can really understand how to use this

395
00:25:12,840 --> 00:25:16,080
Speaker 5: responsibly is through documentation. It's one of the areas where

396
00:25:16,080 --> 00:25:19,320
Speaker 5: there's a clear consensus in the community. So how do

397
00:25:19,359 --> 00:25:22,240
Speaker 5: you document the models that you are using, making sure

398
00:25:22,280 --> 00:25:24,359
Speaker 5: that you've got a registry in place. How do you

399
00:25:24,480 --> 00:25:27,040
Speaker 5: document the data that you are using and where that

400
00:25:27,119 --> 00:25:29,560
Speaker 5: data comes from. This is sort of the first system,

401
00:25:29,680 --> 00:25:33,000
Speaker 5: first line of defense in terms of understanding both what

402
00:25:33,200 --> 00:25:35,239
Speaker 5: is in place and what you need to do in

403
00:25:35,320 --> 00:25:38,720
Speaker 5: order to monitor it moving forward. And then secondly, once

404
00:25:38,760 --> 00:25:41,600
Speaker 5: you've got an understanding of how you're already using the system,

405
00:25:42,000 --> 00:25:44,240
Speaker 5: look at ways in which you could begin to pilot

406
00:25:44,400 --> 00:25:47,160
Speaker 5: or iterate in a low risk way using these systems

407
00:25:47,160 --> 00:25:50,080
Speaker 5: to really begin to see how and what structures you

408
00:25:50,160 --> 00:25:52,680
Speaker 5: need to have in place to use it moving forward.

409
00:25:53,040 --> 00:25:57,080
Speaker 5: And then thirdly, make sure that you structure a team

410
00:25:57,119 --> 00:26:00,199
Speaker 5: in place internally that's able to do some of this

411
00:26:00,320 --> 00:26:05,720
Speaker 5: cross departmental monitoring, Knowledge sharing and learning boards are very

412
00:26:05,800 --> 00:26:08,800
Speaker 5: very interested in this technology, So thinking about how you

413
00:26:08,840 --> 00:26:11,200
Speaker 5: can have a system or a team in place internally

414
00:26:11,240 --> 00:26:14,119
Speaker 5: that's reporting to your board, giving them a sense of

415
00:26:14,160 --> 00:26:18,040
Speaker 5: both the opportunities that it identifies for you and the

416
00:26:18,080 --> 00:26:21,359
Speaker 5: additional risk mitigation and management you might be putting into place.

417
00:26:21,720 --> 00:26:25,160
Speaker 5: And then once you have those things into place, you're

418
00:26:25,280 --> 00:26:28,679
Speaker 5: really going to need to understand how you work with

419
00:26:28,800 --> 00:26:31,879
Speaker 5: the most valuable asset you have, which is your people.

420
00:26:32,520 --> 00:26:35,520
Speaker 5: How do you make sure that AI systems are working

421
00:26:35,880 --> 00:26:38,359
Speaker 5: for the workers, making sure that they're going into place.

422
00:26:38,440 --> 00:26:42,080
Speaker 5: The most important and impressive implementations we see are those

423
00:26:42,119 --> 00:26:44,560
Speaker 5: where you have the workers who are going to be

424
00:26:44,600 --> 00:26:48,600
Speaker 5: engaged in this process central to figuring out how to

425
00:26:48,680 --> 00:26:52,520
Speaker 5: develop and deploy it in order to really enhance their work.

426
00:26:52,560 --> 00:26:55,280
Speaker 5: It's a core part of a set of Shared Prosperity

427
00:26:55,280 --> 00:26:57,480
Speaker 5: guidelines that we issued last year.

428
00:26:58,320 --> 00:27:04,080
Speaker 4: And then, from the side of policy makers, how should

429
00:27:04,119 --> 00:27:09,680
Speaker 4: policy makers think about the balance between innovation and regulation.

430
00:27:10,400 --> 00:27:13,000
Speaker 5: Yeah, it's so interesting, isn't it that we always think of,

431
00:27:13,119 --> 00:27:17,640
Speaker 5: you know, innovation and regulation as being two sides of

432
00:27:17,680 --> 00:27:22,000
Speaker 5: a coin, when in fact, so much innovation comes from

433
00:27:22,760 --> 00:27:26,800
Speaker 5: having a clear set of guardrails and regulation in place.

434
00:27:27,119 --> 00:27:29,800
Speaker 5: We think about all of the innovation that's happened in

435
00:27:30,040 --> 00:27:35,480
Speaker 5: the automotive industry, right we can drive faster because we

436
00:27:35,720 --> 00:27:38,960
Speaker 5: have breaks, we can drive faster because we have seat

437
00:27:38,960 --> 00:27:42,000
Speaker 5: belts in place. So I think it's often interesting to

438
00:27:42,040 --> 00:27:43,760
Speaker 5: me that we think about the two as being on

439
00:27:43,880 --> 00:27:46,720
Speaker 5: either side of the coin, but an actual fact, you

440
00:27:46,880 --> 00:27:52,600
Speaker 5: can't be innovative without being responsible as well. And so

441
00:27:53,800 --> 00:27:56,280
Speaker 5: I think from a policy maker perspective, what we have

442
00:27:56,359 --> 00:28:00,040
Speaker 5: been really encouraging them to do is to understand that

443
00:28:00,080 --> 00:28:04,520
Speaker 5: you've got foundational regulation in place that works for you. Nationally,

444
00:28:04,560 --> 00:28:08,480
Speaker 5: this could be ensuring that you have strong privacy protections

445
00:28:08,520 --> 00:28:12,880
Speaker 5: in place. It could be ensuring that you are understanding

446
00:28:12,920 --> 00:28:17,240
Speaker 5: potential online harms, particularly to vulnerable communities, and then look

447
00:28:17,240 --> 00:28:20,520
Speaker 5: at what you need to be doing internationally to being

448
00:28:20,600 --> 00:28:24,639
Speaker 5: both competitive and sustainable. There's all sorts of mechanisms that

449
00:28:24,680 --> 00:28:27,040
Speaker 5: are in place right now at the international level to

450
00:28:27,040 --> 00:28:30,480
Speaker 5: think about how do we build an interoperable space for

451
00:28:30,560 --> 00:28:32,320
Speaker 5: these technologies moving forward.

452
00:28:32,880 --> 00:28:36,479
Speaker 4: We've been talking in various ways about what it means

453
00:28:36,640 --> 00:28:42,280
Speaker 4: to responsibly develop AI, and if you're going to boil

454
00:28:42,360 --> 00:28:46,160
Speaker 4: that down, you know the essential concerns that people should

455
00:28:46,160 --> 00:28:48,960
Speaker 4: be thinking about, like what are the key things to

456
00:28:49,040 --> 00:28:51,920
Speaker 4: think about in responsible AI?

457
00:28:52,680 --> 00:28:56,920
Speaker 5: So if you are a company, if we're talking specifically

458
00:28:57,000 --> 00:29:01,479
Speaker 5: through the company lens, when we're thinking about responsese of AI,

459
00:29:02,160 --> 00:29:07,440
Speaker 5: the most important difference between this form of AI technologies

460
00:29:07,480 --> 00:29:10,760
Speaker 5: and other forms of technologies that we have used previously

461
00:29:11,440 --> 00:29:15,719
Speaker 5: is the integration of data and the training models that

462
00:29:15,800 --> 00:29:18,040
Speaker 5: go on top of that data. So when we think

463
00:29:18,080 --> 00:29:21,920
Speaker 5: about responsibility, first and foremost, you need to think about

464
00:29:21,920 --> 00:29:26,120
Speaker 5: your data. Where did it come from? What consent and

465
00:29:26,160 --> 00:29:30,480
Speaker 5: disclosure requirements do you have on it? Are you privacy protecting?

466
00:29:30,920 --> 00:29:34,120
Speaker 5: You can't be thinking about AI within your company without

467
00:29:34,120 --> 00:29:36,880
Speaker 5: thinking about data, and that's both your training data. But

468
00:29:36,960 --> 00:29:41,520
Speaker 5: then once you're using your systems and integrating and interacting

469
00:29:41,560 --> 00:29:44,040
Speaker 5: with your consumers, how are you protecting the data that's

470
00:29:44,080 --> 00:29:48,120
Speaker 5: coming out of those systems as well? And then secondly

471
00:29:48,440 --> 00:29:53,280
Speaker 5: is when you're thinking about how to deploy that AI system,

472
00:29:53,720 --> 00:29:56,040
Speaker 5: the most important thing you want to think about is

473
00:29:56,400 --> 00:30:00,280
Speaker 5: are we being transparent about how it's being you with

474
00:30:00,320 --> 00:30:04,000
Speaker 5: our clients and our partners. So you know, the idea

475
00:30:04,040 --> 00:30:07,160
Speaker 5: that if I'm a customer, I should know when I'm

476
00:30:07,200 --> 00:30:11,520
Speaker 5: interacting with an AI system, I should know when I'm

477
00:30:11,560 --> 00:30:14,440
Speaker 5: interacting with a human. So I think those two pieces

478
00:30:14,560 --> 00:30:17,600
Speaker 5: are the fundamentals. And then of course you want to

479
00:30:17,640 --> 00:30:21,480
Speaker 5: be thinking carefully about, you know, making sure that whatever

480
00:30:21,760 --> 00:30:25,800
Speaker 5: jurisdiction you're operating in, you're meeting all of the legal

481
00:30:25,840 --> 00:30:29,200
Speaker 5: requirements with regard to the services and products that you're offering.

482
00:30:29,720 --> 00:30:34,840
Speaker 4: Let's finish with the speed round, complete the sentence. In

483
00:30:34,920 --> 00:30:36,800
Speaker 4: five years, AI will.

484
00:30:37,800 --> 00:30:43,280
Speaker 5: Will drive equity, justice, and shared prosperity if we choose

485
00:30:43,840 --> 00:30:46,720
Speaker 5: to set that future trajectory for this technology.

486
00:30:47,680 --> 00:30:51,480
Speaker 4: What is the number one thing that people misunderstand about AI.

487
00:30:52,600 --> 00:30:56,440
Speaker 5: AI is not good, and AI is not bad, But

488
00:30:56,600 --> 00:31:01,719
Speaker 5: AI is also not neutral. It is a product of

489
00:31:01,760 --> 00:31:06,000
Speaker 5: the choices we make as humans about how we deploy

490
00:31:06,080 --> 00:31:06,880
Speaker 5: it in the world.

491
00:31:08,280 --> 00:31:11,520
Speaker 4: What advice would you give yourself ten years ago to

492
00:31:11,720 --> 00:31:17,000
Speaker 4: better prepare yourself for today?

493
00:31:17,720 --> 00:31:22,760
Speaker 5: Ten years ago, I wish that I had known just

494
00:31:23,160 --> 00:31:30,920
Speaker 5: how fundamental the enduring questions of ethics and responsibility would

495
00:31:31,000 --> 00:31:36,280
Speaker 5: be as we developed this technology moving forward, So many

496
00:31:36,360 --> 00:31:40,000
Speaker 5: of the questions that we ask about AI are questions

497
00:31:40,000 --> 00:31:44,760
Speaker 5: about ourselves and the way in which we use technology,

498
00:31:45,320 --> 00:31:47,960
Speaker 5: and the way in which technology can advance the work

499
00:31:48,000 --> 00:31:48,560
Speaker 5: we're doing.

500
00:31:49,720 --> 00:31:52,080
Speaker 4: How do you use AI in your day to day

501
00:31:52,120 --> 00:31:52,800
Speaker 4: life today?

502
00:31:53,400 --> 00:31:56,760
Speaker 5: I use AI all day every day. So whether it's

503
00:31:56,840 --> 00:32:00,360
Speaker 5: my bird app when I go out for my learning

504
00:32:00,400 --> 00:32:03,640
Speaker 5: walk helping me to better identify birds that I see,

505
00:32:03,920 --> 00:32:07,040
Speaker 5: or whether it is my mapping app that's helping me

506
00:32:07,120 --> 00:32:10,680
Speaker 5: to get more speedily through traffic to whatever meeting I

507
00:32:10,800 --> 00:32:13,840
Speaker 5: need to go to, I use AI all the time.

508
00:32:14,360 --> 00:32:18,160
Speaker 5: I really enjoy using some of the generative AI chatbots

509
00:32:18,760 --> 00:32:21,680
Speaker 5: more for fun than for anything else. As a creative

510
00:32:21,720 --> 00:32:25,520
Speaker 5: partner in thinking through ideas and integrating it into all

511
00:32:25,640 --> 00:32:28,680
Speaker 5: aspects of our lives. Is just so much about the

512
00:32:28,680 --> 00:32:30,000
Speaker 5: way in which we live today.

513
00:32:31,280 --> 00:32:35,520
Speaker 4: So people use the word open to mean different things,

514
00:32:36,120 --> 00:32:39,360
Speaker 4: even just in the context of technology. How do you

515
00:32:39,400 --> 00:32:41,600
Speaker 4: define open in the context.

516
00:32:41,160 --> 00:32:41,760
Speaker 2: Of your work.

517
00:32:42,400 --> 00:32:44,560
Speaker 5: So there is the question of open as it is

518
00:32:44,680 --> 00:32:48,480
Speaker 5: deployed to technology, which we've talked a lot about. But

519
00:32:48,600 --> 00:32:52,920
Speaker 5: I do think a big piece of PAI is open minded.

520
00:32:53,800 --> 00:32:57,240
Speaker 5: We need to be open minded truly to listen to,

521
00:32:57,800 --> 00:33:01,960
Speaker 5: for example, what a civil society advocate might say about

522
00:33:01,960 --> 00:33:04,200
Speaker 5: what they're seeing in terms of the way in which

523
00:33:04,280 --> 00:33:08,160
Speaker 5: AI is interacting in a particular community. Or we need

524
00:33:08,200 --> 00:33:11,000
Speaker 5: to be open minded to hear from a technologist about

525
00:33:11,000 --> 00:33:13,680
Speaker 5: their hopes and dreams of where this technology might go

526
00:33:13,760 --> 00:33:17,960
Speaker 5: moving forward. And we need to have those conversations listening

527
00:33:18,040 --> 00:33:21,400
Speaker 5: to each other to really identify how we're going to

528
00:33:21,440 --> 00:33:25,680
Speaker 5: meet the challenge and opportunity of AI today. So open

529
00:33:26,640 --> 00:33:31,960
Speaker 5: is just fundamental to the partnership on AI. I often

530
00:33:32,000 --> 00:33:35,040
Speaker 5: call it an experiment in open innovation.

531
00:33:36,680 --> 00:33:38,400
Speaker 4: Rebecca, thank you so much for your time.

532
00:33:39,280 --> 00:33:41,160
Speaker 5: It is my pleasure. Thank you for having me.

533
00:33:43,680 --> 00:33:46,440
Speaker 3: Thank you to Rebecca and Jacob for that engaging discussion

534
00:33:46,760 --> 00:33:49,640
Speaker 3: about some of the most pressing issues facing the future

535
00:33:49,720 --> 00:33:53,840
Speaker 3: of AI. As Rebecca emphasized, whether you're thinking about data

536
00:33:53,840 --> 00:33:58,520
Speaker 3: privacy or disclosure, transparency and openness are key to solving

537
00:33:58,640 --> 00:34:05,560
Speaker 3: challenges and capitalizing on new opportunities by developing best practices

538
00:34:05,600 --> 00:34:10,000
Speaker 3: and resources. Partnership on AI is building out the guardrails

539
00:34:10,280 --> 00:34:13,279
Speaker 3: to support the release of open source models and the

540
00:34:13,320 --> 00:34:18,239
Speaker 3: practice of post deployment monitoring. By sharing their work with

541
00:34:18,280 --> 00:34:23,640
Speaker 3: the broader community, Rebecca and Pai are demonstrating how working responsibly,

542
00:34:24,040 --> 00:34:30,640
Speaker 3: ethically and openly can help drive innovation. Smart Talks with

543
00:34:30,680 --> 00:34:35,160
Speaker 3: IBM is produced by Matt Romano, Joey Fishground, Amy Gaines McQuaid,

544
00:34:35,600 --> 00:34:39,759
Speaker 3: and Jacob Goldstein. We're edited by Lydia jen Kott. Our

545
00:34:39,840 --> 00:34:44,400
Speaker 3: engineers are Sarah Brugaer and Ben Holliday. Theme song by Gramoscope.

546
00:34:44,560 --> 00:34:47,799
Speaker 3: Special thanks to the eight Bar and IBM teams, as

547
00:34:47,840 --> 00:34:51,320
Speaker 3: well as the Pushkin marketing team. Smart Talks with IBM

548
00:34:51,440 --> 00:34:55,280
Speaker 3: is a production of Pushkin Industries and Ruby Studio at iHeartMedia.

549
00:34:55,960 --> 00:34:59,399
Speaker 3: To find more Pushkin podcasts, listen on the iHeartRadio app,

550
00:34:59,600 --> 00:35:04,720
Speaker 3: Apple Podcasts, or wherever you listen to podcasts. I'm Malcolm Glapwell.

551
00:35:11,760 --> 00:35:15,480
Speaker 3: This is a paid advertisement from IBM. The conversations on

552
00:35:15,520 --> 00:35:33,640
Speaker 3: this podcast don't necessarily represent IBM's positions, strategies or opinions,