WEBVTT - Smart Talks with IBM: Scaling AI With Purpose 0:00:00.120 --> 0:00:02.880 Hey everyone, it's Robert and Joe here. Today we've got 0:00:02.880 --> 0:00:05.120 something a little different to share with you. It's a 0:00:05.160 --> 0:00:08.760 new season of the Smart Talks with IBM podcast series. 0:00:09.240 --> 0:00:12.039 This season, on smart Talks, Malcolm Gladwell and team are 0:00:12.080 --> 0:00:15.280 diving into the transformative world of artificial intelligence with a 0:00:15.320 --> 0:00:18.640 fresh perspective on the concept of open What does open 0:00:18.720 --> 0:00:21.919 really mean in the context of AI. It can mean 0:00:22.040 --> 0:00:25.639 open source code or open data, but it also encompasses 0:00:25.720 --> 0:00:30.800 fostering an ecosystem of ideas, ensuring diverse perspectives are heard, 0:00:31.160 --> 0:00:33.559 and enabling new levels of transparency. 0:00:33.880 --> 0:00:37.120 Join hosts from your favorite pushkin podcasts as they explore 0:00:37.120 --> 0:00:40.960 how openness and AI is reshaping industries, driving innovation, and 0:00:41.000 --> 0:00:44.880 redefining what's possible. You'll hear from industry experts and leaders 0:00:44.880 --> 0:00:48.360 about the implication and possibilities of open AI, and of course, 0:00:48.720 --> 0:00:50.920 Malcolm Gladwell will be there to guide you through the 0:00:50.960 --> 0:00:52.720 season with his unique insights. 0:00:53.000 --> 0:00:55.560 Look out for new episodes of Smart Talks every other 0:00:55.640 --> 0:00:59.200 week on the iHeartRadio app, Apple Podcasts, or wherever you 0:00:59.240 --> 0:01:02.520 get your podcast and learn more at IBM dot com, 0:01:02.560 --> 0:01:14.160 Slash smart Talks. 0:01:11.200 --> 0:01:19.240 Pushkin Hello, Hello, Welcome to Smart Talks with IBM, a 0:01:19.280 --> 0:01:24.600 podcast from Pushkin Industries, iHeartRadio and IBM. I'm Malcolm Gladwell. 0:01:25.600 --> 0:01:28.840 This season, we're diving back into the world of artificial intelligence, 0:01:29.200 --> 0:01:32.200 but with a focus on the powerful concept of open 0:01:32.800 --> 0:01:38.880 its possibilities, implications, and misconceptions. We'll look at openness from 0:01:38.880 --> 0:01:41.679 a variety of angles and explore how the concept is 0:01:41.720 --> 0:01:45.759 already reshaping industries, ways of doing business and our very 0:01:45.840 --> 0:01:51.080 notion of what's possible. In today's episode, Jacob Goldstein sits 0:01:51.120 --> 0:01:55.080 down with Rebecca Finley, the CEO of the Partnership on Ai, 0:01:55.800 --> 0:01:59.720 a nonprofit group grappling with important questions around the future 0:01:59.720 --> 0:02:04.040 of AI. Their conversation focuses on Rebecca's work bringing together 0:02:04.360 --> 0:02:08.280 a community of diverse stakeholders to help shape the conversation 0:02:08.800 --> 0:02:14.280 around accountable AI governance. Rebecca explains why transparency is so 0:02:14.400 --> 0:02:19.440 crucial for scaling the technology responsibly, and she highlights how 0:02:19.480 --> 0:02:22.960 working with groups like the AI Alliance can provide valuable 0:02:23.000 --> 0:02:27.559 insights in order to build the resources, infrastructure, and community 0:02:27.919 --> 0:02:32.800 around releasing open source models. So, without further ado, let's 0:02:32.880 --> 0:02:34.120 get to that conversation. 0:02:41.240 --> 0:02:42.919 Can you just say your name and your job? 0:02:43.400 --> 0:02:46.600 My name is Rebecca Finley. I am the CEO of 0:02:46.639 --> 0:02:50.320 the Partnership on AI to Benefit People and Society, often 0:02:50.400 --> 0:02:52.280 referred to as PAI. 0:02:53.200 --> 0:02:55.760 How did you get here? What was your job before 0:02:55.800 --> 0:02:57.639 you have the job that you have now? 0:02:58.600 --> 0:03:03.640 I came to about three years ago having had the 0:03:03.680 --> 0:03:08.760 opportunity to work for the Canadian Institute for Advance Research, 0:03:09.320 --> 0:03:14.040 developing and deploying all of their programs related to the 0:03:14.080 --> 0:03:19.639 intersection of technology and society, and one of the areas 0:03:19.840 --> 0:03:24.160 that the Canadian Institute had been funding since nineteen eighty 0:03:24.200 --> 0:03:28.040 two was research into artificial intelligence. 0:03:28.280 --> 0:03:30.280 Wow, early, they were early. 0:03:31.360 --> 0:03:35.680 It was a very early commitment and an ongoing commitment 0:03:35.800 --> 0:03:41.160 at the Institute to fund long term fundamental questions of 0:03:41.240 --> 0:03:50.160 scientific importance in interdisciplinary research programs that were often committed 0:03:50.200 --> 0:03:53.720 and funded to for well over a decade. The AI 0:03:53.920 --> 0:03:57.000 Robotics and Society program that kicked off the work at 0:03:57.040 --> 0:04:02.880 the Institute eventually became a program very much focused on 0:04:03.360 --> 0:04:08.480 deep learning and reinforcement learning, neural networks. All of the 0:04:09.160 --> 0:04:14.400 current iteration of AI, or certainly the pregenerative AI iteration 0:04:14.560 --> 0:04:19.080 of AI that led to this transformation that we've seen 0:04:19.160 --> 0:04:22.440 in terms of online search and all sorts of ways 0:04:22.440 --> 0:04:25.360 in which predictive AI has been deployed. So I had 0:04:25.360 --> 0:04:28.760 the opportunity to see the very early days of that 0:04:28.880 --> 0:04:33.159 research coming together, and when in the early sort of 0:04:33.200 --> 0:04:38.480 two thousand and twenty and tens, when compute capability came 0:04:38.560 --> 0:04:43.080 together with data capability through some of the Internet companies 0:04:43.080 --> 0:04:47.080 and otherwise, and we really saw this technology start to 0:04:47.120 --> 0:04:50.120 take off, I had the opportunity to start up a 0:04:50.160 --> 0:04:55.280 program specifically focused on the impacts of AI in society. 0:04:55.960 --> 0:04:58.720 There was, as you know, at that time, some concerns 0:04:58.760 --> 0:05:03.240 both about the potential for the technology, but also in 0:05:03.360 --> 0:05:05.920 terms of what we were seeing around data sets and 0:05:06.200 --> 0:05:10.440 bias and discrimination and potential impact on future jobs. And 0:05:10.560 --> 0:05:14.760 so bringing a whole group of experts, whether they were 0:05:14.760 --> 0:05:20.520 ethicists or lawyers or economists sociologists into the discussion about 0:05:20.560 --> 0:05:24.000 AI was core to that new program and continues to 0:05:24.040 --> 0:05:27.720 be core to my commitment to bringing diverse perspectives together 0:05:27.839 --> 0:05:31.640 to solve the challenges and opportunities that AI offers today. 0:05:32.720 --> 0:05:35.279 So specifically, what is your job now? What is the 0:05:35.320 --> 0:05:37.360 work you do? What is the work that PAI does. 0:05:38.440 --> 0:05:42.320 I like to answer that question by asking two questions. 0:05:42.760 --> 0:05:46.120 First and foremost, do you believe that the world is 0:05:46.360 --> 0:05:50.279 more divided today than it ever has been in recent history. 0:05:51.160 --> 0:05:55.000 And do you believe that if we don't create spaces 0:05:55.560 --> 0:05:58.880 for very different perspectives to come together, we won't be 0:05:58.960 --> 0:06:01.719 able to solve the challenges that are in front of 0:06:01.720 --> 0:06:05.680 the world today. My answer to both of those questions is, yes, 0:06:06.200 --> 0:06:09.760 we're more divided, and two, we need to seek out 0:06:09.880 --> 0:06:14.600 those spaces where those very different perspectives can come together 0:06:15.120 --> 0:06:18.320 to solve those great challenges. And that's what I get 0:06:18.360 --> 0:06:21.880 to do as CEO of the Partnership on AI. We 0:06:21.880 --> 0:06:26.599 were begun in twenty sixteen with a fundamental commitment to 0:06:26.839 --> 0:06:33.320 bringing together experts, whether they were in industry, academia, civil society, 0:06:33.480 --> 0:06:37.720 or philanthropy, coming together to identify what are the most 0:06:37.760 --> 0:06:41.479 important questions when we think about developing AI centered on 0:06:41.560 --> 0:06:44.920 people and communities, and then how do we begin to 0:06:45.040 --> 0:06:47.880 develop the solutions to make sure we benefit appropriately. 0:06:48.880 --> 0:06:54.159 So that's a very big picture set of ideas. I'm 0:06:54.200 --> 0:06:56.520 curious on a sort of more day to day level. 0:06:56.520 --> 0:06:58.839 I mean, you talk about collaborating with all these different 0:06:58.960 --> 0:07:01.280 kinds of people, all these different groups, what does that 0:07:01.320 --> 0:07:04.040 actually look like. What are some specific examples of how 0:07:04.080 --> 0:07:04.960 you do this work. 0:07:05.520 --> 0:07:09.040 So right now we have about one hundred and twenty 0:07:09.160 --> 0:07:14.760 partners in sixteen countries. They come together through working groups 0:07:14.840 --> 0:07:17.760 that we look at through a variety of different perspectives. 0:07:17.800 --> 0:07:21.640 It could be AI, labor and the economy. It could 0:07:21.720 --> 0:07:26.400 be how do you build a healthy information ecosystem. It 0:07:26.440 --> 0:07:29.640 could be how do you bring more diverse perspectives into 0:07:29.760 --> 0:07:34.200 the inclusive and equitable development of AI. It could be 0:07:34.320 --> 0:07:38.600 what are the emerging opportunities with these very very large 0:07:38.640 --> 0:07:42.560 foundation model applications and how do you deploy those safely? 0:07:43.040 --> 0:07:46.600 And these groups come together most importantly to say what 0:07:46.720 --> 0:07:50.160 are the questions we need to answer collectively, So they 0:07:50.160 --> 0:07:52.960 come together in working groups. I have an amazing staff 0:07:53.040 --> 0:07:56.560 team who hold the pen on synthesizing research and data 0:07:56.920 --> 0:08:02.480 and evidence, developing frameworks, best practice resources, all sorts of 0:08:02.520 --> 0:08:04.960 things that we can offer up to the community, be 0:08:05.080 --> 0:08:08.720 they in industry or in policy, to say this is 0:08:08.800 --> 0:08:11.480 how we can well, this is what good looks like, 0:08:11.560 --> 0:08:12.960 and this is how we can do it on a 0:08:13.040 --> 0:08:15.000 day to day basis. So that's what we do, and 0:08:15.040 --> 0:08:18.360 then we publish our materials. It's all open. We make 0:08:18.440 --> 0:08:20.680 sure that we get them into the hands of those 0:08:20.680 --> 0:08:23.360 communities that can use them, and then we drive and 0:08:23.440 --> 0:08:25.960 work with those communities to put them into practice. 0:08:26.600 --> 0:08:29.600 You use the word open there and describing your publications. 0:08:30.520 --> 0:08:33.280 I know, in the world of AI on the sort 0:08:33.280 --> 0:08:37.079 of technical side, there's a lot of debate, say, or 0:08:37.200 --> 0:08:42.160 discussion about kind of open versus closed AI, And I'm 0:08:42.200 --> 0:08:46.320 curious how you kind of encounter that particular discussion. What 0:08:46.480 --> 0:08:48.480 is your view on open versus closed AI? 0:08:49.559 --> 0:08:55.360 So the current discussion between open and closed release of 0:08:55.559 --> 0:09:00.400 AI models came once we saw a CHAT, GPT and 0:09:00.520 --> 0:09:05.319 other very large generative AI systems being deployed out into 0:09:05.320 --> 0:09:10.679 the hands of consumers around the world, and there emerged 0:09:11.000 --> 0:09:15.960 some fear about the potential of these models to act 0:09:16.000 --> 0:09:19.479 in all sorts of catastrophic ways. So there were concerns 0:09:19.520 --> 0:09:23.679 that the models could be deployed with regard to different 0:09:24.240 --> 0:09:29.880 development of viruses or biomedical weapons or even nuclear weapons, 0:09:30.000 --> 0:09:33.800 or through manipulation or otherwise. So this are emerged about 0:09:34.040 --> 0:09:40.080 over the last eighteen months, this real concern that these models, 0:09:40.200 --> 0:09:44.520 if deployed openly, could lead to some level of truly 0:09:44.640 --> 0:09:50.280 catastrophic risk. And what emerged is actually that we discovered 0:09:50.840 --> 0:09:52.720 that through a whole bunch of work that's been done 0:09:52.720 --> 0:09:56.120 over the last little while that releasing them openly has 0:09:56.120 --> 0:09:58.480 not led and doesn't appear to be leading in any 0:09:58.520 --> 0:10:03.760 way to catastrophic risk. In facts, releasing them openly allows 0:10:03.800 --> 0:10:08.559 for much more greater scrutiny and understanding of the safety 0:10:08.600 --> 0:10:11.200 measures that have been put into place, And so what 0:10:11.360 --> 0:10:15.000 happened was sort of the pendulum swung very much towards 0:10:15.040 --> 0:10:18.600 concerned about really catastrophic risk and safety over the last year, 0:10:18.640 --> 0:10:21.240 and over the last year we've seen it swing back 0:10:21.320 --> 0:10:24.000 as we learn more and more about how these models 0:10:24.320 --> 0:10:26.880 are being used and how they are being deployed into 0:10:26.920 --> 0:10:32.880 the world. My feeling is we must approach this work openly, 0:10:33.280 --> 0:10:36.520 and it's not just open release of models or what 0:10:36.600 --> 0:10:40.520 we think of as traditional open source forms of model 0:10:40.600 --> 0:10:43.760 development or otherwise, but we really need to think about 0:10:43.800 --> 0:10:48.520 how do we build an open innovation ecosystem that fundamentally 0:10:48.559 --> 0:10:52.640 allows both for the innovation to be shared with many people, 0:10:52.720 --> 0:10:56.440 but also for safety and security to be rigorously upheld. 0:10:56.880 --> 0:11:00.319 So when you talk about this kind of broader idea 0:11:00.400 --> 0:11:04.960 of open innovation beyond open source or you know, transparency 0:11:04.960 --> 0:11:08.239 and models, like, what do you mean sort of specifically 0:11:08.280 --> 0:11:09.800 how does that look in the world. 0:11:10.160 --> 0:11:13.760 So I have three particular points of view when it 0:11:13.800 --> 0:11:16.400 comes to open innovation, because I think we need to 0:11:16.440 --> 0:11:19.880 think both both upstream around the research that is driving 0:11:19.880 --> 0:11:23.000 these models and downstream in terms of the benefits of 0:11:23.040 --> 0:11:26.760 these models to others. So, first and foremost, what we 0:11:26.840 --> 0:11:29.440 have known in terms of how AI has been developed, 0:11:29.440 --> 0:11:31.480 and yes, I had an opportunity to see it when 0:11:31.520 --> 0:11:34.559 I was at the Canadian Institute for Advanced Research is 0:11:34.640 --> 0:11:40.800 a very open form of scientific publication and rigorous peer review. 0:11:41.120 --> 0:11:44.200 And what happens when we release openly is you have 0:11:44.240 --> 0:11:48.440 an opportunity for the research to be interrogated to determine 0:11:48.480 --> 0:11:51.760 the quality and significance of that, but then also for 0:11:51.880 --> 0:11:54.679 it to be picked up by many others. And then secondly, 0:11:55.160 --> 0:11:59.240 openness for me is about transparency. We released a set 0:11:59.280 --> 0:12:02.840 of very strong recommendations last year around the way in 0:12:02.880 --> 0:12:07.160 which these very large foundation models could be deployed safely. 0:12:07.800 --> 0:12:11.880 They're all about disclosure. They're all about disclosure and documentation 0:12:12.040 --> 0:12:15.359 right from the early days pre R and D development 0:12:15.360 --> 0:12:17.840 of these systems, right in terms of thinking about what's 0:12:17.880 --> 0:12:20.439 in the training data and how is it being used? 0:12:20.600 --> 0:12:24.640 All the way through to post deployment monitoring and disclosure. 0:12:25.160 --> 0:12:28.400 So I really think that this is important transparency through it. 0:12:28.440 --> 0:12:31.360 And then the third piece is openness in terms of 0:12:31.400 --> 0:12:34.760 who was around the table to benefit from this technology. 0:12:35.200 --> 0:12:37.320 We know that if we're really going to see these 0:12:37.360 --> 0:12:42.679 new models being successful deployed into education or healthcare or 0:12:42.760 --> 0:12:46.520 climate and sustainability, we need to have those experts in 0:12:46.600 --> 0:12:49.920 those communities at the table charting this and making sure 0:12:50.000 --> 0:12:52.320 that the technology is working for them. So those are 0:12:52.320 --> 0:12:54.000 the three ways I think about openness. 0:12:55.080 --> 0:12:58.480 Is there like a particular project that you've worked on 0:12:58.640 --> 0:13:02.560 that you feel like you know reflects your approach to 0:13:02.720 --> 0:13:03.720 responsible AI. 0:13:04.760 --> 0:13:07.720 So there's a really interesting project that we have underway 0:13:07.720 --> 0:13:11.880 at PAI that is looking at responsible practices squarely when 0:13:11.960 --> 0:13:15.480 it comes to the use of synthetic media. And what 0:13:15.520 --> 0:13:19.120 we heard from our community was that they were looking 0:13:19.160 --> 0:13:22.680 for a clear code of conduct about what does it 0:13:22.800 --> 0:13:25.720 mean to be responsible in this space? And so what 0:13:25.920 --> 0:13:28.920 happened is we pulled together a number of working groups 0:13:28.960 --> 0:13:32.800 to come together. They included industry representatives, They also included 0:13:33.120 --> 0:13:38.880 civil society organizations like WITNESS, a number of academic institutions 0:13:38.920 --> 0:13:41.960 and otherwise, and what we heard was that there were 0:13:42.240 --> 0:13:48.160 clear requirements that creators could take, that developers of the 0:13:48.200 --> 0:13:51.280 technology could take, and then also distributors. So when we 0:13:51.320 --> 0:13:55.839 think about those generative AI systems being deployed across platforms 0:13:55.840 --> 0:13:59.040 and otherwise, and we came up with a framework for 0:13:59.120 --> 0:14:02.480 what responsibility looks like. What does it mean to have consent, 0:14:02.640 --> 0:14:06.240 what does it mean to disclose responsibly, what does it 0:14:06.320 --> 0:14:10.560 mean to embed technology into it? So, for example, we've 0:14:10.559 --> 0:14:13.600 heard many people talk about the importance of water marking 0:14:13.679 --> 0:14:16.079 systems right and making sure that we have a way 0:14:16.120 --> 0:14:18.360 to water mark them. But what we know from the 0:14:18.440 --> 0:14:22.680 technology is that is a very very complex and complicated problem, 0:14:22.920 --> 0:14:26.160 and what might work on a technical level certainly hits 0:14:26.160 --> 0:14:29.160 a whole new set of complications when we start labeling 0:14:29.240 --> 0:14:32.320 and disclosing out to the public about what that technology 0:14:32.360 --> 0:14:35.960 actually means. All of these, I believe are solvable problems, 0:14:35.960 --> 0:14:39.560 but they all needed to have a clear code underneath 0:14:39.640 --> 0:14:41.840 them that was saying this is what we will commit to. 0:14:42.160 --> 0:14:45.440 And we now have a number of organizations, many many 0:14:45.480 --> 0:14:48.720 of the large technology companies but also many of the 0:14:49.200 --> 0:14:52.320 small startups who are operating in this based civil society 0:14:52.360 --> 0:14:56.160 and media organizations like the BBC and the CBC who's 0:14:56.200 --> 0:14:59.640 have signed on. And one of the really exciting pieces 0:14:59.720 --> 0:15:03.640 of that is that we're now seeing how it's changing practice. 0:15:03.920 --> 0:15:06.640 So a year in we asked each of our partners 0:15:06.720 --> 0:15:09.920 to come up with a clear case study about how 0:15:09.960 --> 0:15:13.160 that work has changed the way they are making decisions, 0:15:13.600 --> 0:15:18.200 deploying technology and ensuring that they're being responsible in their use. 0:15:18.240 --> 0:15:21.160 And that is creating now a whole resource online that 0:15:21.200 --> 0:15:23.560 we're able to share with others about what does it 0:15:23.640 --> 0:15:26.960 mean to be responsible in this place. There's so much 0:15:27.000 --> 0:15:29.160 more work to be done, and the exciting thing is 0:15:29.200 --> 0:15:31.360 once you have a foundation like this in place, we 0:15:31.440 --> 0:15:35.000 can continue to build on it. So much interest now 0:15:35.040 --> 0:15:38.000 in the policy space, for example, about this work as well. 0:15:39.080 --> 0:15:42.880 Are there any specific examples of those sort of case 0:15:42.920 --> 0:15:47.640 studies or the real world experiences that say media organizations 0:15:47.680 --> 0:15:50.000 had that are interesting that are illuminating. 0:15:50.400 --> 0:15:55.400 Yes. So, for example, what we saw with the BBC 0:15:56.000 --> 0:15:59.120 is that they're developing a lot of content as a 0:15:59.120 --> 0:16:02.560 public broadcast are both in terms of their news coverage 0:16:02.560 --> 0:16:05.200 but also in terms of some of the resources that 0:16:05.200 --> 0:16:08.840 they are developing for the British public as well. And 0:16:08.880 --> 0:16:11.320 what they talked about was the way in which they 0:16:11.360 --> 0:16:16.960 had used synthetic media in a very very sensitive environment 0:16:17.080 --> 0:16:21.520 where they were hearing from individuals talk about personal experiences, 0:16:21.880 --> 0:16:25.000 but wanted to have some way to change the face 0:16:25.240 --> 0:16:28.440 entirely in terms of the individuals who were speaking. So 0:16:28.560 --> 0:16:31.720 that's a very complicated ethical question, right, how do you 0:16:31.840 --> 0:16:34.760 do that responsibly? And what is the way in which 0:16:34.800 --> 0:16:38.280 you use that technology, and most importantly, how do you 0:16:38.400 --> 0:16:41.240 disclose it? So their case study looked at that in 0:16:41.320 --> 0:16:45.200 some real detail about the process they went through to 0:16:45.280 --> 0:16:49.000 make the decision responsibly to do what they chose, how 0:16:49.000 --> 0:16:51.360 they intended to use the technology in that space. 0:16:52.120 --> 0:16:55.119 As you describe your work and some of these studies, 0:16:55.280 --> 0:17:00.320 the idea of transparency seems to be a theme. Talk 0:17:00.360 --> 0:17:02.680 about the importance of transparency in this kind of work. 0:17:03.840 --> 0:17:08.760 Yeah, transparency is fundamental to responsibility. I always like to 0:17:08.800 --> 0:17:12.600 say it's not accountability in a complete sense, but it 0:17:12.680 --> 0:17:16.639 is a first step to driving accountability more fully, so, 0:17:17.160 --> 0:17:20.440 when we think about how these systems are developed, they're 0:17:20.440 --> 0:17:25.679 often developed behind closed doors inside companies who are making 0:17:25.760 --> 0:17:29.800 decisions about what and how these products will work from 0:17:29.800 --> 0:17:35.040 a business perspective, and what disclosure and transparency can provide 0:17:35.119 --> 0:17:38.480 is some sense of the decisions that were made leading 0:17:38.560 --> 0:17:41.359 up to the way in which those models were deployed. 0:17:41.440 --> 0:17:46.840 So this could be ensuring that individual's private information was 0:17:46.880 --> 0:17:51.199 protected through the process and won't be inadvertently disclosed or otherwise. 0:17:51.640 --> 0:17:54.679 It could be providing some sense of how well the 0:17:54.720 --> 0:17:58.439 system performs against a whole level of quality measures. So 0:17:58.520 --> 0:18:01.320 we have all of these different types of evaluations and 0:18:01.359 --> 0:18:04.640 a measures that are emerging about the quality of these 0:18:04.680 --> 0:18:08.520 systems as they're deployed. Being transparent about how they perform 0:18:08.600 --> 0:18:11.720 against these systems is really crucial to that as well. 0:18:11.960 --> 0:18:15.040 We have a whole ecosystem that's starting to emerge around 0:18:15.119 --> 0:18:18.119 auditing of these systems. So what does that look like 0:18:18.200 --> 0:18:20.600 we think about auditors and all sorts of other sectors 0:18:20.600 --> 0:18:23.119 of the economy. What does it look like to be 0:18:23.200 --> 0:18:26.359 auditing these systems to ensure that they're meeting all of 0:18:26.400 --> 0:18:30.120 those both legal but additional ethical requirements that we want 0:18:30.160 --> 0:18:31.320 to make sure that are in place. 0:18:32.640 --> 0:18:37.160 What are some of the hardest ethical dilemmas you've come 0:18:37.280 --> 0:18:39.359 up against in AI policy. 0:18:40.600 --> 0:18:43.959 Well, the interesting thing about AI policy right is what 0:18:44.000 --> 0:18:48.280 it works very simply in one setting, can be highly 0:18:48.320 --> 0:18:51.679 complicated in another setting. And so, for example, I have 0:18:51.720 --> 0:18:54.560 an app that I adore. It's an app on my 0:18:54.720 --> 0:18:57.719 phone that allows me to take a photo of a bird, 0:18:58.280 --> 0:19:00.520 and it will help me to better understand and you know, 0:19:00.520 --> 0:19:02.879 what that bird is, and give me all sorts of 0:19:02.920 --> 0:19:07.080 information about that bird. Now, it's probably right most of 0:19:07.119 --> 0:19:09.639 the time, and it's certainly right enough of the time 0:19:09.680 --> 0:19:12.640 to give me great pleasure and delight when I'm out walking. 0:19:13.200 --> 0:19:16.679 You could think about that exact same technology applied. So 0:19:16.880 --> 0:19:20.120 for example, now you're a security guard and you're working 0:19:20.680 --> 0:19:24.280 in a shopping plaza, and you're able to take photos 0:19:24.320 --> 0:19:27.719 of individuals who you may think are acting suspiciously in 0:19:27.760 --> 0:19:30.320 some way and match that photo up with some sort 0:19:30.359 --> 0:19:34.560 of a database of individuals that may have been found, 0:19:34.600 --> 0:19:37.400 you know, to have some sort of connection to other 0:19:37.440 --> 0:19:39.880 criminal behavior in the past. Right, So what goes from 0:19:39.920 --> 0:19:43.359 being a delightful Oh, isn't this an interesting bird? To 0:19:43.480 --> 0:19:47.879 a very very creepy What is this say about surveillance 0:19:47.920 --> 0:19:51.560 and privacy and access to public spaces? And that is 0:19:51.600 --> 0:19:54.800 the nature of AI. So much of the concern about 0:19:54.840 --> 0:20:00.840 the ethical use and deployment of AI is how organization 0:20:01.520 --> 0:20:06.080 is making the choices within the social and systemic structure 0:20:06.600 --> 0:20:09.760 they sit. So so much about the ethics of AI 0:20:10.040 --> 0:20:13.280 is understanding what is the use case, how is it 0:20:13.320 --> 0:20:17.080 being used, how is it being constrained? How does it 0:20:17.160 --> 0:20:20.439 start to infringe upon what we think of as the 0:20:20.520 --> 0:20:24.920 human rights of an individual to privacy? And so you 0:20:25.080 --> 0:20:28.280 have to constantly be thinking about ethics. What could work 0:20:28.400 --> 0:20:31.600 very well in one situation absolutely doesn't work in another. 0:20:31.920 --> 0:20:35.440 We often talk about these as socio technical questions. Right, 0:20:35.840 --> 0:20:39.199 just because the technology works doesn't actually mean that it 0:20:39.240 --> 0:20:41.200 should be used and deployed. 0:20:41.960 --> 0:20:46.959 What's an example of where the partnership on AI influence 0:20:47.200 --> 0:20:50.640 changes either in policy or in industry practice. 0:20:51.840 --> 0:20:54.520 We talked a little bit about the framework for Synthetic 0:20:54.600 --> 0:20:59.000 Media and how that has allowed companies and media organizations 0:20:59.000 --> 0:21:02.120 and civil society or organizations to really think deeply about 0:21:02.119 --> 0:21:04.960 the way in which they're using this. Another area that 0:21:05.000 --> 0:21:10.960 we focused on has been around responsible deployment of foundation 0:21:11.240 --> 0:21:14.040 and large scale models. So, as I said, we issued 0:21:14.080 --> 0:21:18.240 a set of recommendations last year that really laid out 0:21:18.480 --> 0:21:22.399 for these very large developers and deployers of foundation and 0:21:22.480 --> 0:21:27.639 frontier models, what does good look like right from R 0:21:27.680 --> 0:21:30.720 and D through to deployment monitoring, and it has been 0:21:30.880 --> 0:21:34.480 very encouraging to see that that work has been picked 0:21:34.560 --> 0:21:38.840 up by companies and really articulated as part of the 0:21:38.840 --> 0:21:43.520 fabric of the deployment of their foundation models and systems 0:21:43.520 --> 0:21:46.640 moving forward. So much of this work is around creating 0:21:46.720 --> 0:21:50.280 clear definitions of what we're meaning as the technology evolves 0:21:50.640 --> 0:21:53.280 and clear sets of responsibility. So it's great to see 0:21:53.280 --> 0:21:56.600 that work getting picked up. The NTIA in the United 0:21:56.600 --> 0:22:01.119 States just released a report on open models and the 0:22:01.160 --> 0:22:03.879 release of open models. Great to see our work sited 0:22:03.920 --> 0:22:07.320 there as contributing to that analysis. Great to see some 0:22:07.400 --> 0:22:10.720 of our definitions and synthetic media getting picked up by 0:22:10.800 --> 0:22:14.960 legislators in different countries. Really just it's important, I think, 0:22:15.000 --> 0:22:17.679 for us to build capacity, knowledge and understanding and our 0:22:17.720 --> 0:22:22.320 policy makers in this moment as the technology is evolving 0:22:22.400 --> 0:22:24.080 and accelerating in its development. 0:22:25.200 --> 0:22:28.800 What's the AI Alliance and why did Partnership on AI 0:22:28.880 --> 0:22:29.640 decide to join? 0:22:30.200 --> 0:22:33.720 So you had asked about the debate between open versus 0:22:33.840 --> 0:22:38.320 closed models and how that has evolved over the last year, 0:22:38.680 --> 0:22:43.159 and the AI Alliance was a community of organizations that 0:22:43.320 --> 0:22:47.600 came together to really think about, okay, if we support 0:22:48.080 --> 0:22:51.600 open release of models, what does that look like and 0:22:51.640 --> 0:22:54.119 what does the community need? And so that's about one 0:22:54.240 --> 0:22:58.800 hundred organizations. IBM, one of our founding partners, is also 0:22:58.880 --> 0:23:02.080 one of the founding partner of the AI Alliance. It's 0:23:02.119 --> 0:23:06.120 a community that brings together a number of academic institutions 0:23:06.520 --> 0:23:09.959 many countries around the world, and they're really focused on 0:23:10.520 --> 0:23:15.639 how do you build the resources and infrastructure and community 0:23:15.800 --> 0:23:19.600 around what open source in these large scale models really mean. 0:23:19.720 --> 0:23:23.199 So that could be open data sets, that could be 0:23:23.320 --> 0:23:27.919 open technology development. Really building on that understanding that we 0:23:28.000 --> 0:23:31.520 need an infrastructure in place and a community engaged in 0:23:31.600 --> 0:23:36.040 thinking about safety and innovation through the open lens. 0:23:36.880 --> 0:23:41.000 This approach brings together organizations and experts from around the 0:23:41.000 --> 0:23:46.960 globe with different backgrounds, experiences, and perspectives to transparently and 0:23:47.119 --> 0:23:51.919 openly address the challenges and opportunities today. I poses the 0:23:51.960 --> 0:23:56.760 collaborative nature of the AI Alliance encourages discussion, debate, and innovation. 0:23:57.560 --> 0:24:00.639 Through these efforts, IBM is helping to build the community 0:24:00.960 --> 0:24:04.640 around transparent open technology. 0:24:05.280 --> 0:24:08.359 So I want to talk about the future for a minute. 0:24:08.560 --> 0:24:11.840 I'm sure is what you see as the biggest obstacles 0:24:11.920 --> 0:24:16.359 to widespread adoption of responsible AI practices. 0:24:17.080 --> 0:24:22.399 One of the biggest obstacles today is an inability and 0:24:22.600 --> 0:24:26.720 really a lack of understanding about how to use these 0:24:26.800 --> 0:24:31.280 models and how they can most effectively drive forward a 0:24:31.359 --> 0:24:35.959 company's commitment to whatever products and services it might be deploying. 0:24:36.359 --> 0:24:39.720 So I always recommend a couple of things for companies 0:24:39.880 --> 0:24:43.119 really to think about this and to get started. One 0:24:43.440 --> 0:24:47.760 is think about how you are already using AI across 0:24:47.800 --> 0:24:51.560 all of your business products and services, Because already AI 0:24:51.880 --> 0:24:55.840 is integrated into our workforces and into our workstreams, and 0:24:55.840 --> 0:24:58.840 into the way in which companies are communicating with their 0:24:58.880 --> 0:25:02.360 clients every day. So understand how you are already using 0:25:02.440 --> 0:25:06.919 it and understand how you are integrating oversight and monitoring 0:25:06.960 --> 0:25:09.720 into those One of the best and clearest ways in 0:25:09.760 --> 0:25:12.720 which a company can really understand how to use this 0:25:12.840 --> 0:25:16.080 responsibly is through documentation. It's one of the areas where 0:25:16.080 --> 0:25:19.320 there's a clear consensus in the community. So how do 0:25:19.359 --> 0:25:22.240 you document the models that you are using, making sure 0:25:22.280 --> 0:25:24.359 that you've got a registry in place. How do you 0:25:24.480 --> 0:25:27.040 document the data that you are using and where that 0:25:27.119 --> 0:25:29.560 data comes from. This is sort of the first system, 0:25:29.680 --> 0:25:33.000