Bản chép lời đầy đủ của AIxMultilateralism: Philanthropy's Role in Inclusive, Ethical AI

--> 00:02:43.988 sense of in order to build a taxonomy of philanthropy in Switzerland. 00:02:44.288 --> 00:02:48.468 Where does the philanthropic center stand with respect to AI and then sort of 00:02:48.468 --> 00:02:51.268 build this research project that clearly looked at that. 00:02:51.768 --> 00:02:57.568 First of all, are philanthropic organizations ready to learn and embrace AI? 00:02:57.568 --> 00:03:02.468 And if so, or to which extent, that was one of the research questions. 00:03:02.668 --> 00:03:07.528 And the other one is AI is sort of this super fast moving technological revolution. 00:03:07.968 --> 00:03:12.408 What is philanthropy doing in order to sort of help society benefit from it, 00:03:12.468 --> 00:03:17.248 especially the ones who are least able to sort of influence the debate that 00:03:17.248 --> 00:03:19.328 maybe could benefit the most from, you know, 00:03:19.448 --> 00:03:24.648 these tools and have some benefits from this technology instead of having only 00:03:24.648 --> 00:03:26.668 the usual suspects, you know, the big debts, 00:03:26.888 --> 00:03:31.548 writing all the benefits and selling all the products and maybe excluding most of the people. 00:03:31.943 --> 00:03:36.203 And over to you now, Hubert, why do you think researching the links between 00:03:36.203 --> 00:03:38.703 AI and philanthropy matter? 00:03:39.523 --> 00:03:44.683 So, philanthropic organizations showcased in the past and showcased today their 00:03:44.683 --> 00:03:49.883 systems-oriented approach and mission-driven action when it comes to solving 00:03:49.883 --> 00:03:52.863 grand challenges like global healthcare, poverty, 00:03:53.363 --> 00:03:55.423 education, human rights abuse. 00:03:56.363 --> 00:04:02.443 So it has the right incentives, the know-how, the means to contribute to the 00:04:02.443 --> 00:04:05.683 grand challenges that AI brings with it and the risks. 00:04:05.923 --> 00:04:12.303 So we've seen with the exponential adoption also the increase in risks and negative impacts of AI. 00:04:12.303 --> 00:04:17.423 So, for example, disinformation, illegal content creation, the impact on jobs, 00:04:17.763 --> 00:04:23.823 biased HR, hiring, or even in financial practices when we talk about loans that 00:04:23.823 --> 00:04:26.323 are based on biased data sets potentially. 00:04:27.123 --> 00:04:33.103 Energy consumption by these energy hungry data centers, IP and data privacy risks, etc. 00:04:33.103 --> 00:04:38.883 So to tackle or bring a perspective and solutions there, but also on promoting 00:04:38.883 --> 00:04:44.043 the beneficial use of AI when it comes to using leveraging AI for good to deliver 00:04:44.043 --> 00:04:46.463 on the UN SDGs, for example, 00:04:47.023 --> 00:04:52.183 but also for philanthropic organizations with this mission-driven and long-term 00:04:52.183 --> 00:04:58.003 orientation to contribute to the governance of AI and filling representation gaps. 00:04:58.003 --> 00:05:04.743 So here, not only the global north-south divide, but also unheard interest groups, societal groups. 00:05:04.923 --> 00:05:09.403 And even if you look at economic stakeholders, supporting less resourceful and 00:05:09.403 --> 00:05:13.023 very small companies, which are often also put on the sidelines. 00:05:13.163 --> 00:05:16.703 So on the sidelines, observing this massive AI transformation, 00:05:16.943 --> 00:05:19.563 but lack the resources to be part of it. 00:05:19.883 --> 00:05:22.043 So various dimensional aspects. 00:05:22.590 --> 00:05:25.510 Definitely a lot of aspects and linkages there. 00:05:25.770 --> 00:05:29.710 A big part of the project that you're working on is the publishing of a book, 00:05:29.850 --> 00:05:34.610 which is open access, called The Ruchelidge Handbook of Artificial Intelligence 00:05:34.610 --> 00:05:39.730 and Philanthropy, edited by Yu Giuseppe as well as Milos Maricic. 00:05:39.730 --> 00:05:46.870 And this kind of looks at two big perspectives, how AI can be used within philanthropy itself, 00:05:47.110 --> 00:05:51.450 which you've briefly touched upon, but also how philanthropy can impact the 00:05:51.450 --> 00:05:56.230 inclusive and responsible development and use of AI, which you mentioned, Hubert. 00:05:56.870 --> 00:06:02.950 So to you Giuseppe, what do you think the book tells us about how the philanthropic 00:06:02.950 --> 00:06:04.670 sector is harnessing AI today? 00:06:05.010 --> 00:06:10.510 And what's the potential for it to support how money and resources are leveraged for actual impact? 00:06:11.070 --> 00:06:15.550 Well, the potential is for sure immense. You don't really know what actually 00:06:15.550 --> 00:06:19.930 AI tools, how are they quickly evolving and how they can support all the different 00:06:19.930 --> 00:06:21.390 philanthropic operations. 00:06:21.750 --> 00:06:26.830 So there is a lot of potential. Now in the book we start looking at a few cases 00:06:26.830 --> 00:06:31.910 where AI has been used by a philanthropic organization in a very successful 00:06:31.910 --> 00:06:33.830 way in order to become more efficient. 00:06:34.090 --> 00:06:39.210 They are actually using AI in order to find maybe players that are so involved 00:06:39.210 --> 00:06:42.890 in the work they're doing that they don't have the time to write an application. 00:06:43.070 --> 00:06:47.670 So it could improve the efficacy and diversification of organizations that can 00:06:47.670 --> 00:06:48.990 be supported by philanthropy. 00:06:49.579 --> 00:06:53.039 On another side, if we think of a more traditional organization, 00:06:53.339 --> 00:07:00.679 we have one of the chapters that looks at how AI has been used in order to build new metrics. 00:07:01.239 --> 00:07:05.979 So there is a very wide range of metrics that can be used of measures and so 00:07:05.979 --> 00:07:07.939 on that vary from one organization to the other. 00:07:08.359 --> 00:07:12.399 What we also did in terms of research, there's this

organization in the Balkans 00:07:12.399 --> 00:07:16.419 that is also using AI to create a map, you know who's who and who is working 00:07:16.419 --> 00:07:20.199 in philanthropy for which reasons and so on in different parts of the world. 00:07:20.319 --> 00:07:25.499 So these are all different aspects that AI can support for philanthropic corporations. 00:07:26.059 --> 00:07:29.879 Maybe if you were to put it all under one umbrella, I think what AI can really 00:07:29.879 --> 00:07:34.119 help is to promote collaboration and the finding of each other, 00:07:34.219 --> 00:07:38.539 whether it's beneficiaries finding donors or whether it's donors identifying 00:07:38.539 --> 00:07:42.879 other donors who are working so they don't reinvent the wheel and so on. 00:07:43.019 --> 00:07:46.819 That's kind of what AI see as enabling in the philanthropic landscape. 00:07:47.589 --> 00:07:51.149 Really interesting. And seeing your examples in the book of how philanthropic 00:07:51.149 --> 00:07:56.369 organizations are harnessing AI to improve processes, to better find suitable 00:07:56.369 --> 00:08:00.169 donors, enhance metrics, and how they measure impact as well. 00:08:00.449 --> 00:08:06.309 I want to look now at the potential of philanthropy to support AI governance, 00:08:06.489 --> 00:08:07.969 which is a massive issue today. 00:08:08.169 --> 00:08:12.629 Talk about it regionally, nationally, and also at the international level how 00:08:12.629 --> 00:08:16.929 we can better govern AI, its development, its use, its deployment. 00:08:17.349 --> 00:08:19.529 And equal access around the world. 00:08:19.949 --> 00:08:23.969 And part of your research is in AI governance and philanthropy. 00:08:23.969 --> 00:08:28.709 You have an article out on the topic. And for those listeners who access our 00:08:28.709 --> 00:08:31.869 resources at the Library and Archives, we do have access to that article. 00:08:31.889 --> 00:08:33.149 So please reach out to us. 00:08:33.309 --> 00:08:38.569 And you argue in that article that philanthropic organizations can and should 00:08:38.569 --> 00:08:42.649 play a key role in AI governance. Hubert, to you, tell us why. 00:08:43.009 --> 00:08:48.729 That's, I think, a key question because AI, as we know and as we see, 00:08:48.849 --> 00:08:50.749 will increasingly become omnipresent. 00:08:50.949 --> 00:08:57.569 So in the way we work, we interact, we learn, and hence also bringing with that 00:08:57.569 --> 00:09:03.349 the positive and negative effects, but quite quickly on a societal and systemic level. 00:09:03.649 --> 00:09:09.609 And with that, we have this AI ecosystem that is quite complex and interconnected. 00:09:09.609 --> 00:09:12.769 Connected among very different stakeholders of society. 00:09:13.089 --> 00:09:19.689 And here, what we look at is how each of these stakeholders can leverage their unique strengths, 00:09:19.809 --> 00:09:25.249 because we see that they are quite complementary in making AI safer and also 00:09:25.249 --> 00:09:28.749 to make sure that AI works for us humans and not the other way around. 00:09:29.049 --> 00:09:33.929 And here, if we look, for example, at different stakeholders, so businesses as one. 00:09:34.689 --> 00:09:40.109 So where companies can act positively or promote responsible AI on the layer 00:09:40.109 --> 00:09:44.269 of product design, deployment, and how users access AI tools, 00:09:44.869 --> 00:09:47.889 they have the unique strengths and resourcefulness that they bring. 00:09:48.109 --> 00:09:54.129 If you look at governments, they can act in terms of policies that promote safe 00:09:54.129 --> 00:09:56.449 AI and safeguard citizens. 00:09:56.889 --> 00:10:01.929 And here their unique lever is enforceability. So thinking about hard governance or regulation. 00:10:01.929 --> 00:10:08.509 And if we look at academia, it's not only about producing scientific knowledge 00:10:08.509 --> 00:10:13.389 and disseminating it, but also educating possibly AI developers and users. 00:10:13.389 --> 00:10:17.829 And last but not least, philanthropic organizations, I think, 00:10:17.969 --> 00:10:24.649 has a unique niche to our responsibility or room to support projects for the 00:10:24.649 --> 00:10:30.389 public good and also act as a sounding board to call out on gaps, 00:10:30.609 --> 00:10:35.129 for example, governance or other negative effects that AI brings along and to 00:10:35.129 --> 00:10:40.449 add this to the public discourse and to make sure that the AI ecosystem is more inclusive. 00:10:40.449 --> 00:10:44.689 And we're seeing fantastic initiatives and, of course, more can be done, 00:10:44.729 --> 00:10:49.149 but it's to support or develop themselves open source models or alternatives. 00:10:49.829 --> 00:10:55.929 Valuation and audit tools, AI literacy programs out there, and also to fund 00:10:55.929 --> 00:10:58.549 international organizations like the UNESCO, 00:10:58.729 --> 00:11:02.589 who is doing very important work on soft governance, as we know, 00:11:02.669 --> 00:11:06.609 and also to make sure marginalized communities are included, 00:11:06.949 --> 00:11:09.269 have a voice and double clicking on that. 00:11:09.269 --> 00:11:11.409 That also like raw languages, that 00:11:11.409 --> 00:11:15.209 these are part of data sets that Dutch language models are trained on. 00:11:15.489 --> 00:11:20.169 And all this to say, to leverage really their collaborative approach that philanthropic 00:11:20.169 --> 00:11:24.709 organizations engage in, their mission-driven and oftentimes human-centered approach. 00:11:24.989 --> 00:11:31.289 And all this to have a more representative and fairer ecosystem to really set 00:11:31.289 --> 00:11:35.729 guardrails and to kind of have this balance of positive-negative effects was 00:11:35.729 --> 00:11:36.869 heavy on the positive side. 00:11:37.435 --> 00:11:43.035 And so what I hear is that philanthropy can play a very critical role in supporting

00:11:43.035 --> 00:11:45.275 the benefits of AI for more people. 00:11:46.155 --> 00:11:51.415 Giuseppe, I want to ask you now a bit more about AI governance because it is a prolific theme today. 00:11:51.795 --> 00:11:56.215 And in your work, in your research, you cite that the OECD counts, 00:11:56.275 --> 00:12:00.955 in fact, more than 1000 AI governance initiatives in terms of soft governance. 00:12:01.355 --> 00:12:04.895 And of course, there are also some hard governance initiatives for AI, 00:12:05.195 --> 00:12:07.335 but the majority today are soft governance. 00:12:07.955 --> 00:12:13.135 Where do you see the biggest opportunities for philanthropy to support AI governance 00:12:13.135 --> 00:12:16.735 in this very complex layered world of frameworks and principles? 00:12:17.396 --> 00:12:20.656 So really nice, really good point, really good question. 00:12:20.916 --> 00:12:25.016 So I think in general, what philanthropy could do is to try to, 00:12:25.516 --> 00:12:29.036 among these principles, among this quality noise, what is the real signal? 00:12:29.756 --> 00:12:32.636 What are the principles that are going to matter the most? 00:12:32.836 --> 00:12:36.256 And to come back to what Hubert was saying before, for example. 00:12:36.816 --> 00:12:40.756 What principles should we follow in order to make sure that the data industry 00:12:40.756 --> 00:12:46.696 trained models is actually representative of the majority of the population, not only a few groups. 00:12:46.696 --> 00:12:49.996 So that's, I think, the role that philanthropy should take. 00:12:50.196 --> 00:12:54.956 It's not really to produce new governance or to enter too much in ethical debates, 00:12:55.076 --> 00:13:00.956 but it's really to either fund or by themselves to try to identify the principles 00:13:00.956 --> 00:13:04.616 that have been put forward to identify the ones that are the most crucial ones. 00:13:04.756 --> 00:13:09.676 What are the ones that have actually the strongest impact on society and that 00:13:09.676 --> 00:13:14.456 we really need to put into place and to ensure that not only they're there. 00:13:14.856 --> 00:13:18.196 Right, because also in the cases of legislations or hard governance, 00:13:18.576 --> 00:13:22.636 in many cases, we do have the good, we know what the best practices are, 00:13:22.716 --> 00:13:25.836 but no one is wondering that these are followed or we know what the rules and 00:13:25.836 --> 00:13:30.976 laws are, but we have little organizations sort of ensuring that these are respected. 00:13:31.196 --> 00:13:34.676 So that is where philanthropy can have a really strong role. 00:13:34.996 --> 00:13:39.116 Blowing away a bit of smoke, you know, in this mirror game, it's too many principles 00:13:39.116 --> 00:13:42.396 that sometimes are just to distract from the real problems. 00:13:42.396 --> 00:13:47.096 Philanthropy should really have a role in supporting research and initiatives 00:13:47.096 --> 00:13:51.876 that can identify the most important principles and then identify the risks, 00:13:52.056 --> 00:13:54.516 the dangers of technology, of this technology, 00:13:54.836 --> 00:13:57.796 and then identify the principles that can mitigate this. 00:13:57.936 --> 00:14:02.916 And on the more positive angles, so identify which are the strongest opportunities 00:14:02.916 --> 00:14:05.876 and then support initiatives that promote these opportunities. 00:14:06.846 --> 00:14:10.986 Yeah. Hubert, do you agree with that? And are there examples of this happening at the moment? 00:14:12.026 --> 00:14:16.266 No, very good points, but maybe a few additions and maybe one with zooming out 00:14:16.266 --> 00:14:17.726 and zooming into academia. 00:14:18.046 --> 00:14:21.846 I know there has been some effort trying to make sense of all these principles 00:14:21.846 --> 00:14:22.826 out there in AI governance. 00:14:23.329 --> 00:14:27.969 And to digest them and cluster them in a few. And I recall one paper, 00:14:28.109 --> 00:14:30.529 it's a few years old, from Jobin et al. 00:14:30.989 --> 00:14:36.489 Going through almost like 80 or 90, I don't remember, principles and distilling 00:14:36.489 --> 00:14:40.149 like five core principles that have been mentioned in one way or the other. 00:14:40.309 --> 00:14:43.769 And these work then like transparency, justice and fairness, 00:14:44.049 --> 00:14:46.209 non-malfeasance, responsibility and privacy. 00:14:46.489 --> 00:14:51.389 But these also develop, I believe, over time, especially now with generative 00:14:51.389 --> 00:14:54.349 AI and other kind of next frontiers. 00:14:54.589 --> 00:14:58.749 But there's been important work there. But I go back to philanthropic organizations. 00:14:58.989 --> 00:15:03.709 So I fully agree with what Giuseppe laid out through funding or other mechanisms, 00:15:03.749 --> 00:15:07.329 indeed, to promote governance in one form or the other. 00:15:07.869 --> 00:15:12.709 And often here hard or soft law or even hybrid approach that many called on 00:15:12.709 --> 00:15:16.849 recently, indeed, to kind of make sense of the principles. 00:15:17.469 --> 00:15:19.509 And as Giuseppe said, to cut through 00:15:19.509 --> 00:15:23.109 the noise and to see which ones are impactful, ones are actionable. 00:15:24.029 --> 00:15:27.369 And what I see, however, like philanthropic organizations to play an active 00:15:27.369 --> 00:15:33.329 role in is to really call out on gaps, to be this sounding board for especially 00:15:33.329 --> 00:15:36.469 underrepresented groups, but also in general. 00:15:36.769 --> 00:15:41.789 And so to call on gaps, out gaps, but also to measure the impact, 00:15:41.889 --> 00:15:45.369 which I think is the biggest challenge in governance today is the implementation, 00:15:45.709 --> 00:15:49.069 the operationalization of these principles and governance mechanisms, 00:15:49.089 --> 00:15:52.849 and also to see and to measure are these working, yes or not?

00:15:53.189 --> 00:15:57.309 Can we observe the positive effects? I think that's the next challenge and we 00:15:57.309 --> 00:16:01.449 need philanthropic organizations among the many other actors to be active and responsible. 00:16:02.169 --> 00:16:07.189 Yeah, and you also mentioned briefly something I'd like to dive into a bit deeper, 00:16:07.489 --> 00:16:12.929 which is that many kind of actions on AI governance are lacking voices from 00:16:12.929 --> 00:16:16.729 the global majority, from communities really affected by AI systems. 00:16:17.389 --> 00:16:22.769 How can then philanthropic organizations help close these representation gaps. 00:16:23.455 --> 00:16:28.175 So there are various ways on how philanthropy can step in here. 00:16:28.395 --> 00:16:32.735 So if we take their unique capacities, their resources, 00:16:33.175 --> 00:16:37.455 their assets, their funding mechanisms, their expertise, the convening power, 00:16:37.475 --> 00:16:44.435 and also some more tech-savvy foundations to bring in tech expertise and actual tools. 00:16:44.435 --> 00:16:49.735 So to bring that to the AI stack, and with that I mean like the hardware, 00:16:49.895 --> 00:16:53.755 the infrastructure, the data sets, the model design, development, 00:16:53.755 --> 00:16:55.415 and the use of these models. 00:16:55.955 --> 00:17:01.515 And I think what we can see also is how philanthropic organizations can act 00:17:01.515 --> 00:17:06.455 in each of these layers to close the representation gap and to make AI safer 00:17:06.455 --> 00:17:10.575 and promote a responsible approach to AI and dominant use. 00:17:10.575 --> 00:17:14.475 So if we drill down into each of these layers, so for example, 00:17:14.895 --> 00:17:19.415 philanthropic organizations can fund initiatives to lower the barrier to access 00:17:19.415 --> 00:17:25.155 this expensive hardware infrastructure to have a more participatory and globally 00:17:25.155 --> 00:17:26.755 inclusive AI ecosystem. 00:17:26.755 --> 00:17:33.195 And if we go down to the data sets to support projects to create more representative 00:17:33.195 --> 00:17:38.355 data sets, I mentioned earlier to include especially raw languages and endangered 00:17:38.355 --> 00:17:39.635 languages and cultures. 00:17:40.075 --> 00:17:45.375 And here we have, to name one, the Wikimedia Foundation having one of the largest 00:17:45.375 --> 00:17:47.615 open data sets out there. 00:17:48.135 --> 00:17:53.835 And similarly, if you go on the model level to also promote projects or develop 00:17:53.835 --> 00:17:55.475 themselves open source alternatives. 00:17:56.016 --> 00:18:00.236 Which are often capital intensive with uncertain revenue streams. 00:18:01.256 --> 00:18:06.376 And then also if you look at the use, so also to promote or engage in advocacy 00:18:06.376 --> 00:18:12.356 campaigns to promote the beneficial use of AI, for example, water scarcity or human rights abuse. 00:18:12.496 --> 00:18:14.816 So how can AI kind of

help there be leveraged there? 00:18:15.096 --> 00:18:20.936 And also to bring attention to governance gaps, put it in the public discourse, 00:18:21.276 --> 00:18:26.656 to call out on representation gaps and make governance more actionable, 00:18:27.076 --> 00:18:34.896 so to avoid the talk-no-walk approach and to be the one keeping everybody on track or accountable. 00:18:35.616 --> 00:18:41.256 And all this to say to avoid that major governments, big corporations are the 00:18:41.256 --> 00:18:44.536 only ones, because they're very important, but they shouldn't be the only ones 00:18:44.536 --> 00:18:48.676 to decide how AI is designed, developed, and regulated, 00:18:49.096 --> 00:18:54.036 but they have a more participatory and indeed an ecosystem, AI ecosystem. 00:18:54.816 --> 00:18:59.356 Thanks. And thanks also for sharing those examples of how it's happening in practice at the moment. 00:18:59.636 --> 00:19:05.176 I also want to look at maybe some of the risks of philanthropic funds because 00:19:05.176 --> 00:19:09.296 we've seen these major philanthropic movements like effective altruism pouring 00:19:09.296 --> 00:19:14.516 a lot of money into AI research in Silicon Valley, often kind of looking at capabilities, 00:19:15.056 --> 00:19:18.956 maybe not looking so much at safety or global equity or responsible development. 00:19:19.156 --> 00:19:21.996 So I want to know your opinions there. Maybe to you, Gillespie, 00:19:22.196 --> 00:19:26.636 first. What do you see as the risks when philanthropy actually funds AI development? 00:19:27.345 --> 00:19:30.505 I don't think it's specific to AI development. 00:19:30.785 --> 00:19:34.685 It's like the same risk that we would see when philanthropy engages in most 00:19:34.685 --> 00:19:37.285 of its activities and supporting initiatives. 00:19:37.545 --> 00:19:41.705 The fact that it's private capital, there is no democratic control, 00:19:41.705 --> 00:19:45.685 so to say, over this private capital that is pouring into different initiatives. 00:19:45.885 --> 00:19:50.265 It could be in education, right? A certain educational viewpoint could be supported. 00:19:50.585 --> 00:19:57.065 In Brazil, for example, there is a large foundation that is supporting a private education movement. 00:19:57.545 --> 00:20:02.145 No one can sort of limit what a private can do with their capital supporting 00:20:02.145 --> 00:20:06.145 different initiatives as long as it's disrespecting the laws and governance 00:20:06.145 --> 00:20:08.685 principles that we alluded to before. 00:20:08.865 --> 00:20:14.165 So the risk is that we have a few decision makers that identify what they believe 00:20:14.165 --> 00:20:17.245 to be a good solution and they try to promote it. 00:20:17.625 --> 00:20:22.685 The risk is that if it's not the right solution, then it could cause some damage, 00:20:22.685 --> 00:20:27.185 but that's a bit up to also both the beneficiaries, the state, 00:20:27.685 --> 00:20:33.945 and the other privates to monitor how this phenotropic capital is used and whether 00:20:33.945 --> 00:20:38.565 the initiative is supported if they are actually creating more damages than benefits. 00:20:39.129 --> 00:20:43.509 I think it's really easy at some points to jump to the negative conclusion and 00:20:43.509 --> 00:20:47.369 take the one bad example of philanthropy, sort of when it's wrong, so to say. 00:20:48.029 --> 00:20:51.889 But not to be forgotten that it happens many more times for four properties, right? 00:20:52.549 --> 00:20:56.969 There are so many negative externalities produced and most of the time, 00:20:57.029 --> 00:21:01.449 philanthropy is actually at least has a good intention to try to provide solutions. 00:21:01.869 --> 00:21:06.369 And we should actually promote philanthropy that takes risks and fails, right? 00:21:06.529 --> 00:21:11.589 What philanthropy is therefore to have money that is used to take calculated 00:21:11.589 --> 00:21:15.769 risks that could bring a very large reward if it works. 00:21:15.769 --> 00:21:19.509 And if it doesn't work, it was capital that others, like a state, 00:21:19.609 --> 00:21:24.649 for example, didn't have to spend in something that was maybe too risky for a state. 00:21:24.849 --> 00:21:29.249 And then if you think of a sort of a term from the business world, 00:21:29.429 --> 00:21:33.889 if you find the unicorn solution to have, but more responsible use of the eye. 00:21:34.549 --> 00:21:39.549 Thanks to philanthropy, that's a great opportunity to take. What do you think, Hubert? 00:21:39.966 --> 00:21:45.446 Yeah, I agree with Giuseppe and especially with the fact that it's not specific to AI necessarily. 00:21:46.086 --> 00:21:50.466 As indeed there have been situations in the past where philanthropic action 00:21:50.466 --> 00:21:55.966 was not focused on the root cause or core issue at hand, but rather on a symptom 00:21:55.966 --> 00:21:58.446 which then amplified negative externalities. 00:21:58.726 --> 00:22:03.086 And here I think it's indeed important to have to adopt like a strategic and 00:22:03.086 --> 00:22:08.446 outcome-based view and to also continuously, which is also a big challenge, 00:22:08.966 --> 00:22:14.006 I know, but like ideally to put effort behind measuring progress so as to guarantee 00:22:14.006 --> 00:22:19.966 the positive and hope for sustainable long-term effect of philanthropic action. 00:22:20.186 --> 00:22:24.786 And also with that, then to work with the public for which philanthropic action 00:22:24.786 --> 00:22:26.906 is meant for the public good. 00:22:27.106 --> 00:22:32.306 So to work with the public to recalibrate also work that maybe doesn't result 00:22:32.306 --> 00:22:34.646 in the effects outcome that was planned for. 00:22:35.106 --> 00:22:40.586 So I think there needs to be some dynamic and collaborative action or mechanisms there. 00:22:40.786 --> 00:22:46.266 And also, I think we alluded to it earlier on the fragmentation of

the philanthropic 00:22:46.266 --> 00:22:52.686 landscape, which I think is another risk where work is redundant or duplicative 00:22:52.686 --> 00:22:55.586 at the expense of positive impact. 00:22:55.586 --> 00:23:01.126 And the last point, I think, which was implied, I think is also indeed something 00:23:01.126 --> 00:23:07.646 can be done or the incentives sometimes are unclear of philanthropic organizations. 00:23:08.486 --> 00:23:12.226 So I think there's also room for improvement to increase transparency and accountability 00:23:12.226 --> 00:23:17.706 mechanisms there so as to have kind of an orchestrated, coordinated approach 00:23:17.706 --> 00:23:20.186 that is outcome based and impact based. 00:23:20.908 --> 00:23:24.588 I want to now look at what all this means for multilateralism. 00:23:24.808 --> 00:23:28.048 We've mentioned a little bit about collectivity. In your research together, 00:23:28.048 --> 00:23:32.808 you look at also partnerships for how philanthropy can have an impact in AI 00:23:32.808 --> 00:23:33.948 development and governance. 00:23:34.328 --> 00:23:38.508 The UN General Assembly just last year established two kind of major mechanisms 00:23:38.508 --> 00:23:42.888 for how we work on AI collaboratively, the Global Dialogue on AI Governance, 00:23:42.928 --> 00:23:47.548 which will meet yearly, and also an independent international scientific panel on AI. 00:23:48.088 --> 00:23:53.168 What is your view? Should philanthropy be present in these multilateral mechanisms? 00:23:53.488 --> 00:23:55.828 And what is their role in bringing different partners together? 00:23:56.287 --> 00:23:59.887 Yeah, thanks. This is a great question. So I think it's particularly important 00:23:59.887 --> 00:24:04.667 that philanthropy has a voice also, especially philanthropy that has been engaging 00:24:04.667 --> 00:24:07.067 with the ethics and the responsible use of AI. 00:24:07.407 --> 00:24:13.567 The reason why this is so important is that when it comes to the digital world, AI, consequently, 00:24:14.007 --> 00:24:19.347 it's really hard to take the traditional logic of setting, you know, my state, 00:24:19.627 --> 00:24:23.387 my rules, my laws apply, you have to do, you know, you have to respect them 00:24:23.387 --> 00:24:27.087 because most of the times, you know, companies are an AI company or digital 00:24:27.087 --> 00:24:30.167 companies based on state X, but operating worldwide. 00:24:30.787 --> 00:24:33.907 So having international standards is super important. 00:24:34.527 --> 00:24:39.207 So I think at the multilateral level, hopefully with a strong UN presence, 00:24:39.727 --> 00:24:42.247 strong government participation is international based. 00:24:42.487 --> 00:24:45.687 There is a charter as we know, right? 00:24:46.007 --> 00:24:50.687 The UNESCO has also proposed principles for analytical AI and philanthropy is 00:24:50.687 --> 00:24:55.927 fundamental there, because the philanthropy is sort of bringing in the

logic of the private sector, 00:24:56.147 --> 00:25:02.427 but a logic of the private sector that is keeping in mind, at least that's what the intention stated, 00:25:02.807 --> 00:25:04.067 keeping in mind, so the logic is 00:25:04.067 --> 00:25:06.607 a private sector, but keeping in mind the promotion of the public group. 00:25:06.747 --> 00:25:13.187 So they can really bring this unique voice and they can also take risk and say, as we mentioned before, 00:25:13.447 --> 00:25:17.207 when we are part of this debate and we recognize we should take this approach, 00:25:17.287 --> 00:25:22.147 then okay, we can provide funding to support sort of the implementation of these approaches. 00:25:22.407 --> 00:25:27.427 So I think it's really important to have gilatropia that is a multilateral table 00:25:27.427 --> 00:25:32.907 because of the unique perspective it brings and the unique opportunities of action it can support. 00:25:33.187 --> 00:25:37.287 I'll ask you the same, Hubert, and maybe also how you think kind of philanthropic 00:25:37.287 --> 00:25:40.887 data stuff that we don't necessarily see at the multilateral level, 00:25:41.027 --> 00:25:45.427 like on the ground information from societies can feed into multilateral work? 00:25:46.160 --> 00:25:49.500 I'm very happy to answer, if I may, just to your question before, 00:25:49.780 --> 00:25:54.380 because I think the Global Dialogue and Scientific Panel are indeed very important initiatives. 00:25:54.640 --> 00:25:58.760 And I think also thriving to have a lot of representation, if I'm not mistaken, 00:25:58.820 --> 00:26:04.800 one of the few or first initiatives to allow for a representation of each UN 00:26:04.800 --> 00:26:07.060 member state, which I think is huge. 00:26:07.700 --> 00:26:11.780 So participation of philanthropy organizations, I think, are indeed twofold important. 00:26:11.780 --> 00:26:14.820 So one on the representation side, look at the civil society layer, 00:26:14.980 --> 00:26:15.900 and I agree with Giuseppe, 00:26:16.740 --> 00:26:20.300 to especially philanthropic organizations that have a track record of a positive 00:26:20.300 --> 00:26:24.180 impact or engagement on governance, for example, 00:26:24.980 --> 00:26:29.500 but also to bring philanthropic organizations on board to convince them that 00:26:29.500 --> 00:26:36.280 it's a cause worthwhile to invest in time, effort, and assets to promote good AI, responsible AI. 00:26:36.280 --> 00:26:43.360 And your question on how data and enabled insights can feed into multilateral work. 00:26:43.500 --> 00:26:49.120 So we published a few years ago a paper on how natural language processing NLP 00:26:49.120 --> 00:26:52.880 can be leveraged to not only track philanthropic activities, 00:26:52.880 --> 00:26:54.540 we looked at Switzerland in particular, 00:26:54.760 --> 00:27:01.260 but also their connections to SDGs and also with that to uncover SDG-aligned 00:27:01.260 --> 00:27:03.800 partnerships to promote

the impact. 00:27:04.680 --> 00:27:08.500 What it helps to have a data-driven and AI-driven approach in philanthropic 00:27:08.500 --> 00:27:13.480 activity is also to avoid, indeed, the redundant duplicative work that is done 00:27:13.480 --> 00:27:15.560 at the expense of impact a lot. 00:27:16.020 --> 00:27:22.620 What it helps is to really leverage the strengths of every stakeholder through 00:27:22.620 --> 00:27:26.020 this more informed way to act and optimize for impact. 00:27:26.360 --> 00:27:29.740 This has been such a fascinating dive into philanthropy and AI. 00:27:30.300 --> 00:27:34.620 Giuseppe, one final question for you. How would you like to see philanthropy 00:27:34.620 --> 00:27:36.980 kind of supporting air development in the near future? 00:27:37.644 --> 00:27:41.564 What I would like to see is to really embrace this 00:27:41.564 --> 00:27:47.084 technology that is having an incredible impact in recent years and to let us 00:27:47.084 --> 00:27:52.304 slow down in thinking how to use it for what in novel fields but to really start 00:27:52.304 --> 00:27:57.924 a debate or to support the ongoing debates on how to use AI properly, 00:27:58.124 --> 00:28:02.044 how not to just use it because you can use it but to only use it for problems 00:28:02.044 --> 00:28:06.584 that are actually going to be strongly supported and complemented by AI. 00:28:06.944 --> 00:28:12.564 Also, keeping in mind all the secondary externalities that come from using AI, 00:28:12.704 --> 00:28:18.464 like incredible use of water and of electricity that's running AI includes. 00:28:18.664 --> 00:28:24.584 So what I would really hope to see philanthropy doing is championing the responsible 00:28:24.584 --> 00:28:30.664 use of AI in a way that it's going to be beneficial at most for society and 00:28:30.664 --> 00:28:33.864 limiting as much as possible the downsides of it. 00:28:33.984 --> 00:28:39.624 But if in five years we will look back and see for what are we using AI, 00:28:39.824 --> 00:28:43.504 how is it complementing our operations and for what we're not using and so on, 00:28:43.604 --> 00:28:49.184 if we manage to find that sweet spot where we are really integrating AI and 00:28:49.184 --> 00:28:52.604 helping us transition and work, for example, better together, 00:28:52.804 --> 00:28:58.664 as Huber said before, leveraging novel tools and the ability to digest incredible 00:28:58.664 --> 00:29:03.104 amount of data to work in synchrony among different philanthropic organizations 00:29:03.104 --> 00:29:07.284 or different organizations that can be philanthropy, 00:29:07.584 --> 00:29:10.924 can be UN agencies, can be privates, can be governments. 00:29:11.124 --> 00:29:17.044 If we were to manage to have a true systems approach to these big challenges 00:29:17.044 --> 00:29:19.784 we have, that will be a successful philanthropy. 00:29:19.964 --> 00:29:24.704 When we really have this sort of network perspective, this relations perspective 00:29:24.704 --> 00:29:29.284 where we're all working together in a collaborative fashion That would be a 00:29:29.284 --> 00:29:33.604 great success for philanthropy, having promoted the responsible use of AI. 00:29:34.104 --> 00:29:37.704 And what would be nice to see is that philanthropy managed to make all the actors 00:29:37.704 --> 00:29:42.204 benefiting or touched by AI collaborate instead of competing against each other. 00:29:42.731 --> 00:29:45.871 Yeah. Hubert, do you agree? I agree. 00:29:46.191 --> 00:29:52.331 And how I would see philanthropy fully integrated is also as an established sounding board. 00:29:52.331 --> 00:29:57.731 So that calls out on gaps, for example, governance, but also of unfair, 00:29:57.951 --> 00:30:03.511 of an unfair impact on certain groups of society that is maybe overseen by decision makers. 00:30:03.511 --> 00:30:10.011 For example, if AI has unproportionate, big impact on a certain job in society. 00:30:10.011 --> 00:30:15.891 Also philanthropy organizations that call for action and that are actively involved 00:30:15.891 --> 00:30:20.931 in initiatives and panels that tackle these big questions of AI governance and 00:30:20.931 --> 00:30:22.771 responsible design and use. 00:30:23.031 --> 00:30:27.131 But also to see a philanthropic organization as an independent voice that is 00:30:27.131 --> 00:30:30.551 also accountable for its very cause. It wants to contribute. 00:30:30.751 --> 00:30:37.911 So to have this end-to-end, a healthy interplay of these different actors. Really interesting. 00:30:38.531 --> 00:30:43.471 Normally we ask our guests also to recommend an open access resource for our listeners. 00:30:43.911 --> 00:30:48.731 Of course, the book we mentioned earlier that you published on the topic is 00:30:48.731 --> 00:30:51.111 a fantastic deep dive and is open access. 00:30:51.451 --> 00:30:53.891 Are there any other recommendations you'd like to share? 00:30:54.531 --> 00:30:59.011 Maybe on the more ethical side, I think there is a fantastic book called The 00:30:59.011 --> 00:31:03.411 Green and the Blue, Digital Politics in Philosophical Discussion by Luciano Floridi. 00:31:03.651 --> 00:31:09.211 It's a fascinating book, really how AI is changing the perspective of how we 00:31:09.211 --> 00:31:14.651 can and should interact and how we should interpret AI and how through this 00:31:14.651 --> 00:31:18.491 perspective, then we would find a better way of using the tool. 00:31:19.051 --> 00:31:22.631 Thank you so much for sharing. We'll make sure both books are in the show notes. 00:31:23.951 --> 00:31:29.591 Thank you so much for joining us for this podcast. Thank you for this wonderful 00:31:29.591 --> 00:31:31.471 opportunity. Thank you.