AIxMultilateralism: Philanthropy's Role in Inclusive, Ethical AI

WEBVTT
00:00:15.702 --> 00:00:21.722
I'm Natalie Alexander. Welcome to AI and Multilateralism, a playlist at the
00:00:21.722 --> 00:00:23.082
UN Library and Archives Geneva,
00:00:23.422 --> 00:00:28.922
where we're joined by experts who help us unpack the many ideas at the nexus
00:00:28.922 --> 00:00:31.482
of AI and international cooperation.
00:00:31.482 --> 00:00:37.042
Should philanthropy be a part of the development and governance of AI?
00:00:37.422 --> 00:00:42.482
We're exploring this question with Hubert Halopé and Giuseppe Ugazio,
00:00:42.702 --> 00:00:47.602
both part of the team at the Artificial Intelligence and Philanthropy Project
00:00:47.602 --> 00:00:49.542
at the University of Geneva.
00:00:49.542 --> 00:00:54.082
We look at the potential for philanthropy to support more representative,
00:00:54.522 --> 00:01:02.042
inclusive AI, as well as how it can drive forward AI governance from something written to reality.
00:01:16.642 --> 00:01:19.982
Hubert and Giuseppe, thank you so much for joining us on the series.
00:01:20.402 --> 00:01:24.522
Thank you for having us. Thank you, Nathalie. It's great to be here with you today.
00:01:24.942 --> 00:01:27.962
What we're going to cover today, we haven't covered in this series before,
00:01:28.182 --> 00:01:32.462
and that's the link between AI and philanthropy. And you're both part of the
00:01:32.462 --> 00:01:36.722
team in what's called the Artificial Intelligence and Philanthropy Project at
00:01:36.722 --> 00:01:37.882
the University of Geneva.
00:01:38.342 --> 00:01:43.802
Giuseppe, firstly to you, why does this project exist? We started this project by necessity.
00:01:44.042 --> 00:01:49.642
We started this project as we needed to use AI in order to understand a bit
00:01:49.642 --> 00:01:53.722
better the philanthropic landscape here in Switzerland and abroad.
00:01:53.722 --> 00:01:58.002
We realized there is not really too much data, too much information around to
00:01:58.002 --> 00:02:01.302
understand, you know, where do philanthropic organizations work,
00:02:01.322 --> 00:02:05.442
in which areas, for which reasons, which operative modes.
00:02:05.908 --> 00:02:11.988
And so we started adopting AI tools to support what we could find about philanthropy.
00:02:12.288 --> 00:02:15.628
And then with this data, we had to make sense of it. And as you may know,
00:02:15.728 --> 00:02:19.368
here in Switzerland, we have over 17,000 philanthropic organizations.
00:02:19.688 --> 00:02:23.308
So for a couple of researchers to make sense of all these documents,
00:02:23.328 --> 00:02:26.388
it would take forever and probably be quite imprecise.
00:02:26.548 --> 00:02:29.428
And by the time it would be done, probably this would be already outdated.
00:02:29.608 --> 00:02:34.408
So we turned to a specific form of AI. So we will use natural language processing
00:02:34.408 --> 00:02:40.208
to extrapolate some of the key trends, some of the key words we could then make
00:02:40.208 --> 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.