The following are the outputs of the captioning taken during an IGF intervention. Although it is largely accurate, in some cases it may be incomplete or inaccurate due to inaudible passages or transcription errors. It is posted as an aid, but should not be treated as an authoritative record.
***
>> MARK IRURA: Good evening, good morning. Hi, everyone. Thank you for joining our session. My name is Mark Irura. I will be moderating this session. I think we will start with introductions. I will introduce the panel. We have three participants who are online and three on stage.
I will start with Deshni from Fair Forward, from South Africa, Artificial Intelligence for All. The secretary for Technology and Impact. She is especially proficient in skills of prototyping local language within the local community, scaling it to science. Hosts bootcamps conducted for women conducted across three African countries. She has co‑developed South Africa's AI Maturity Assessment Framework. She has especially worked on Language-Accessible Policy Hub for AI Policy Playbook with Global South policymakers. Desh describes herself as a bridge‑builder and co‑creator and advisory board member of the South African AI Association, co-founder of Africa-Asia AI Policymaker Network, working group members on AI strategy recommendations for South Africa and featured in “100 Brilliant Women in AI Ethics.”
Next I will introduce (?) on my left, Principle Researcher at Digital Uganda, Voice Technology for African languages and is AI and Open Data organisation on a mission to democratize access to information in African languages. Founded in 2018, the company builds large‑scale voice and text datasets and develops voice AI tools to bridge the language divide and preserve linguistic diversity. Spanning 17 African languages, they have recorded countless hours of speech samples for models for global impact. Grounded in the tradition of Uganda, community uplift through collective uplift, Digital Uganda unites community, developers, governments and NGOs to build open‑source language infrastructure by Africans for the world.
On the far right, on my far right I have Dr. Lilian Diana Awuor Wanzare, lecturer at department of computer science at Maseno University. Her such interests are artificial intelligence and machine learning. In particular, natural language processing. You will hear the term LLP a lot in this panel. Building text processing tools for resource languages. Served as principle investigator for several research projects funded by BMGF, the Lacuna Fund, Canadian Development Aid & NLP, which is a Kenyan language corpus for NLP and learning research, a project that looks at building datasets for training and NLP tools for underserved languages, particularly Kenya, with use cases geared towards agriculture, language and health, particularly sign language, particularly Kenyans, using virtual signing. She holds a PhD in Computational Linguistics and MSc in Language Science from Salem University in Germany.
Online, I will start with Dr. Melissa Omino. Melissa is director of the Centre for Intellectual Property and Law, CIPIC, leading the Policy Hub and Data Policy Centre. Her research direction is focused on utilizing an African lens and human rights lens. Part of the research conducted under her leadership involved mapping AI applications in Africa. As initial step in answering the question of what determines African AI and problems African AI should aim to solve. Dr. Melissa Omino is an Intellectual Property expert and served as advisory board member in several projects that intersect between AI and IP. This also includes driving a national strategy process and she's led an advisory for global entity funding AI research in Africa.
We also have Elikplim Sabblah. He is a technical advisor working for the Fair Forward Programme within the Digital Transformation Centre, DTC Ghana, and is a German technical corporation. In this role Elikplim focuses on AI policy, accessibility and capacity‑building to foster inclusive and sustainable AI development in Ghana. Elikplim's worked on the development of Ghana's National Strategy, collaborating with the Ministry of Communication, Digital Technology and Innovation through the commission. With a strong background in data science, monitoring, evaluation, project management and stakeholder engagement, Eli is working towards enhancing accessibility, local innovation responsible AI adoption in Ghana.
Last, but definitely not least, we have Ms. Viola Ochola, director of Access to Information. She's an advocate for the high court of Kenya and legal practitioner with administrative law, (?) human rights. Her experience spans over 15 years. And holds an MBA in strategic management and extensive experience in public and private sector. She is the immediate former ‑‑ Complex Investigations and Legal Services and Commission on Administrative Justice, Kenya. Viola is an Open Government Leadership Fellow and member of the technical committee on Open Government Partnership, the Kenya chapter, in her capacity as lead to access information commitment. She is passionate about open governance and empowerment of citizenry to access services and benefit from opportunities offered by government.
The reason I have gone through the elaborate introduction is for you to know who is talking to us in these topics this evening. And also for you to look up the panelists and reach out via LinkedIn and ask questions and connect and continue to engage on the topic.
Our topic today is Exploring New Data Governance Mechanisms for Language Day to Driving NLP Systems in Africa.
The issue of licensing of language has already come up in various workshops. Today we want to have a more practical discussion that looks at research that is currently going on in this topic. Language is culture and culture is identity. Yet the digital identity of Africa is skewed, manipulated, commercialized. The language data collection is characterized by a significant disparity between large‑scale publicly accessible resources and numerous small, isolated projects.
The Mozilla Foundation seeks to positively impact the way in which local language data is viewed, collected, scored and utilized. Currently Mozilla Common Voice is the most diverse multi ‑language open speech corpus, holding more than 30,000 hours in more than 180 languages and example of successful community initiative that is also a digital public good. It is a community platform, as well as lab for linguistic inclusion and for traversing data governance issues in NLP. But there has been an awakening and sentiment change among the language communities. This is what you will delve into today. Speaker who's have datasets and some issues including equitable investment, locally sensitive community control and the dynamics around power and building the technology.
So having set the background for the problem, we are going to highlight the unintended and intended effects of the CC0 Open Public Licence on communities and language data. And we want to look at governance and policy, how do they intersect and what are some solutions that are being worked on to try and resolve that, the problem.
I will go straight to you, Lilian. I will begin with a question on how can AI training data licences be adapted to protect cultural serenity and ensure equitable benefit, especially for those who have been marginalized.
>> LILIAN DIANA AWUOR WANZARE: Thank you so much, Mark, for introduction. When comes to AI, data is core; and when comes to (?), language and language. As Mark as mentioned here, is really more than just, you know, group of one. It embodies aspirations of community and culture of different communities. If you look at that and think about this data, how can it be licensed in a way that still promotes the cultural values from where they come from, okay. I think about it as one community sentence. How do we go about collecting the data from the community themselves, okay? How do we manage the use of this data along the journey it's been used in NLP systems.
In community centre, there are a lot of things that go into it. One is constant. As they are going to provide the data, as they properly informed. This is a continuous process. An understanding of the journey of the data as goes around, being developed and moves across, as being used for different systems, okay. Now how do we balance the issue within the licensing, the issues of open sharing vis‑a‑vis benefit sharing, okay? Those things should not be mutually exclusive. We can still have open sharing and benefit to share. How can this be embodied within existing licencing to have both? In such a way that here is, we still do open sharing to facilitate development of tools, development of, you know, systems that promote the language.
But still, from where the data comes from. It is no longer (?) Have benefit of tools are going to be developed from them. How does this move not just from this community but the larger language community. Those are tools in themselves but the larger community holds the particular words, languages. Think about it in the last bit as I close, in this licensing ecosystem, another different ways to look about licensing and licensing in general, in this ecosystem how transparent is it to different views on different come nations that support requirements be able to really pull things together that are aligned to your values.
And there is no one‑size‑fits all. There can be different ways that still supports the community centre, the communities, but still allows for open sharing and development (?) That would be my opening remark.
>> MARK IRURA: Thank you so much, Lilian. I will come to you, Melissa. Lilian has mentioned something to do with the different needs and different requirements over the entire, let me call it, language or AI language life cycle. She's talked about values. And to it I want to throw in the benefit.ly not just make it economic but when we think about benefit. I would like you to help us unpack that in view of the question, like how do we think about sovereignty. But also enable these things, this kind of walk you are currently undertaking.
>> MELISSA OMINO: Thank you so much, Mark. Hope you can hear me.
>> MARK IRURA: Yes.
>> MELISSA OMINO: Excellent. When we think about the benefit, I don't think we here should be discussing the benefit without referencing the language community. Because that is where the benefits should flow. So in part of the work CIPIC is doing with University of Victoria is reaching out to language communities, and I'm being specific and this term, language community, because there is also a data community that exists and is African data developers who actually collate into the sets for natural language processing.
We think that community, the language community, should be able to speak for itself and say what type of benefit that they would require for the particular use of that language dataset. Has already been mentioned by Dr. Lilian the different types of uses might require different types of benefits or might actually require a different thought as to what a benefit would mean. A lot of resistance towards having these language communities speak about quote unquote benefit is it is automatically assumed to be a monetary thing or a royalty‑based thing. But essentially, we are saying it should be given up to the community to decide what that should be. Most my discussions with various communities, including the (?) Via the (?) University, wants something sustainable and community‑based, everyone can interact and benefit from. A monetary or royalty benefit doesn't quite meet that mark.
So essentially, we need to think about the harmful dynamic created with the current use of language data sets and fact these being commodified in AI systems primarily serves dominant languages and wealthy corporations, while marginalized communities receive no benefits, no matter how you define it, and often leave their cultural protocols, values, practices violated. So a benefit could be the respect of the cultural knowledge the language carries or even a shared or access to this AI tool that's been built using the language data.
So can a licensing framework deliver this? I think it can. That is what the new dual licence of (?) digital licence is supposed to do, provide avenue where this conversation about what type of benefit would flow to a community would start from. We are sort of trying to fit it into what currently governs the language data set regime, which is copyright licensing. So we came up with an alternative licence with elements of copyright but also elements of recognition of cultural knowledge and giving a voice to the community to negotiate about what they would want as a benefit.
Here I would have to signal the Creative Commons community, where I am a board member, who just yesterday released publicly their work on preference signaling that would work hand‑in‑hand with Creative Commons licenses, giving data stewards of sets used by AI to be able to say what they would prefer that data set to be used for or as. So this is actually signaling that this act of benefit recognition, benefit‑sharing is something worked on and needs to be worked on. And maybe it is not for us to determine, because it would just be us imposing our thoughts on these language communities, but bringing language communities to the forefront so they can speak for themselves as to what they would like, thank you.
>> MARK IRURA: Thanks, Melissa. I appreciate specifically your comments and would benefit obviously one of the things that is a point in the continent is the issue of avoiding, you know, the colonisation through language and through AI. I think this is an important question to ask. I will come to you and ask you about policy.
When we think about policy and think about policy framework, what are good principles we can incorporate to think about, you know, equity and anti‑extractiveness, so there is mutual benefit. We do not stifle innovation, as Lilian is saying, but grow in advance. We still need a Commons to be able to move forward.
>> MELISSA OMINO: Thanks, Mark. I think in other view equity, we are required to think about communities as having ownership and not just a group that would provide consent. Ownership and consent are two completely different things. The traditional data sharing regime treats this as sources rather than partners. This extracts value while leaving these communities with just the risks and harms. So there has to be a shift where there is community data sovereignty and I think (?) Has alluded to this and you alluded to that, Mark, where we legally recognise communities as collective data stewards with inherent rights to govern data about their members, territories and cultural knowledge, which is where language would fall into.
Individual consent is not enough when data affects an entire community. We need graduated consent that requires community consultation before individual agreements. The community gets to weigh in on whether that serves collective interest and get to voice what collective interests are. This includes verification rather than one‑time permission and complete transparency about who is benefitting and how and also deferring community veto power over harmful applications.
If someone profits from community data, the community must benefit too. This means a mandatory benefit‑sharing requirement where communities might get a percentage of profits, if that is what they want, or might get capacity‑building investments in infrastructure, education and priority access to products developed using their community data. This is not coming from me, this is from consultations I have had with specific community members.
So in order to prevent exploitation or to make this shift to these new ‑‑ utopia I'm speaking about, we need strong anti‑extractive safeguards so data should not be shared without going back to the community for permission. Communities should be able to reclaim their data and take it elsewhere if they figure like, which requires regular audits shared publicly within months to have accountable. All should have penalties for violating community agreements. I must admit here my bias as a lawyer. I'm really thinking about legal frameworks and structures, so that is why I'm talking about accountability enforcement mechanisms.
I think that works currently in language data‑sharing regime because they are using agreements being copyright licences to govern sharing of this data. So the conversation ‑‑ rather what I'm trying to highlight here is it is automatically about power and not just viewing data as a tool and a data governance regime, so not just about privacy. It is really about where is the wealth and power concentrated and how can we then distribute this in an equitable manner. So legal frameworks would be one of the policy considerations that I would think of, but I also think that governments, when coming up with their AI strategies and policies, which a plethora of those have happened on the African continent. They need to centre culture as one of the main pillars of their strategy. I know the Kenyan strategy does that. It does mention culture is an important factor. It does mention responsible and ethical AI, which this would be a pillar this conversation would fall under. It also talks about model development for problem‑solving on the continent. You cannot talk about model development for problem‑solving if you do not think about language datasets.
So I think that this is essentially how we can get to a balance. It is not about closing off the data; it is about ensuring it is an equitable exchange between those who want to collect and use the data and the communities that have preserved and curated. Again, I say there are two communities that exist: Language community, suffered historically in procuring the language particularly in context of Africa; then data community, who put in effort, who views their skills and knowledge in creating these data sets and who have an interaction with those who fund these activities.
So there needs to be a balance for ‑‑ let's talk about these three parties in this context. Those who would like to use the dataset, those who have curated the languages and preserved them and those who have created the dataset.
>> MARK IRURA: Thanks, Melissa. I'm looking at you. Melissa has taken us to utopia, to (?) But coming back to what exists now we could latch onto. Even us as Deshni gives her remarks, I will ask you to ‑‑ ask you, Viola, to be on standby to give us a different perspective, if there is anything Deshni have missed out on, so over to you, Deshni.
>> DESHNI GOVENDER: So I think it is important to point out when we mention the concept of extractive practices, it is not always foreign versus local context. So cross‑border issue because I think these practices often happen within countries in the continent under the guise of open collaboration concept. Do I think policy protections that cover digital work should also actually take their foundational basis from existing protections that afforded to cultural and indigenous communities which exist in a civil context. So assuming those foundational building blocks exist, then policy protection can almost come into play in two ways. Sorry. Policy protection can come into play in two ways, which is as a source for human rights, because that is really important protecting labour rights and gig worker who's often do the un‑sexy work of labeling data, of training algorithms, but also coming as a counter‑leverage point in context of open source and digital public goods. We have heard the speakers mention the concept of quid pro quo, if you take something, give something back.ly just run through very quickly a few points. So fair sharing is one way.
Then my co-panelist Melissa mentioned the Noodle licence but the Incuba licence developed. Another way is if a commercial actor has to cross‑subsidize public maintenance in an open source ‑‑ for Open Source AI resources, what would that look like? Does it come with conditions? But the use of open grounds or language‑term partnerships that actually benefit the community. One example was a grant Google did, had given to Ghana NLP with few conditions that the community could use as they saw fit. I think the other one for that AI policy could include, which doesn't often happen and should, is having where there are foreign investors or foreign partners, including local partners equal collaborators. Oftentimes local partners come in as just consultants. When you have equal collaborator you have co‑owner of Kopura (?), often done by MOUs or just general contracts.
I think policies should make AI developers accountable and could look like impact reports or independent audits. I will mention quickly something I came across, before I hand over to Viola. In my research something that is called the Nagoya Protocol. This actually exists in the biodiversity space, requiring fair and equitable sharing of benefits in the use of Genetic Resources, like plants, animals, microorganisms, et cetera. I feel if we want to learn we could learn if parallels like this. Establishing something like the linguistic protocol for use of African languages in AI could be a great policy tool for regional principle, codes of conduct. I guess another could be the AI policy playbook launched at the UNESCO conference a few weeks ago. I will stop here.
>> MARK IRURA: Over to you, Viola.
>> VIOLA OCHOLA: Thank you, Mark. After speaking after Melissa and Deshni, most have taken out, most of policy requirement, but I would still emphasize the data sovereignty and equal data‑sharing that both Melissa and Deshni talked about. The local African community should be able to control their data from the point of collection and up to the point of usage of those AI technologies so that they are able to be part of the process.
So the whole process has to be inclusive. They should not just be there at the point of information give us or data give us but should be involved in the whole process. And Melissa mentioned she is a lawyer so would be biassed around the legal framework. I will also speak ‑‑ I'm also a lawyer. The legal framework around collection of this data has to be very stringent, has to be very robust so local communities are protected from possible exploitation, from the external, big tech, so to speak. So even at the point of using the benefits, whatever way they may define these benefits, they are able to benefit from that so that it is not an issue that they feel are being exploited.
Quickly, the aspect of community ownership should not just be something that is entrenched in the law but should be actually mechanisms operationalized within the ecosystem, within the African countries so these local communities can be reached. Because sometimes you'll realise some of these communities that are in very remote areas in the African continent, and sometimes even in terms of their digital infrastructure, they cannot even access some of these benefits or some of these ‑‑ what the external parties want to develop.
So it will be important for the governments, at least African governments, to ensure the infrastructure is available so that these communities can be able to reach out to these quote/unquote investors who might want to develop this AI technologies using the languages.
With that, like I said, the engagements have to be very meaningful. It shouldn't just be like one of my co‑panelists said, something that you are just called to give information or to give data. You have to be aware and understand what exactly you are giving out and the possible repercussions of that. And finally, I will speak to another policy perspective. That one of building the capacity and skills development of the African nations. Because you realise sometimes the issue is the lack of skills and the lack of the capacity to do this within the continent.
So it is important for the various policy frameworks to be able to put in place possible training solutions or skills development strategies so that some of these technologies are home‑grown and home‑owned also so that you then now even develop a framework from which you can transfer the knowledge locally, beyond just waiting for the external parties to come in.
And this is not necessary to be done within the country. You can also collaborate with the big tech to be able to develop the skills within the continent. Their skills will be developed from there. So I think I'll stop there, thank you.
>> MARK IRURA: Thanks, Viola. Also thank you for, like, such a broad response. You covered infrastructure, you covered capacity‑building and this speaks to an equal system approach. Like you can't just develop infrastructure only. You can't just build capacity. You can't just develop policy.
So I will look at you, Sam, now. When we are coming up with national AI strategies, the goal is think about where you want to go and what do you want to achieve. I will also ask you, Eli, to share experiences from Ghana, since you have gone through this cycle. I will start with you some. It is an abstract question but also a simple question. Very simple. Can governments support community‑led governance? Could government partner ‑‑ it is always top‑down, it is always, this is what you need to do. What do you think about these strategies to help support the growth of the AI-unique system coming up.
>> SAMUEL RUTUNDA: Thank you, mark. I think the AI strategies or AI policies, they help within these three categories. First they raise awareness. Usually once something becomes a strategy or a policy, it makes people to know about it. So AI, once it is implemented, people are looking at all the components of AI, which currently the major one is the language component. Second, it creates a working framework that governments and other entities can use as a guideline or as a framework to follow. Then it also adds some accountability because they have to explain something. This helps us. Where in the absence of that policy, this could not have been.
Then in terms of what it create, it starts creating a discussion. I mean now when you go to them, you can have a base of how you can discuss, some place from where to start the discussion and they can look and say oh, we have a plan, policy or a strategy and this is what it says. Then the thing about languages cross‑cutting, and touches many aspects of everyday life and starts creating synergies. So for example, someone in health can say oh, actually we are thinking of using this tool. But they don't know how to do it. Given there is a policy, they have where to ask. It is that even as a community start saying how about we work within the health. For example, medicinal plans. Is it something we can capture within or languages.
So it creates synergies and collaborations. Then ultimately, the goal is to raise resources. With these discussions and with these collaborations how as a country we start streamlining how we raise resources, because there is a need to raise the resources. Yeah, I think that is what I will say.
>> MARK IRURA: Thank you, Sam. I will invite you, Eli, to also contribute to that point. Bearing also we have a global audience and we have also ways that we are trying to see and build this ecosystem in a way that others could learn from us.
>> ELIKPLIM SABBLAH: Right, thank you very much, Mark. I continue to say governments should definitely support our communities to take ownership or lead data governance so far as language data sets is concerned. Government should actually empower local communities. By thinking about the idea of national strategies, the AI policies and AI strategies. Then looking at the way that the (?) Has been drafted it includes which ‑‑ includes local communities and major stakeholders. So just by that definition through stakeholder consultations and ecosystem analysis and research, SWOT Analysis, all that process should already include communities that are existing in the space. If that is the case, then in the first step and we put in the communities take ownership of whatever comes out of data ‑‑ what data governance is concerned in a particular country.
Now what I have learned in Ghana is currently we have our draft national AI strategy and undergoing review. Throughout the review processes, we reach out to various groups, trying to understand their specific needs and what they would like to see in the review document. It has been consistently spoken of how they need to see representation in there or how they have to ‑‑ or they would like to be empowered to be able to given datasets governed within same space.
In the draft, there is a pillar that actually speaks to this. Pillar 5 which says strategy seeks to provide data collectors with guidelines and principles for collecting data, storing and sharing it. I think this creates an avenue for government to empower local communities to take the lead or ownership as far as governance is concerned. If the government ‑‑ if their strategy would actually pinpoint specific principles and guidelines that these communities need to take. That would eventually influence how ‑‑ the level of ownership they be able to take of data governance system in the country.
So I think a lot has been said already. We also need to take a look at adopting alternative licences and models. Noodle has been mentioned by Dr. Melissa on the call, in this session. And this position work out well for all the communities involved.
>> MARK IRURA: Thanks, Eli. I think this is ‑‑ this is something also that all this comes up with me. And this morning in a session I had it. So I want to put an open question to the panel. So we've talked about rules, regulations and not talked about money. Some from this panel asked the difficult question about money. A lot of the challenges even that came up earlier was the procurement systems. Is there ‑‑ because procurement provides an opportunity for these communities, develop communities that Melissa mentioned. Even talked about like people who are in remote areas, they cannot benefit because there is no infrastructure, no connectivity. So to this panel and anyone who might have a thought on it, the issue about public procurement and ability to procure innovation. That conversation with government. Not just in Africa but globally, because I think that is also an issue.
Do you have any reflections on it? We have presentations from government but will not put her on the spot but tell you anyone who has a view. Like what could we do in this regard. So that even as we talk about governance, procurement becomes an issue, thinking about procuring this. Any thoughts?
>> SAMUEL RUTUNDA: Let me start. Usually, I don't know, I was talking to someone and say government is run by accountants. Accountants, they want facts. They want what is this going to do. Then still in the early stage of language technology, but particularly within our domain, especially for low resource languages so it is difficult to show the facts. It is something to say oh, I will take a chance and then I will see. Yeah, but I think there is a need to take a chance. For example, when we worked the beginning with Common Voice for Rwanda, there was no policy, no AI ecosystem, there was nothing. Then there was a leap forward to say, okay. Let's take a chance.
Now six years, I think, 30,000 hours have been collected, I think. There is at least, last time I checked, one in ten African languages that were done. So there is a need to take those chance. But then that requires us to talk to people and to convince and change mentalities to say okay. This is what happened. Then another thing, currently I'm also looking, although we are talking about language, we should look at the settings. For these technologies to be used, there is maybe access to the Internet or the digital literacy and others, so I will have to look globally, but there is a changing of mind sets to deploy some use cases and learn from it before having first having proofs so you can deploy.
>> MARK IRURA: Thanks. Anyone else with a view?
>> DESHNI GOVENDER: I think I would come in for, one of the things we know about African language or NLP from indigenous languages, a lot of time it is oral. Particularly for African languages but for other cultures. The problem with having culture or language that is intended for oral knowledge, it means it is also shaped by tone, shaped by cadence, by who is telling the story and the meaning attached to it and also communal use. The problem is it creates a little bit of an NLP design flaw. For example, a design challenge in how do you actually codify knowledge that is not as easy as taking something that is, you know, a book then making it digital.
So the point I'm trying to make is when we talk about procurement and what it is we need to do, we need to understand what asset we are actually working with. It is kind of hard to understand the asset you are working with if you are not even sure how to put it into create an asset value or ‑‑ you know it is an asset but don't foe how to make this tangible and in a form that somebody says oh that, is actually interesting, I'm willing to invest in it or willing to do this or that. It is the difficult part of trying to actually unpack that and then unpack it properly and in a way that you actually same and preserve and protect the cultures and nuances that come with trying to take this raw material that is an asset to the people but then make it a tangible and international value that you can say cool as a country we have this and this and now how to use this as a tool to come in for infrastructure development, knowledge sharing but still protect the people.
>> MARK IRURA: Thanks, Deshni. Melissa.
>> MELISSA OMINO: I'm going to ask you a very lawyerly question, when you talk about procurement, are you talking about funding? When you think about procurement, I think of funding. In the local context, I really think the challenges on government, to move away from looking to other people to save us. I'm stealing that from Dr. Albert, the keynote at COSA. He said no one will save us. We need to think of ways we can locally invest in natural language processing so we can then call the shots or really have the terms, put down the terms of how the language data would be used. I think this is something that government is very much aware of.
A lot of conversation around the Kenyan AI strategy is how will it be implemented. The Kenyan government made decision to keep the implementation plan away from public purview but there is a plan there. There are key performance indicators there and there are key partners identified to help with that AI implementation strategy.
Essentially the conversation we are having, we are at the beginning cycle of natural language processing and the next person in room can say that. We need to talk about collection when we talk about language data and building models that will utilize this language data. That is why we are up in arms about having that open and free‑for‑all, because it will minimise the ability for local companies to invest in that language data and build models, because the market will thoroughly thrash them. If you are talking about market economics, demand, supply, et cetera, which also as a lawyer I might not be very good at. That is the end of my disclaimers.
So I think when we talk about procurement, we need to talk about funding and also stop looking outside. We need to think about locally on the African continent how can we fund. At the key AI Summit this year there was a conversation about infrastructure, about having data centres, which is very integral to how do we control who can access and use the data. There was a conversation about starting to have particular data centres in particular regions. The question was, will it be accessible to African developers or creating centres for others to use on the continent in order to be compliant with data governance regimes.
So I would say for public procurement to make sense we need to first think about funding. To think about funding we must challenge local investors to put their money where their mouth is and invest locally. Not just in data collection but in the development of models. Because, as far as know, nobody outside is actually funding the development of models in a path to truly have African AI, thank you.
>> MARK IRURA: Thank you, Melissa. I will come to you Viola. If you are online and have a question you'd like to pose, please put it in the chat. Over to you, Viola.
>> VIOLA OCHOLA: Thank you, mark. Mine will be quick. Melissa has talked about funding because, you can't talk about procurement without the funding bit. There is the other aspect of procurement, the process. I believe that is where the challenge you are speaking on was. The question is, does even the procurement officer understand what it is.
In government, where I am, there is always a process. In Kenya there is the Public Procurement Act that outlines the process. Part of the process is you need to give specifications and you say this is the end product. Sometimes the procurement person is not aware of AI, let alone even, you know, any other thing. So it will be difficult for such a person to even appreciate where you are coming from if you are to procure this.
So maybe as a way forward, and I know Kenya has developed this strategy. It is very fresh. It was launched in March, at least I will tell you. We need to perhaps just build the capacity of some of these key offices. For example, the procurement of government so they are able to appreciate this may not necessarily be tangible item we are looking at but could be something else. So that is number one. Number two, because the laws as we have them now do not appreciate such thing, we need to review the laws so that they capture these angles. This law should not only be reviewed by lawyers.
Melissa knows, you need to have the technical capacity to be able to put it in the laws in a way that it will inform what you want to get at the end of the tunnel. So I think I will stop there with respect to procurement, thank you.
>> MARK IRURA: There is a friend of mine who says for government procuring a bucket of milk and procuring (?) It is not supposed to be like that. I will come to you, Eli. I will ask a question. Let me start with you, Lilian, because you work with communities. What sort of skills would communities need to build in order to govern their own language technologies effectively. Like so yes, we are saying government ‑‑ these are issues, right. Before they come to ‑‑ they did a number of skills, so we can even talk about the governance of the language. What do we need?
>> VIOLA OCHOLA: Thank you so much, Mark. It is interesting because as we walk through this community ‑‑ we have this Mozilla Project that we go to the community and understand what exactly they want in terms of language to be used in AI. When you pose this to them they like, it is just my language. Not exactly something bigger.
Actually, there is a lack of exposure or lack of knowledge in terms of what can be used. But if you start unpacking the possibilities, then they say ah‑ha, lot of interest to work with and seeing how their children and future can benefit from it and seeing less aspect of capacity‑building. One, they need base‑line knowledge. From people collecting the data, as Melissa said the system. The people collecting the data, as they collect it, are they aware how do we package it and fit objective X, AI model there. Not just how we packet the collection. Then in terms of people who are going to develop the particular models, okay, do they understand how the data comes in or used or just sit down and hope data centre because we find students are looking for data but have no idea how data is collected. They have to be able to think about based on this problem, how is the whole pipeline and wanting the government come into play and think about if you want to govern this system, what really. For them, they really not have a governance framework.
If somebody wants to use all that, how do they come in? If I want to share my data, how do they come in. If media wants to share the data, how do they come in. All these data generated, how do they come, the benefit structure. You can see then they really don't know what comes together to develop these models. There is a disjointedness in terms of how this comes in vis‑a‑vis model. Somebody once asked ,I want a model to help me in chemistry and read some of what you are seeing. First of all, do you have chemical (?) before we test? To start, the utopia of this thing is magical but no understanding of how do we get there and how do all stakeholders come into play to make us get there, maybe that needs to be put into it, thank you.
>> MARK IRURA: Thanks. Eli, I will ask you -- maybe almost to wrap it up or talk a little bit about anything to do with community work, right. Since we are at this place where we are thinking about governance, of products developed for and by these communities and probably in collaboration with them.
>> ELIKPLIM SABBLAH: Thank you, Mark. For the past responses from other panelists, one theme is connecting all of it. You can hear a lot about maybe outreach and community sensitization and all that. I think you have to understand that ‑‑ some definite skills have to be built. We need people who in communities understand digital rights, who understand the importance of data and who understand linguistics or skills in linguistics to be able to maximize the opportunity that this technology brings to their communities.
Now one of the things I have come to understand is that sometimes there is community fatigue regarding contributing today that collection schemes. So the sensitization would make them understand and immediate benefits in terms of monetary terms or whatever in the immediate sense or ‑‑ but it goes along with contributing to something bigger that can actually benefit them immediately and also the nation as a whole. So I think it is important for us to understand the need for outreach programmes, to reach out to people in communities to let them understand artificial intelligence ‑‑ we did research trying to understand how women (?) are using AI and LP tools to interact with their customers and partners and all that. We came to them, and most of them are probably using tools that have AI algorithms working but they don't even know. Also some express a level of fatigue, as I already mentioned, that tired of contributing to data collection schemes and stuff like that. But we actually need people with indigenous knowledge and indigenous experience to exhibit to these things.
One other thing I wanted to also point out is the need for us to let these models we are developing on the African continent to represent African culture. One culture is shared ownership of resources. When you talk about African culture and oral condition, proverbs and expressions and stories don't have proprietary ownership; it belongs to the community that. Should be reflected in models in data collection activities so the data and models are open access to all. I think I mashed up a lot of things but basically that is what I wanted to end with, thank you.
>> MARK IRURA: Thanks, Eli. So I will ‑‑ a question has come. We have just run out of time. Lillian, in one circle, how to bridge between building capacity for local communities in AI beyond collection and increasing usage of AI models within those same communities. 30 seconds, please.
>> LILIAN DIANA AWUOR WANZARE: It is about partnership and collaboration. We have local ecosystem, funders, internal players, how do we come to the collaborative view to make this possible. Cannot be distracted, that is the whole effort within the whole ecosystem.
>> MARK IRURA: Thank you, thank you. I don't want to recap what has been said. I began with an elaborate introduction of everyone. Maybe I didn't introduce myself properly, I'm Mark Irura, I work with the Mozilla Foundation. You can follow us online. Each one of us, you can press subscribe button and like ‑‑ no, subscribe to LinkedIn and feel free to ask about the work and this work and about what they are doing. Thank you so much. Thank you so much. Thank you for being part of this panel, we really appreciate it, thank you.
