Bridging the Intelligence Divide: AI Capacity Building and Knowledge Sharing for the Global South
This discussion focused on the challenges and opportunities related to AI capacity building in Africa, with speakers addressing engineering development, urban planning, data sovereignty, labour rights, and linguistic inclusion.
Prof. Ke Gong opened by introducing the Engineering Capacity Building for Africa programme, initiated by the World Federation of Engineering Organizations, noting that fewer than 7% of Africa's measurable SDGs are on track . The programme aims to leverage digital technology integrated with engineering expertise, establishing training centres across African countries in partnership with local organisations . Pilot trainings in Kenya, Uganda, and Cape Verde yielded positive feedback, with a key lesson being the necessity of working closely with local partners rather than remotely . Prof. Felix Krawczyk built on this by recommending a focus on urban planning combined with AI and open geospatial data, given Africa's rapid population growth and infrastructure challenges . He also raised the question of data ownership, arguing for a fundamental right to data for communities, templates for data-sharing agreements, and potential taxation on AI outputs to return value to data contributors .
Prof. Christoph Stückelberger broadened the analysis to six interrelated dimensions of AI sovereignty: human values, critical minerals, capital, data sovereignty, green energy, and governance . He highlighted that Africa holds vast cobalt reserves essential to AI hardware, yet faces neocolonial dynamics in how those resources are exploited , and that African capital held in Geneva banks is not being reinvested on the continent . Dr. Xiao Zhang stressed that infrastructure investment and human capacity building must proceed in parallel, and that training programmes should be tailored to local needs rather than designed externally .
Dr. Ricardo Israel Robles Pelayo drew attention to the largely invisible labour of data workers in the Global South, citing documented cases in Kenya where workers reviewing harmful content were paid as little as $2 per hour and dismissed when attempting to unionise , in violation of international labour standards . Mr. Christofer Dutz highlighted the language barrier as a significant obstacle to participation, advocating for AI-powered multilingual tools .
Overall, the discussion underscored that bridging the AI intelligence divide requires not only technical and infrastructural investment, but also equitable governance, cultural sensitivity, labour protections, and inclusive multilingual participation .
Overall Purpose
- The discussion is a multi-speaker panel session focused on addressing the "intelligence divide" between the Global North and Global South, with particular emphasis on Africa. The session brings together academics, policymakers, technologists, and civil society representatives to explore how AI capacity building, digital infrastructure, data sovereignty, labour rights, and ethical governance can be advanced to ensure African nations become equal partners in the global AI ecosystem rather than mere consumers or raw material suppliers.
- --
Major Discussion Points
- Africa's critical gap in sustainable development and the need for engineering capacity building: Africa is severely behind on the global sustainable development agenda, with less than 7% of measurable SDGs on track . The World Federation of Engineering Organizations has launched a programme to leverage digital technology integrated with engineering to address this, including pilot training centres in Kenya, Uganda, and Cape Verde . A key lesson from the pilots was the necessity of working closely with local organisers and responding to real, on-the-ground needs rather than imposing externally designed solutions .
- Urban planning, AI, and open geospatial data as tools for African development: Prof. Krawczyk argued that Africa's rapid urban population growth makes urban engineering and planning a critical priority . He proposed combining open geospatial data platforms with AI technologies to develop urban planning strategies, and suggested building courses that train urban planners in AI capacity and vice versa . He also raised the important question of data ownership, arguing for a fundamental right to data for communities, templates for data-sharing agreements, and potential taxation on AI outputs to return value to the communities whose data underpins AI systems .
- AI sovereignty in Africa requires addressing six interconnected dimensions: Prof. Stückelberger presented a holistic framework arguing that AI capacity in Africa cannot be achieved by focusing solely on technology or finance . He identified six interrelated factors: people and ethical values (overcoming corruption and nepotism) ; control over critical minerals such as cobalt, on which the global AI industry depends ; access to investment capital and the need to shift the narrative from risk to opportunity ; data sovereignty and the urgent need for African data centres ; green energy that balances AI infrastructure needs with people-centred access ; and good governance and politics as the most critical enabling factor .
- Labour rights and the exploitation of Global South data workers: Dr. Robles Pelayo highlighted a largely unnamed divide - between those who design AI and those whose invisible labour makes it possible . He cited documented cases in Kenya where workers reviewing harmful content for AI training were paid as little as $2 per hour, suffered post-traumatic stress disorder and clinical depression, and were dismissed when attempting to unionise . He argued this constitutes a violation of international labour law and called for independent audits of working conditions in data supply chains, inclusion of data workers' unions in AI governance dialogues, and a redefinition of capacity building as genuine co-design rather than cheap labour extraction .
- Practical, layered, and locally adapted capacity building is essential: Dr. Zhang and Mr. Dutz both emphasised that capacity building must be pragmatic and context-sensitive. Dr. Zhang argued that infrastructure investment and human capacity development must proceed in parallel rather than sequentially , that programmes must be designed around local needs rather than pre-packaged solutions from Geneva or Washington , and that training should be built in layers tailored to different audiences - policymakers, network operators, and software developers - rather than using a one-size-fits-all approach . Mr. Dutz added that language barriers represent a significant form of exclusion, and that AI translation tools should be deployed to make open-source knowledge and community participation accessible in languages beyond English .
- --
Overall Tone
- The overall tone of the discussion is earnest, collaborative, and solutions-oriented, with an undercurrent of urgency. Speakers consistently acknowledged the severity of Africa's exclusion from global AI development while remaining constructive and forward-looking. The tone is largely collegial and respectful across speakers, reflecting a shared commitment to equity and inclusion.
- However, the tone shifts noticeably at points. Prof. Stückelberger introduced a more sobering and politically charged register when discussing the exploitation of African mineral resources, the confidential US-DRC deal, and the billions of African capital sitting in Geneva banks rather than being reinvested on the continent . Similarly, Dr. Robles Pelayo brought a distinctly more critical and advocacy-driven tone when exposing the labour exploitation of data workers, framing it explicitly as a violation of international law rather than merely an ethical concern . The audience contributions, particularly from the Kenyan government representative, added a tone of cautious optimism and national pride, asserting Kenya's rule of law and innovation capacity . By the close of the session, the tone returned to collaborative and forward-looking, with calls for continued dialogue and cross-sector partnership.
Expanded Summary: AI Capacity Building in Africa - Engineering, Data Sovereignty, Labour Justice, and Inclusive Participation
#
Overview and Context
This multi-speaker panel session, held as part of an international forum and chaired by Prof. Lina Dai, brought together academics, policymakers, technologists, and civil society representatives to address the "intelligence divide" between the Global North and Global South, with particular emphasis on Africa. The session explored how AI capacity building, digital infrastructure, data sovereignty, labour rights, and ethical governance can be advanced to ensure African nations become equal partners in the global AI ecosystem rather than mere consumers or suppliers of raw materials. The discussion evolved through distinct phases - from technical and infrastructural framing, through structural and political economy critique, to practical recommendations - revealing both broad areas of consensus and significant underlying disagreements about the nature of the problem and the appropriate remedies.
---
#
Engineering Capacity Building for Africa: The Foundational Challenge
Prof. Ke Gong opened the session by introducing the Engineering Capacity Building for Africa programme, an initiative of the World Federation of Engineering Organizations, of which he serves as chair . The programme was established to address Africa's severe lag in the global sustainable development agenda: according to the African Sustainable Development Report, fewer than 7% of Africa's 32 measurable SDGs are on track, and no African sub-region is on course . The programme's approach is to leverage digital technology integrated with engineering expertise, with the aim of establishing training and capacity building centres across African countries in cooperation with local member organisations, and to introduce international expertise and tools to support this effort .
Prof. Gong reported that the programme had already commenced pilot training in Kenya, Uganda, and a third location (referred to in the transcript as "Kapowelda," possibly a transcription of a location made famous by FIFA), generating overwhelmingly positive feedback . A critical lesson from these pilots was that programmes must be grounded in real, on-the-ground needs and must work closely with local engineering organisations rather than operating remotely . This emphasis on local engagement over externally designed solutions became a recurring theme throughout the session, echoed by multiple subsequent speakers.
---
#
Urban Planning, Open Geospatial Data, and AI Integration
Prof. Felix Krawczyk, drawing on his involvement in the IPCC special report on cities - including a chapter with a strong focus on Africa - built on Prof. Gong's foundation by recommending that the capacity building programme be expanded to include urban planning integrated with AI and open geospatial data technologies . He argued that Africa and Asia are experiencing particularly rapid population growth, generating urgent challenges in sanitation, street networks, and public transport - areas where urban planning capacity is often lacking but where foundational decisions made now will shape long-term outcomes .
Prof. Krawczyk proposed that open geospatial data, now widely available at building and street level, could serve as a starting point for AI-assisted urban planning processes . He suggested developing courses that attract urban planners to gain AI skills and AI practitioners to gain urban planning knowledge, creating a dual-track educational model . He also highlighted the existence of open datasets for Africa - including Microsoft's open building dataset - and argued that combining multiple open data sources in open formats would constitute a "tremendous open public good" for African countries, estimated to cost in the low millions of dollars .
---
#
Data Ownership, Community Rights, and the Question of Value Return
Prof. Krawczyk then raised what became one of the session's most consequential themes: the question of who owns data collected from communities by AI providers . Using the example of an AI company collecting community health data to develop solutions - retaining both the data and the resulting intellectual property - he argued that something must flow back to the communities whose data underpins AI systems . He proposed three concrete mechanisms to address this: first, establishing a fundamental right to data for communities, which, while an individual data point has little worth, gives communities a negotiating position ; second, developing templates for communities to share data in exchange for capacity building or financial returns ; and third, introducing a form of taxation on AI outputs - such as tokens - to redistribute value to the people whose data contributed to AI development . He acknowledged that token taxation is not a formally perfect mechanism but noted that tokens can be measured and therefore taxed in some form .
This framing - of data as a collective community asset from which value is currently being extracted without adequate return - directly influenced subsequent speakers and reoriented the discussion from the question of how Africa can access AI to the more fundamental question of on whose terms AI development is occurring.
---
#
Six Interrelated Dimensions of AI Sovereignty in Africa
Prof. Christoph Stückelberger, drawing on nearly five decades of engagement with Africa including recent visits to educational programmes and mining operations, presented the most structurally comprehensive analysis of the session . He argued that AI capacity in Africa cannot be achieved by focusing solely on technical or financial dimensions, and proposed a six-factor framework of interrelated dimensions that must be addressed together .
The first dimension is people and values: Africa has a large, young, and increasingly well-educated population, but the challenge lies in the ethical values needed to overcome corruption, nepotism, and tribalism in service of the common good - obstacles that, while not unique to Africa, are significant barriers to achieving shared goals .
The second dimension is critical minerals: Africa holds vast reserves of cobalt and other minerals essential to AI hardware, with the DRC alone supplying approximately 80% of global cobalt, of which 80% is controlled by Chinese companies . African governments are increasingly asserting their right not merely to export raw materials but to industrialise and process these minerals domestically . This has generated legal and geopolitical tensions, including the DRC's attempts to renegotiate or cancel long-term mining contracts .
The third dimension is capital: Prof. Stückelberger identified investment capital as the most serious bottleneck for AI development in Africa, noting that billions of African private capital held in Geneva and London are not being reinvested on the continent due to risk perception . He argued that a new narrative framing Africa as an opportunity rather than a risk is needed to attract both foreign and domestic capital , while insisting that foreign capital must operate on a win-win basis rather than reproducing neocolonial dominance . He cited the example of a reportedly confidential deal between the United States and the DRC in which land occupied by artisanal miners was allegedly promised to American companies - noting that figures cited in this context were unclear, with references to both 600,000 and 2 million artisanal miners in the DRC - illustrating the conflict potential embedded in foreign investment .
The fourth dimension is data sovereignty: approximately 70% of global data centres are located in the United States, with the remainder concentrated in China and Europe, leaving Africa with very little . Prof. Stückelberger argued that Africa urgently needs its own data centres, noting that dependency on American, Chinese, and European infrastructure is a direct challenge to data sovereignty . He acknowledged that the African Union and member states are working on data sovereignty legislation, but noted that achieving unified African data governance requires overcoming competing national interests .
The fifth dimension is green energy: data centres require enormous amounts of energy, and Prof. Stückelberger warned that if new energy sources are directed primarily to AI infrastructure and industrialisation, communities will be left without power for basic needs such as cooking and school lighting, potentially leading to social unrest . Energy production must therefore be balanced between people-centred access and AI infrastructure needs .
The sixth and most critical dimension is politics and governance: good governance and political will are, in Prof. Stückelberger's assessment, the most critical factors for achieving sovereign AI capacities in Africa . He concluded by urging the audience not to look only at technical or financial factors but to treat all six dimensions as interrelated .
---
#
Labour Justice and the Invisible Workforce of AI
Dr. Ricardo Israel Robles Pelayo introduced what he described as a divide that had not yet been named in the discussion: the divide between those who design AI and those whose invisible labour makes it possible . He argued that AI is not autonomous but depends on millions of workers who label data, classify images, and review extremely harmful content to train AI models . Most of these workers operate under conditions that violate international law .
The most documented case, he noted, is in Kenya, where OpenAI paid a subcontracted company $12.50 per hour, but the Kenyan workers reviewing hours of extreme violence and torture received only $2 per hour . When 184 of these workers attempted to unionise, they were dismissed . Dr. Robles Pelayo highlighted a "curious paradox": these workers are teaching AI to detect the very harmful content that is making them sick, suffering post-traumatic stress disorder, clinical depression, and insomnia, while simultaneously rendering their own jobs redundant .
He argued that this constitutes not merely an ethical problem but a violation of existing international norms, including ILO Conventions 87 and 88 on freedom of association and collective bargaining, the International Covenant on Economic, Social and Cultural Rights, and the United States Guiding Principles on Supply Chains (as stated in the transcript, though this may be a transcription of a reference to the UN Guiding Principles on Business and Human Rights) . He also referenced the most recent international global labour standards specifically for platform workers, which require collective bargaining, the right to explanation of AI-driven decisions, and the right to human review . He noted that the Global South is already responding through legal channels: the Montauk case against Meta reached the Kenyan Court of Appeal, which confirmed jurisdiction, with 184 claims and a demand for billions of dollars - a separate reference to the number 184 from the earlier figure of 184 workers dismissed for attempting to unionise . In 2023, workers formed an Africa-first content moderation workers' union .
Dr. Robles Pelayo concluded with three concrete calls to action: no capacity building project should be funded without an independent audit of working conditions throughout the data supply chain ; data workers and their union representatives must be included in global AI governance dialogues ; and capacity building in the Global South must be fundamentally redefined as co-design of governance and algorithmic auditing, not as a mechanism for producing cheap labour .
---
#
Practical, Layered, and Locally Adapted Capacity Building
Dr. Xiao Zhang, drawing on more than twenty years of work on digital issues, offered a pragmatic and action-oriented perspective . She noted that ministers at the forum had been reluctant to admit that their infrastructure is poor, reflecting the real challenge facing developing countries in the AI age . She made three principal points.
First, Africa should not wait for full infrastructure completion before beginning human capacity building . Basic infrastructure investment must continue , but capacity building can begin immediately using existing mobile devices, cloud-based AI services, open-source models that run on modest hardware, and multilingual interfaces . Data sharing can be built on trust and developed incrementally .
Second, capacity building must be demand-driven and grounded in real local needs rather than pre-packaged solutions designed in Geneva or Washington . Programmes should identify exactly what problems local technicians are facing and what policies local regulators are struggling to write .
Third, training should be built in layers rather than delivered as a single bulk programme . Different audiences have different needs: policymakers need to understand AI governance and risk frameworks but do not need to write software; network operators need practical tools; software developers need technical skills . A one-size-fits-all approach is likely a waste of time .
---
#
Language Barriers and Inclusive Participation in Open-Source Communities
Mr. Christofer Dutz, Director of the Apache Software Foundation, highlighted the language barrier as a significant and often overlooked structural form of exclusion in global technology communities . He noted that the vast majority of knowledge and open-source resources are available only in English, effectively making non-English speakers second-class citizens in these communities . He pointed to the irony that during the very session being discussed, requests for subtitles to be enabled were declined .
Mr. Dutz described the Apache Software Foundation's Travel Assistance Committee, which sponsors participants from Africa and other regions to attend international conferences, but noted that physical travel and visa access remain significant barriers . He also observed that major AI companies are currently making large donations to open-source foundations, and argued that these resources should be directed towards multilingual inclusion . He argued that AI translation tools now make it technically feasible - indeed, a "solved problem" - to make resources available simultaneously in multiple languages, and that the primary constraint is resources rather than technology . He called for open-source foundations to use AI-powered translation to enable participation in any project in any language, framing this as a public good for global inclusion .
In response to an audience question about Chinese-language inclusion, Mr. Dutz noted that Chinese technology communities tend to use platforms such as WeChat that are inaccessible to people outside China , and that he regularly has to remind Apache project communities that discussions and decisions must occur on publicly accessible mailing lists rather than closed platforms . He argued that making Chinese communities more aware of inclusive communication tools is needed not just for Africa but for all participants outside China .
---
#
National Responses: Kenya's AI Strategy and the Rule of Law
A representative of the Kenyan government, speaking as a principal secretary, offered a national perspective on the intelligence divide . She acknowledged that the Global South has increasingly been positioned as a consumer and raw material source rather than an equal partner in AI development , and called for greater international collaboration in AI capacity building, including on infrastructure and data centres .
She described Kenya's national AI strategy 2025-2030, which places capacity building at its core, investing in local talent, digital infrastructure, digital hubs, and public Wi-Fi to enable citizens to participate as equal partners in global AI development . She noted that Kenya's digital economy master plan provides a strategic roadmap for leveraging AI to strengthen the economy, create quality jobs, and position Kenya as a leader in the fourth industrial revolution . She emphasised that national efforts alone are insufficient and that global solidarity is needed .
In direct response to Dr. Robles Pelayo's account of labour exploitation of Kenyan data workers, the representative acknowledged the issue but expressed confidence in Kenya's rule of law, noting that large technology companies have been guided on minimum labour requirements and that "the law will prevail" . She also highlighted Kenya's strengths as an innovation hub, noting that Kenyans are globally recognised for innovation and that the country has a highly educated youth population with connections to leading universities worldwide .
---
#
South African Perspectives: Unionisation, Technology Adoption, and Higher Education
An audience member from South Africa offered a complementary perspective, identifying several structural barriers to AI adoption specific to the South African context . She noted that South Africa's highly unionised labour environment creates specific challenges for technology adoption, as fears of job displacement slow the introduction of automation even at basic levels - for example, self-checkout machines have not been widely introduced due to union concerns . She also highlighted financial constraints, evolving technology challenges, and the importance of political will in enabling AI adoption .
On the positive side, she identified significant potential benefits of AI for higher education and research, including enhanced accessibility, predictive analytics, improved data management, 24/7 AI-powered support, and personalised learning . She noted that AI policies in higher education institutions are increasingly being driven by researchers concerned about the ethical use of AI tools, particularly to prevent students from being unfairly penalised for using AI in ways that might be misconstrued as plagiarism . She also raised the question of AI translation for Chinese-language content, prompting the exchange with Mr. Dutz about multilingual inclusion .
---
#
China-Africa Cooperation: Intercultural Sensitivity and Mutual Learning
In response to an audience question about how China can better support developing countries , Prof. Stückelberger offered a nuanced assessment. He acknowledged that China is very present in Africa and has significant positive impacts across sectors including technology, mining, hospitality, hospitals, and education . However, he argued that growing tensions between Chinese and African communities require careful intercultural listening and understanding .
He recommended that every China-Africa investment programme include a cultural mutual learning component, noting that this requires dedicated resources . He cited the example of Chinese mining companies in the cobalt-mining region of south-east DRC (referred to in the transcript variously as "Colbasi" and "Colvesi") whose workers live in isolated camps, automatically generating conflict , contrasted with another Chinese mining company that invested millions in a vocational training centre, successfully building community trust . He also highlighted specific cultural differences - such as different approaches to time management and the importance of funeral attendance in African communities - as areas requiring genuine mutual understanding rather than imposition of one culture's norms .
Mr. Dutz added that the technical dimension of Chinese community participation in global open-source projects is also a barrier, as Chinese communities use WeChat and other tools inaccessible to international participants , and called for greater awareness of inclusive communication forms .
---
#
Areas of Consensus and Underlying Disagreements
Beneath the surface-level consensus, the discussion revealed significant disagreements that merit attention. The most fundamental was between the predominantly technocratic and optimistic framing of most speakers and Dr. Robles Pelayo's structural critique that AI development is already built on exploited African labour , challenging the premise that AI represents an unambiguous opportunity for Africa. There was also a meaningful disagreement about whether capacity building should be framed as technical skills training or as genuine co-design of governance and labour justice . Prof. Krawczyk and Prof. Stückelberger, while sharing a concern about data extraction, proposed different remedies: Krawczyk focused on community-level data rights and token taxation , while Stückelberger emphasised national data centre infrastructure and AU legislation . The Kenyan government representative's confidence in domestic rule of law as sufficient to address AI labour exploitation stood in notable tension with Dr. Robles Pelayo's documented evidence of systematic violations in Kenya specifically .
Despite these disagreements, all speakers agreed that Africa faces a severe and multidimensional AI and digital capacity gap requiring urgent, targeted, and locally grounded responses . There was broad agreement that capacity building must be demand-driven and co-designed with local stakeholders rather than imposed from outside . Data sovereignty emerged as a shared concern, with multiple speakers arguing that Africa needs its own data infrastructure, legal frameworks, and community rights over data . Speakers also converged on the view that technical solutions alone are insufficient and that governance, ethics, labour rights, cultural sensitivity, and political will are equally critical .
---
#
Conclusions and Forward-Looking Recommendations
The session concluded with a broad set of recommendations and action items. Prof. Gong's Engineering Capacity Building for Africa programme is already operational, with plans to scale training centres across African countries in cooperation with local member organisations . Prof. Krawczyk proposed expanding this to include urban planning integrated with AI and open geospatial data, and called for the creation of an open geospatial data infrastructure for Africa as a public good . He also proposed three concrete data governance mechanisms: community data rights, data-sharing templates, and token taxation .
Dr. Robles Pelayo called for independent audits of working conditions in data supply chains as a prerequisite for funding any capacity building project, inclusion of data workers' unions in AI governance dialogues, and a fundamental redefinition of capacity building as co-design . Dr. Zhang advocated for parallel infrastructure and human capacity investment, demand-driven programme design, and layered training tailored to different audiences . Mr. Dutz committed to continuing efforts within the Apache Software Foundation to require publicly accessible communication channels and to deploy AI translation tools for multilingual participation, and called for the significant donations currently being made by major AI companies to open-source foundations to be directed towards these inclusion goals . Prof. Stückelberger recommended embedding cultural mutual learning components in every China-Africa investment programme .
Several critical issues remained unresolved, including how African nations can achieve sufficient unity within the African Union to establish shared data centres ; how to design enforceable mechanisms for returning value from AI companies to communities ; how to attract African private capital back to the continent ; how to balance energy demands between AI infrastructure and household needs ; and how to ensure that large technology companies operating in Africa comply with minimum labour standards beyond the enforcement capacity of individual national courts . The overall implication of the discussion is that any effective AI capacity building initiative for Africa must simultaneously address technical skills, governance co-design, data sovereignty, labour standards, energy trade-offs, and capital investment barriers - a substantially more complex agenda than any single speaker's proposal encompasses, and one that requires genuine multi-stakeholder collaboration grounded in African agency and ownership .
Africa is severely behind on SDGs, with less than 7% of measurable goals on track, necessitating a targeted engineering capacity building programme leveraging digital technology - Engineering capacity gap in Africa
Arg. 1Prof. Ke Gong introduces the Engineering Capacity Building for Africa programme, initiated by the World Federation of Engineering Organizations, to address Africa's severe lag in sustainable development. The programme aims to leverage digital technology and integrate it with engineering professions to build training and capacity building centres across African countries. The goal is to introduce international expertise and tools to support progress on the SDGs.
According to the African Sustainable Development Report, less than 7% of the 32 measurable SDGs are on track for Africa , and no single one of Africa's five sub-regions is on track . The programme plans to build training centres in different parts of African countries in cooperation with member organisations of the Federation .
on: International collaboration and multi-stakeholder partnerships are essential to address Africa's AI capacity challenges
on: Whether capacity building should be framed as technical skills training or as co-design of governance and labour justice
Capacity building must respond to real local needs and work closely with local organisers rather than being designed remotely - Local engagement is essential
Arg. 2Prof. Ke Gong emphasises that effective capacity building requires direct engagement with local communities and organisations rather than remote programme design. Pilot trainings conducted in Kenya, Uganda, and Cape Verde generated very positive feedback, reinforcing the importance of on-the-ground collaboration. The key lesson learned was the necessity of facing real needs and working closely with local engineering organisations.
Pilot training programmes were conducted in Kenya, Uganda, and Cape Verde, and the feedback was described as overwhelmingly positive . The critical lesson drawn from these pilots was that programmes must face real needs and work closely with local organisers rather than operating remotely .
on: Capacity building programmes must be grounded in real local needs and designed with local partners, not imposed from outside
on: The primary barrier to global technology participation for African contributors - language exclusion versus physical and visa access barriers
Capacity building should be demand-driven, addressing real challenges faced by local technicians and policymakers rather than offering pre-packaged solutions designed in Geneva or Washington - Start from real local needs
Arg. 1Dr. Xiao Zhang argues that many capacity building programmes are designed in distant locations such as Geneva or Washington and may not be well adapted to local demands. Instead, programmes should identify the exact problems that local technicians and regulators are actually facing. This targeted approach ensures that training is relevant and effective.
Dr. Zhang noted that some capacity building programmes are designed in Geneva or Washington but used in Africa, and may not be well adapted to local demand . She called for identifying what problems local technicians are really facing and what policies African regulators are struggling to write .
on: Capacity building programmes must be grounded in real local needs and designed with local partners, not imposed from outside
Training should be built in layers tailored to different audiences, such as policymakers needing AI governance understanding and engineers needing practical tools, rather than a one-size-fits-all approach - Layered, targeted training
Arg. 2Dr. Xiao Zhang contends that a single training package cannot serve all audiences effectively, as different stakeholders have fundamentally different needs. Policymakers need to understand AI governance and risk frameworks, while network operators and engineers need practical tools. Building training in layers, with hundreds of complementary programmes, is more effective than a single two-week course.
Dr. Zhang gave the example that policymakers need to understand AI governance and risk frameworks but do not need to write software, while network operators need practical tools . She concluded that a one-size-fits-all training approach is likely a waste of time and advocated for building in layers .
on: Whether capacity building should be framed as technical skills training or as co-design of governance and labour justice
Infrastructure investment must proceed in parallel with human capacity building, using existing mobile technology and cloud-based AI services rather than waiting for full infrastructure completion - Parallel infrastructure and human capacity development
Arg. 3Dr. Xiao Zhang argues that Africa should not wait for full infrastructure completion before beginning human capacity building, as the two must proceed simultaneously. Existing mobile technology, cloud-based AI services, open-source models, and multilingual interfaces can be leveraged immediately. Continued investment in basic infrastructure remains essential, but should not be a prerequisite for starting capacity development.
Dr. Zhang stated that basic infrastructure investment must continue because without it no digital programme can proceed , but simultaneously argued that human capacity building can begin using existing tools such as mobile devices, cloud-based AI services, open-source models on modest hardware, and multilingual interfaces already available in Africa .
on: International collaboration and multi-stakeholder partnerships are essential to address Africa's AI capacity challenges
on: Whether infrastructure completion should precede or run in parallel with human capacity building
AI capacity building should be expanded to include urban planning, integrating AI and geospatial data technologies to address Africa's rapid urbanisation challenges - Urban planning and AI integration
Arg. 1Prof. Krawczyk suggests that the engineering capacity building programme should be expanded to encompass urban planning, given Africa's rapid population growth and associated challenges such as sanitation and public transport. Open geospatial data, now widely available at building and street level, provides a strong foundation for AI-assisted urban planning. Developing courses that combine urban planning expertise with AI and big data skills would address a critical gap.
Prof. Krawczyk noted that Africa and Asia face particularly fast population growth and challenges including sanitation and street network and public transport deficits . He highlighted that open geospatial data is now available at building and street level and can serve as a starting point for urban planning processes , and that Microsoft has built an open building dataset for Africa, though combining multiple open data sources would be more valuable .
on: The intelligence divide between the Global North and Global South in AI is a serious, multidimensional problem requiring urgent collective action
Communities and individuals should have a fundamental right to their data, giving them a negotiation position when AI companies collect and exploit community data - Fundamental right to data
Arg. 2Prof. Krawczyk argues that while an individual data point has little value, community-level data is highly valuable, and communities should be recognised as having a fundamental right to that data. This right would provide communities with a negotiating position when AI companies seek to collect and use their data. Without such a right, communities have no leverage to demand anything in return.
Prof. Krawczyk used the example of an AI provider collecting local health data, finding solutions for the community, but retaining all the health data to develop new products . He argued that a fundamental right to data for people and communities, even if an individual data point has no worth on its own, gives communities a negotiation position .
on: Data sovereignty is a critical and urgent challenge for Africa, requiring African-owned data infrastructure and governance frameworks
on: How data sovereignty should be achieved - through community rights and taxation mechanisms versus national infrastructure and legislation
Templates should be developed allowing communities to share data in exchange for capacity, financial returns, or other benefits - Data sharing templates for communities
Arg. 3Prof. Krawczyk proposes the development of standardised templates that define how communities can share their data with AI companies while receiving something of value in return. These returns could take the form of capacity building, financial streams, or other benefits that help communities finance their future needs. Such templates would operationalise the fundamental right to data by providing practical mechanisms for negotiation.
Prof. Krawczyk proposed developing templates for how communities can share their data and receive something back in return, such as capacity or a money stream to finance their future needs .
AI companies accumulate enormous financial value from knowledge distilled from community data, and mechanisms such as token taxation or data levies should return value to the people whose data was used - Taxation on AI outputs to return value
Arg. 4Prof. Krawczyk observes that AI companies derive tremendous financial value from the accumulated knowledge distilled from community data, yet the communities that generated that data receive nothing in return. He proposes that value could be returned through taxation on data or on AI output tokens, since tokens can be measured and therefore taxed. This approach would apply particularly to development contexts.
Prof. Krawczyk noted that AI companies have tremendous financial value largely because of their knowledge distillation from accumulated data , and proposed that taxation on data or on tokens - which can be measured - could be used to return value to the people whose data was used .
on: How data sovereignty should be achieved - through community rights and taxation mechanisms versus national infrastructure and legislation
Africa urgently needs its own data centres to achieve data sovereignty, as current dependence on US, Chinese, and European data centres undermines African control over its own data - Need for African data centres
Arg. 1Prof. Stückelberger argues that data sovereignty requires physical infrastructure, specifically data centres located on African soil. Currently, approximately 70% of data centres are in the US, with most of the remainder in China and Europe, leaving Africa severely underrepresented. Without African data centres, African data sovereignty remains aspirational rather than real.
Prof. Stückelberger cited figures suggesting approximately 70% of data centres are placed in the US, then in China, then some in Europe, with much less in Africa . He mentioned a billionaire from Zimbabwe seeking to establish data centres in Africa as an important development , and noted that the dependency on American, Chinese, and European data centres is a challenge for data sovereignty .
on: Data sovereignty is a critical and urgent challenge for Africa, requiring African-owned data infrastructure and governance frameworks
on: How data sovereignty should be achieved - through community rights and taxation mechanisms versus national infrastructure and legislation
The African Union and member states are working on data sovereignty legislation, but achieving unified African data governance requires overcoming national partial interests - African data sovereignty legislation
Arg. 2Prof. Stückelberger acknowledges that the African Union is making important efforts to develop legislation for increased data sovereignty. However, he cautions that individual African nations have their own partial interests that can undermine collective action. Finding common ground among African nations to establish shared data infrastructure and governance frameworks remains a significant challenge.
Prof. Stückelberger noted that the AU and member countries are working on legislation for increased data sovereignty , but observed that while the African Union is a great institution, nations have their own partial interests, making it difficult to find common strength for African data centres .
on: How data sovereignty should be achieved - through community rights and taxation mechanisms versus national infrastructure and legislation
Human values, including overcoming corruption, nepotism, and tribalism, are foundational to achieving AI capacity and sovereignty in Africa - Human values as a key factor
Arg. 3Prof. Stückelberger identifies human values as the first and foundational dimension of AI sovereignty, arguing that Africa has sufficient people and a large young educated population. The challenge lies in cultivating ethical values that overcome corruption, nepotism, and tribalism in favour of the common good. He notes these are not uniquely African problems but are nonetheless significant obstacles to the goals being discussed.
Prof. Stückelberger stated that Africa has enough people and millions of well-educated young people, but that the challenge is the ethical values needed to overcome corruption, nepotism, tribalism, and to work for the common good . He acknowledged these problems are not specific to Africa but are present there as well .
Africa holds vast critical mineral resources essential to AI hardware, and African governments are increasingly asserting their right to industrialise and process these minerals domestically rather than merely exporting raw materials - Critical minerals and industrialisation
Arg. 4Prof. Stückelberger highlights that Africa, particularly the DRC, holds the majority of the world's cobalt, which is essential for AI hardware and mobile devices. African governments are now pushing for industrialisation and domestic processing of these minerals rather than simply exporting raw materials. This shift is creating legal and geopolitical tensions with existing long-term mining contracts held by foreign companies.
Prof. Stückelberger noted that approximately 80% of the world's cobalt comes from the DRC, and 80% of that is in the hands of Chinese companies . He described seeing kilometres of trucks exporting raw cobalt , and noted that the DRC is now cancelling some contracts with Chinese companies, leading to legal disputes at the ICC in The Hague .
Investment capital is the most serious bottleneck for AI development in Africa, with billions of African private capital sitting in Geneva and London rather than being reinvested in Africa due to risk perception - Capital investment bottleneck
Arg. 5Prof. Stückelberger identifies capital as the most critical bottleneck for AI sovereignty in Africa, arguing that AI development requires capital sovereignty. Paradoxically, billions of dollars of African private capital are held in banks in Geneva and London but are not reinvested in Africa because of risk perception. Changing this narrative from risk to opportunity is essential to unlock investment.
Prof. Stückelberger stated that the most serious bottleneck is investment capital in Africa, and that AI sovereignty needs capital sovereignty . He noted that private banks in Geneva hold not millions but billions of African capital , which is invested in New York and London rather than Africa due to risk perceptions .
on: Whether foreign investment in Africa is primarily an opportunity to be encouraged or a neocolonial risk to be managed
Green energy production must be balanced between people-centred needs and the energy demands of data centres and industrialisation, to avoid social unrest - Energy balance for AI and communities
Arg. 6Prof. Stückelberger warns that data centres and AI infrastructure require enormous amounts of energy, which risks diverting power away from households, schools, and basic services. If new energy sources are directed primarily to data centres controlled by foreign companies, communities will be left in the dark and social unrest will follow. Energy production must therefore be expanded and prioritised in a people-centred way.
Prof. Stückelberger noted that data centres need a huge amount of energy and raised the dilemma of whether to use energy for big data while it is lacking for households, cooking, and school lighting . He warned that if everything is pushed to data centres and industrialisation, people will take to the streets, citing the scenario of communities being left in the dark while new power stations feed data centres controlled abroad .
Good governance and political will are the most critical factors for achieving sovereign AI capacities in Africa, and all six dimensions — people, minerals, capital, data, energy, and politics — must be addressed as interrelated factors - Six interrelated factors for AI sovereignty
Arg. 7Prof. Stückelberger concludes his analysis by identifying politics and good governance as the most critical factor for AI sovereignty in Africa, alongside the other five dimensions he has outlined. He argues that these six factors — people with values, critical minerals, capital, data sovereignty, green energy, and politics — are deeply interrelated and must be addressed together. Focusing only on technical or financial aspects while ignoring the others will be insufficient.
Prof. Stückelberger stated that politics with good governance is the most critical factor for increased and sovereign AI capacities in Africa , and that multiplied and diversified cooperation with multi-sectoral partners increases leverage, know-how, and stability . He urged the audience not to look only at technical or financial factors but to treat all six as interrelated .
on: International collaboration and multi-stakeholder partnerships are essential to address Africa's AI capacity challenges
China's engagement in Africa has significant positive impacts across sectors, but requires greater intercultural sensitivity, mutual learning, and investment in vocational training to build genuine trust and avoid social conflict - Intercultural sensitivity in China-Africa cooperation
Arg. 8Prof. Stückelberger acknowledges that China is very present in Africa and has many positive impacts across sectors including mining, hospitality, hospitals, and education. However, he argues that growing tensions require careful intercultural listening and mutual learning as a component of every investment programme. He cites the example of Chinese workers living in isolated camps as a source of automatic conflict, contrasted with a Chinese mining company that invested in a vocational training centre and thereby built genuine trust.
Prof. Stückelberger observed that Chinese companies in Colvesia live in camps, which creates conflicts automatically , but contrasted this with a Chinese mining company that invested millions in a vocational training centre, which created trust . He recommended that every China-Africa programme include a cultural mutual learning component, including attention to language and different approaches to time management .
AI is not autonomous but depends on millions of invisible data workers in the Global South who label data and review harmful content under conditions that violate international labour law - Invisible labour underpinning AI
Arg. 1Dr. Robles Pelayo challenges the notion of AI as autonomous, arguing that it depends on millions of workers who perform essential but invisible tasks such as labelling data, classifying images, and reviewing harmful content. Most of these workers operate under conditions that violate international labour law. This labour force is concentrated in the Global South and is systematically exploited.
Dr. Robles Pelayo stated that AI depends on millions of workers who label data, classify images, and review extremely harmful content to teach models, and that most of them work under conditions that violate international law . He described this as invisible human labour .
on: Whether AI labour exploitation in the Global South is primarily a legal violation requiring enforcement or an ethical problem requiring governance reform
Documented cases, such as Kenyan workers paid $2 per hour to review extreme violence for OpenAI while suffering PTSD and clinical depression, illustrate severe exploitation in AI data supply chains - Exploitation of data workers in Kenya
Arg. 2Dr. Robles Pelayo presents Kenya as the most documented case of AI data worker exploitation in the world. Workers were paid only $2 per hour to review hours of extreme violence and torture, while the subcontracted company received $12.50 per hour from OpenAI. These workers suffered serious psychological harm including PTSD and clinical depression, and 184 were dismissed when they attempted to unionise.
Dr. Robles Pelayo cited the case of OpenAI paying a subcontracted company $12.50 per hour, while Kenyan workers reviewing extreme violence received only $2 per hour . He noted that 184 of those workers were dismissed when they attempted to unionise , and that workers suffered post-traumatic stress disorder, clinical depression, and insomnia .
on: Whether AI labour exploitation in the Global South is primarily a legal violation requiring enforcement or an ethical problem requiring governance reform
No capacity building project should be funded without independent audits of working conditions throughout the data supply chain - Independent audits of labour conditions
Arg. 3Dr. Robles Pelayo argues that the exploitation of data workers is a structural problem embedded in AI development supply chains, and that funding for capacity building projects must be conditional on verified compliance with labour standards. Independent audits of working conditions throughout the data supply chain are necessary to ensure that capacity building does not inadvertently subsidise exploitation. This is framed as a matter of existing international legal obligations, not merely ethics.
Dr. Robles Pelayo stated that the condition of labour standards throughout the data supply chain must be addressed, and that no capacity building project should be funded without an independent audit of working conditions .
Data workers and their unions must be included in global AI governance dialogues, as those who build AI systems must participate in the rooms where governance decisions are made - Workers' representation in AI governance
Arg. 4Dr. Robles Pelayo calls for the inclusion of data workers and their union representatives in global AI governance processes. He argues that those who actually build AI systems through their labour have a legitimate stake in how those systems are governed. The formation of the Africa First Content Moderation Worker Union in 2023 is cited as evidence that workers are already organising to assert this right.
Dr. Robles Pelayo noted that in 2023, content moderation workers formed the Africa First Content Moderation Worker Union , and called for data workers' union representation to be included in the global dialogue on AI governance, stating that those who build the systems must be in the room where governance decisions are made .
Capacity building in the Global South must be redefined as co-design of governance and algorithmic auditing, not as a source of cheap labour - Redefining capacity building as co-design
Arg. 5Dr. Robles Pelayo argues that the current model of capacity building in the Global South effectively reduces it to a source of cheap labour for AI companies. He calls for a fundamental redefinition of capacity building as genuine co-design of governance frameworks, algorithmic auditing, and participation in global rule-making. This reframing would position the Global South as an equal partner rather than a subordinate supplier.
Dr. Robles Pelayo called for redefining capacity building not as cheap labour for the Global South but as co-design of governance, algorithmic auditing, and genuine participation in global rule-making .
on: The intelligence divide between the Global North and Global South in AI is a serious, multidimensional problem requiring urgent collective action
on: Whether capacity building should be framed as technical skills training or as co-design of governance and labour justice
Non-English speakers are effectively treated as second-class citizens in global technology communities, as the vast majority of knowledge and open-source resources are available only in English - Language exclusion in technology
Arg. 1Mr. Dutz argues that the dominance of English in global technology communities creates a structural barrier that excludes non-English speakers from full participation. People who are not fluent English speakers find it almost impossible to access the vast majority of available knowledge and resources. This linguistic exclusion compounds other barriers such as visa difficulties and travel costs that prevent African contributors from participating in international technology communities.
Mr. Dutz stated that if you are not a fluent English speaker, you become a sort of second-class citizen, as a lot of information is simply impossible to access . He also noted that during the session itself, several people asked for subtitles to be enabled and were declined each time, illustrating the immediate practical impact of language exclusion .
on: The primary barrier to global technology participation for African contributors - language exclusion versus physical and visa access barriers
AI translation tools now make it technically feasible to make resources available simultaneously in multiple languages, removing language as a barrier to participation in global open-source and technology projects - AI as a tool for language inclusion
Arg. 2Mr. Dutz argues that AI-powered translation has reached a level of maturity where it can effectively bridge language barriers in technology communities. He draws on his personal experience working for a Chinese company where AI translation tools enabled seamless collaboration between English and Chinese speakers, including co-authoring articles. This demonstrates that the technical solution exists, even if resources for implementation remain a challenge.
Mr. Dutz described his experience working for a Chinese company using a chat tool called Feishu, where he and Chinese colleagues could communicate seamlessly and even co-author articles, with him writing in English and colleagues writing in Chinese . He noted that simultaneously translating resources from English to French is now a solved problem technically .
Open-source foundations should use AI-powered translation to enable participation in any language, making global technology communities genuinely inclusive for African and other non-English-speaking contributors - Multilingual open-source participation
Arg. 3Mr. Dutz advocates for open-source foundations, including the Apache Software Foundation, to use modern AI translation technology to make all project resources, email lists, and communications available in multiple languages. Currently, everything in open-source communities revolves around English, which excludes large portions of the global population. He is actively working to change this within Apache, particularly in the context of large AI company donations to open-source foundations.
Mr. Dutz noted that at Apache, everything around email lists and project communications is in English, but he is fighting to use modern technology to make it possible to participate in any project in any language . He mentioned that big AI companies are currently making large donations to open-source foundations, and he is trying to make language inclusion part of the conditions for such donations .
Chinese technology communities tend to use platforms such as WeChat that are inaccessible to people outside China, and greater awareness of inclusive communication tools is needed to enable broader global collaboration - Inclusive communication platforms for global collaboration
Arg. 4Mr. Dutz observes that Chinese technology communities, including those contributing to Apache projects, tend to conduct discussions and decisions on platforms such as WeChat that are not accessible to people outside China. This effectively excludes non-Chinese contributors from key conversations and decision-making processes. He argues that making Chinese communities more aware of the need for inclusive communication tools is essential for genuine global collaboration.
Mr. Dutz noted that at Apache, the rule is that discussions and decisions must happen on public mailing lists, but Asian communities, especially Chinese ones, use WeChat for these discussions . He stated that WeChat and similar tools are not accessible for people outside of China, and called for making Chinese communities more aware of choosing communication forms that are more inclusive .
Kenya's national AI strategy 2025–2030 places capacity building at its core, investing in local talent, digital infrastructure, digital hubs, and public Wi-Fi to enable citizens to participate as equal partners in global AI development - Kenya's national AI strategy
Arg. 1The Kenyan government representative outlines Kenya's comprehensive national AI strategy, which centres on building the capacity of Kenyan citizens to participate in global AI development as equal partners. The strategy includes investment in local talent, extending digital infrastructure across the country, building digital hubs, and providing public Wi-Fi. Kenya positions itself as a leading technology and innovation hub globally and sees AI as an opportunity to assert its capabilities.
The Kenyan principal secretary stated that Kenya's national AI strategy 2025-2030 places capacity building at its core , and described investments in local talent, digital infrastructure including extending digital care across the country, building digital hubs for youth digital services, and providing public Wi-Fi . Kenya's digital economy master plan was cited as a strategic roadmap for leveraging AI to strengthen the economy and create quality jobs .
on: Infrastructure investment and human capacity building must proceed simultaneously rather than sequentially
on: Whether foreign investment in Africa is primarily an opportunity to be encouraged or a neocolonial risk to be managed
National efforts alone are insufficient and must be complemented by global solidarity and international collaboration in AI capacity building and infrastructure - Need for global solidarity alongside national efforts
Arg. 2The Kenyan government representative acknowledges that while Kenya is making significant national investments in AI capacity building, these efforts are insufficient on their own. Global solidarity and international collaboration are essential to address the intelligence divide between the Global North and South. Kenya calls for greater international cooperation encompassing not just human capacity but also infrastructure and data centres.
The Kenyan principal secretary stated that national efforts alone are not sufficient and that global solidarity is needed . Kenya called for greater international collaboration in AI capacity building, including human capacity, infrastructure, and data centres .
on: International collaboration and multi-stakeholder partnerships are essential to address Africa's AI capacity challenges
on: Whether infrastructure completion should precede or run in parallel with human capacity building
South Africa's highly unionised labour environment creates specific challenges for technology adoption, as fears of job displacement slow the introduction of automation even at basic levels such as self-checkout machines - Unionisation and technology adoption challenges in South Africa
Arg. 3An audience member from South Africa highlights that the country's strong union culture creates particular challenges for the adoption of new technologies, including AI. Fears of job displacement have prevented even basic automation such as self-checkout machines from being introduced. This dynamic is presented as a key challenge for the African context when considering AI adoption and capacity building.
The South African audience member noted that South Africa is very unionised, and that even self-checkout machines have not been introduced because of concerns about job displacement . This was presented as a key challenge when looking at the African situation more broadly .
AI offers significant benefits for higher education and research, including enhanced accessibility, predictive analytics, improved data management, and personalised learning, but requires institutional AI policies to govern ethical use - AI benefits and governance in higher education
Arg. 4An audience member from the higher education sector outlines the multiple benefits that AI can bring to research and academic institutions, including enhanced accessibility, predictive analytics for collection development, improved data management, and personalised learning recommendations. However, she notes that institutions are now pushing for AI policies to govern ethical use, particularly to avoid penalising students for using AI tools without understanding the ethical boundaries. The automation of repetitive tasks such as cataloguing and metadata indexing is also highlighted as a significant benefit.
The audience member described benefits including enhanced accessibility, predictive analytics for collection development, improved search and discoverability, better data management and analysis, 24/7 AI-powered chatbot support, and personalised learning recommendations . She noted that higher education institutions are pushing for AI policies to avoid students being penalised for using AI tools without understanding ethical use .
The session should facilitate open dialogue by inviting audience members to raise questions and make contributions after the panel presentations - Facilitating open discussion
Arg. 1Prof. Lina Dai, acting as session moderator, opens the floor to audience participation after the panel presentations, encouraging anyone who wishes to speak or ask questions to raise their hand. This reflects her role in ensuring the session is interactive and inclusive of diverse voices beyond the invited panellists.
Prof. Lina Dai explicitly invited audience members to speak or ask questions after the panel presentations, stating that anyone who wants to speak or make questions can raise their hand , and asked participants to briefly introduce themselves before speaking .
The session covers a broad range of critical perspectives on AI and development, including engineering capacity building, urban planning, ethical dimensions, labour justice, and language inclusion - Breadth of the AI capacity building agenda
Arg. 2Through her introductions of each speaker, Prof. Lina Dai frames the session as encompassing a wide range of interconnected issues relevant to AI capacity building in Africa and the Global South. Her introductions signal that the discussion spans technical, ethical, governance, and social dimensions of the intelligence divide.
Prof. Lina Dai introduced speakers covering engineering capacity building in Africa , AI governance ethics and ethical chances and challenges in Africa , digital issues in developing countries in the era of AI , AI labour justice , and open-source software participation , demonstrating the breadth of topics covered in the session.
Session Knowledge Graph
Speakers · Topics · Arguments · Relationships
Prof. Ke Gong stressed that pilot trainings taught the lesson that programmes must face real needs and work closely with local engineering organisations rather than operating remotely . Dr. Xiao Zhang reinforced this, noting that capacity building programmes designed in Geneva or Washington may not be well adapted to local demand, and called for identifying exactly what problems local technicians and regulators are actually facing . The Kenyan government representative confirmed this from a national perspective, describing Kenya's own strategy as centred on building local talent and enabling citizens to participate as equal partners .
Capacity building must respond to real local needs and work closely with local organisers rather than being designed remotely - Local engagement is essential
Capacity building should be demand-driven, addressing real challenges faced by local technicians and policymakers rather than offering pre-packaged solutions designed in Geneva or Washington - Start from real local needs
Kenya's national AI strategy 2025–2030 places capacity building at its core, investing in local talent, digital infrastructure, digital hubs, and public Wi-Fi to enable citizens to participate as equal partners in global AI development - Kenya's national AI strategy
Prof. Krawczyk raised the question of who owns data, using the example of an AI provider collecting community health data and retaining it for product development , and argued for a fundamental right to data for communities . Prof. Stückelberger elaborated that approximately 70% of data centres are in the US, with very little in Africa , and that dependency on American, Chinese, and European data centres is a direct challenge to data sovereignty . The Kenyan representative called for greater international collaboration including on data centres , confirming that national efforts alone are insufficient .
Communities and individuals should have a fundamental right to their data, giving them a negotiation position when AI companies collect and exploit community data - Fundamental right to data
Africa urgently needs its own data centres to achieve data sovereignty, as current dependence on US, Chinese, and European data centres undermines African control over its own data - Need for African data centres
National efforts alone are insufficient and must be complemented by global solidarity and international collaboration in AI capacity building and infrastructure - Need for global solidarity alongside national efforts
Dr. Xiao Zhang explicitly argued that Africa should not wait for full infrastructure completion before beginning human capacity building, stating that basic infrastructure investment must continue but that human capacity building can begin immediately using existing mobile devices, cloud-based AI services, and open-source models . The Kenyan government representative confirmed this dual approach in practice, describing simultaneous investment in digital infrastructure including extending digital coverage across the country and building digital hubs, alongside investment in local talent and research institutions .
Infrastructure investment must proceed in parallel with human capacity building, using existing mobile technology and cloud-based AI services rather than waiting for full infrastructure completion - Parallel infrastructure and human capacity development
Kenya's national AI strategy 2025–2030 places capacity building at its core, investing in local talent, digital infrastructure, digital hubs, and public Wi-Fi to enable citizens to participate as equal partners in global AI development - Kenya's national AI strategy
Prof. Ke Gong established the scale of the problem, noting that less than 7% of Africa's measurable SDGs are on track and no African sub-region is on track . Prof. Krawczyk highlighted Africa's rapid urbanisation challenges as a specific dimension of the divide . Prof. Stückelberger framed the divide across six interrelated dimensions including people, minerals, capital, data, energy, and politics . Dr. Zhang noted that ministers at the forum were reluctant to admit their infrastructure is poor . Dr. Robles Pelayo argued the divide extends to invisible labour exploitation . The Kenyan representative called for greater international collaboration to build a more equitable AI ecosystem .
Africa is severely behind on SDGs, with less than 7% of measurable goals on track, necessitating a targeted engineering capacity building programme leveraging digital technology - Engineering capacity gap in Africa
AI capacity building should be expanded to include urban planning, integrating AI and geospatial data technologies to address Africa's rapid urbanisation challenges - Urban planning and AI integration
Good governance and political will are the most critical factors for achieving sovereign AI capacities in Africa, and all six dimensions — people, minerals, capital, data, energy, and politics — must be addressed as interrelated factors - Six interrelated factors for AI sovereignty
Capacity building in the Global South must be redefined as co-design of governance and algorithmic auditing, not as a source of cheap labour - Redefining capacity building as co-design
National efforts alone are insufficient and must be complemented by global solidarity and international collaboration in AI capacity building and infrastructure - Need for global solidarity alongside national efforts
Prof. Ke Gong described the Engineering Capacity Building for Africa programme as involving multiple stakeholders including scientific, technical engineering organisations and industrial partners , and building training centres in cooperation with member organisations across African countries . Prof. Stückelberger argued that multiplied and diversified cooperation with multi-sectoral partners increases leverage, know-how, and stability of AI capacities in Africa . Dr. Zhang emphasised that data sharing and cooperation can be built on trust . The Kenyan representative explicitly called for greater international collaboration in AI capacity building and stated that national efforts alone are not sufficient .
Africa is severely behind on SDGs, with less than 7% of measurable goals on track, necessitating a targeted engineering capacity building programme leveraging digital technology - Engineering capacity gap in Africa
Good governance and political will are the most critical factors for achieving sovereign AI capacities in Africa, and all six dimensions — people, minerals, capital, data, energy, and politics — must be addressed as interrelated factors - Six interrelated factors for AI sovereignty
Infrastructure investment must proceed in parallel with human capacity building, using existing mobile technology and cloud-based AI services rather than waiting for full infrastructure completion - Parallel infrastructure and human capacity development
National efforts alone are insufficient and must be complemented by global solidarity and international collaboration in AI capacity building and infrastructure - Need for global solidarity alongside national efforts
Both Prof. Krawczyk and Prof. Stückelberger shared a strong concern about data ownership and the need for mechanisms to return value to communities whose data is exploited. Prof. Krawczyk argued that communities should have a fundamental right to their data as a negotiation position , and proposed taxation on AI tokens or data to return value to people . Prof. Stückelberger similarly emphasised that Africa needs data sovereignty and its own data centres , and that the AU is working on data sovereignty legislation . Both framed data sovereignty as a structural issue requiring both legal rights and physical infrastructure. Both Dr. Robles Pelayo and Prof. Krawczyk raised the issue of AI companies extracting value from communities in the Global South without adequate return. Prof. Krawczyk used the example of an AI provider collecting community health data and retaining it for product development , arguing that communities should receive something back . Dr. Robles Pelayo documented the same dynamic in the labour context, showing that Kenyan workers received only $2 per hour while the subcontracted company received $12.50 per hour from OpenAI . Both called for structural mechanisms to ensure value flows back to communities and workers. Both Mr. Dutz and Dr. Zhang highlighted the importance of making AI capacity building genuinely accessible and inclusive. Mr. Dutz focused on language as a structural barrier, noting that non-English speakers become second-class citizens in technology communities and advocating for AI translation to enable multilingual participation . Dr. Zhang similarly emphasised that capacity building must use multilingual interfaces and tools adapted to local contexts , and that programmes designed in Geneva or Washington may not suit African needs . Both pointed to existing technologies as solutions that can be deployed now. All three speakers converged on the view that capacity building must be genuinely responsive to local needs and co-designed with local stakeholders rather than imposed from outside. Prof. Ke Gong drew on pilot training experience to conclude that programmes must face real needs and work closely with local engineering organisations . Dr. Zhang argued that pre-packaged solutions designed in Geneva or Washington are likely poorly adapted to local demand . Dr. Robles Pelayo went further, arguing that capacity building must be fundamentally redefined as co-design of governance and algorithmic auditing rather than a mechanism for producing cheap labour . Both Prof. Stückelberger and the South African audience member highlighted structural socioeconomic barriers that go beyond technical capacity gaps. Prof. Stückelberger identified capital as the most serious bottleneck, noting that billions of African private capital sit in Geneva and London rather than being reinvested in Africa due to risk perceptions . The South African audience member highlighted that strong union culture and job displacement fears have prevented even basic automation such as self-checkout machines from being introduced in South Africa . Both pointed to the need to address these structural and social dimensions alongside technical capacity building. Both Prof. Krawczyk and Mr. Dutz emphasised the role of open data and open-source resources as public goods that can democratise access to AI capabilities. Prof. Krawczyk highlighted that open geospatial data is now widely available at building and street level and can serve as a foundation for AI-assisted urban planning in Africa , describing this as a tremendous open public good for African countries . Mr. Dutz similarly argued for making open-source project resources available in multiple languages through AI translation, framing this as a public good for global participation . Both saw openness and accessibility of digital resources as key enablers for the Global South.
It was somewhat unexpected that speakers from very different backgrounds - a global ethics professor, a corporate lawyer and labour rights advocate, a climate and urban planning researcher, and a higher education practitioner - all converged on the view that technical solutions alone are insufficient. Prof. Stückelberger explicitly urged the audience not to look only at technical or financial factors but to treat all six dimensions as interrelated . Dr. Robles Pelayo argued that the labour exploitation embedded in AI development is not merely an ethical problem but a violation of existing international norms . Prof. Krawczyk raised governance questions about data ownership that go beyond technical data management . The South African audience member noted that even AI policies in higher education institutions are being driven by researchers concerned about ethical use rather than technical capacity . This convergence across disciplines on the primacy of non-technical factors was a notable area of unexpected consensus.
It was unexpected that language emerged as a point of consensus across speakers who were not primarily focused on linguistic issues. Mr. Dutz made language exclusion a central argument, noting that non-English speakers become second-class citizens in technology communities and that during the session itself requests for subtitles were declined . The South African audience member then directly asked Mr. Dutz about Chinese language inclusion , prompting a broader discussion. Prof. Stückelberger, speaking about China-Africa cooperation, also identified language as a component of the intercultural learning that must be embedded in every investment programme . The audience member from South Africa raised the issue of AI translation for Chinese as well . This convergence on language as a structural barrier - from an open-source foundation director, a global ethics professor, and an African higher education practitioner - was not anticipated given the session's primary focus on engineering and AI capacity building.
Speakers from quite different perspectives - a global ethics professor, a labour rights lawyer, and a government representative - converged on the concern that international engagement with Africa in the AI and technology space risks reproducing extractive or neocolonial patterns. Prof. Stückelberger described foreign capital as needing to be fair as win-win and not with neocolonial dominance , and described the confidential US-DRC deal that reportedly displaced 600,000 artisanal miners as an example of the conflict potential . Dr. Robles Pelayo argued that the current model of capacity building effectively reduces the Global South to a source of cheap labour . The Kenyan government representative noted that the Global South has increasingly been used as a consumer and raw material source rather than an equal partner . This convergence across a government official, an academic ethicist, and a legal advocate on the neocolonial risk was unexpected given their different starting points.
The discussion revealed a high degree of consensus across speakers from diverse backgrounds - engineering, urban planning, global ethics, labour law, digital governance, open-source software, and government - on several core themes. First, all speakers agreed that Africa faces a severe and multidimensional AI and digital capacity gap that requires urgent, targeted, and locally grounded responses . Second, there was broad agreement that capacity building must be demand-driven and co-designed with local stakeholders rather than imposed from outside . Third, data sovereignty emerged as a shared concern, with multiple speakers arguing that Africa needs its own data infrastructure, legal frameworks, and community rights over data . Fourth, speakers converged on the view that technical solutions alone are insufficient and that governance, ethics, labour rights, cultural sensitivity, and political will are equally critical . Fifth, language barriers were identified as a structural form of exclusion that AI tools can help address . Areas of less consensus included the specific mechanisms for returning value from AI companies to communities - with Prof. Krawczyk proposing token taxation and Dr. Robles Pelayo focusing on labour rights and independent audits - and the relative priority of different barriers, with Prof. Stückelberger emphasising capital and governance while Dr. Zhang focused on parallel infrastructure and human capacity development .
Prof. Ke Gong frames capacity building primarily as a technical and engineering skills programme to address Africa's lag on SDGs , and Dr. Xiao Zhang refines this by advocating layered, demand-driven training for different audiences such as policymakers and engineers . Dr. Robles Pelayo fundamentally challenges this framing, arguing that capacity building as currently conceived reduces the Global South to a source of cheap labour for AI companies, and must instead be redefined as genuine co-design of governance, algorithmic auditing, and participation in global rule-making . This represents a deeper disagreement about the political economy of capacity building, not merely its methodology.
Africa is severely behind on SDGs, with less than 7% of measurable goals on track, necessitating a targeted engineering capacity building programme leveraging digital technology - Engineering capacity gap in Africa
Training should be built in layers tailored to different audiences, such as policymakers needing AI governance understanding and engineers needing practical tools, rather than a one-size-fits-all approach - Layered, targeted training
Capacity building in the Global South must be redefined as co-design of governance and algorithmic auditing, not as a source of cheap labour - Redefining capacity building as co-design
Prof. Krawczyk approaches data sovereignty from the bottom up, arguing for a fundamental right to data at the community level , templates for communities to share data in exchange for benefits , and taxation on AI tokens or data to return value to people . Prof. Stückelberger approaches the same issue from the top down, emphasising the need for physical data centre infrastructure on African soil and African Union legislation . While not mutually exclusive, these represent meaningfully different priorities: Krawczyk focuses on individual and community-level rights and financial mechanisms, whereas Stückelberger focuses on national and continental institutional infrastructure and governance.
Communities and individuals should have a fundamental right to their data, giving them a negotiation position when AI companies collect and exploit community data - Fundamental right to data
AI companies accumulate enormous financial value from knowledge distilled from community data, and mechanisms such as token taxation or data levies should return value to the people whose data was used - Taxation on AI outputs to return value
Africa urgently needs its own data centres to achieve data sovereignty, as current dependence on US, Chinese, and European data centres undermines African control over its own data - Need for African data centres
The African Union and member states are working on data sovereignty legislation, but achieving unified African data governance requires overcoming national partial interests - African data sovereignty legislation
Prof. Stückelberger presents a deeply ambivalent picture of foreign investment, warning that foreign capital must be fair and win-win rather than neocolonial , citing the example of a confidential US-DRC deal that allegedly displaced 600,000 artisanal miners , and noting that Chinese companies in Colvesia live in isolated camps creating automatic conflict . The Kenyan government representative, by contrast, takes a more optimistic stance, welcoming international collaboration and expressing confidence that Kenya's rule of law can guide large tech companies towards minimum labour requirements . This reflects a genuine disagreement about whether existing governance frameworks are sufficient to manage the risks of foreign investment.
Investment capital is the most serious bottleneck for AI development in Africa, with billions of African private capital sitting in Geneva and London rather than being reinvested in Africa due to risk perception - Capital investment bottleneck
Kenya's national AI strategy 2025–2030 places capacity building at its core, investing in local talent, digital infrastructure, digital hubs, and public Wi-Fi to enable citizens to participate as equal partners in global AI development - Kenya's national AI strategy
Dr. Xiao Zhang explicitly argues that Africa should not wait for full infrastructure completion before beginning human capacity building, stating definitively that the answer to whether to wait is 'no' , and that existing mobile devices, cloud-based AI services, and open-source models on modest hardware can be used immediately . The Kenyan government representative, while also calling for parallel investment, places greater emphasis on the infrastructure gap as a primary challenge, noting that ministers at the forum were reluctant to admit their infrastructure is poor , and calling for international collaboration specifically on infrastructure including data centres . This suggests a difference in emphasis regarding the urgency and sequencing of infrastructure versus human capacity investment.
Infrastructure investment must proceed in parallel with human capacity building, using existing mobile technology and cloud-based AI services rather than waiting for full infrastructure completion - Parallel infrastructure and human capacity development
National efforts alone are insufficient and must be complemented by global solidarity and international collaboration in AI capacity building and infrastructure - Need for global solidarity alongside national efforts
Mr. Dutz identifies language as the primary structural barrier to participation, arguing that non-English speakers become second-class citizens in global technology communities , and noting that even during the session itself, requests for subtitles were declined . He frames AI translation as a near-solved technical problem that should be deployed to remove this barrier . Prof. Ke Gong, by contrast, emphasises the barrier of remote programme design disconnected from local realities , and the Apache Travel Assistance Committee context highlights physical travel and visa barriers as equally significant. These represent different diagnoses of the primary exclusion mechanism, with different implied solutions.
Non-English speakers are effectively treated as second-class citizens in global technology communities, as the vast majority of knowledge and open-source resources are available only in English - Language exclusion in technology
Capacity building must respond to real local needs and work closely with local organisers rather than being designed remotely - Local engagement is essential
Dr. Robles Pelayo frames the exploitation of Kenyan data workers as a clear violation of existing international law, citing ILO Conventions 87 and 88, the International Covenant on Economic, Social and Cultural Rights, and UN Guiding Principles , and notes that the Kenyan Court of Appeal has confirmed jurisdiction over Meta . The Kenyan government representative, while acknowledging the problem, frames it more as a matter of domestic rule of law and regulatory guidance, expressing confidence that 'the law will prevail' and that big tech companies have been guided on minimum labour requirements . This reflects a disagreement about whether existing legal frameworks are adequate or whether new international governance instruments are needed.
Documented cases, such as Kenyan workers paid $2 per hour to review extreme violence for OpenAI while suffering PTSD and clinical depression, illustrate severe exploitation in AI data supply chains - Exploitation of data workers in Kenya
AI is not autonomous but depends on millions of invisible data workers in the Global South who label data and review harmful content under conditions that violate international labour law - Invisible labour underpinning AI
It is unexpected that the Kenyan government representative, speaking at the same session where Dr. Robles Pelayo had just presented detailed evidence of systematic labour exploitation of Kenyan workers by AI companies , responded by expressing confidence in Kenya's rule of law and noting that big tech companies 'have been guided in terms of the best minimum labour requirements' and that 'the law will prevail' . Rather than engaging with the specific documented violations, the representative pivoted to Kenya's strengths as an innovation hub and the quality of its educated youth . This unexpected divergence suggests a significant gap between the government's self-assessment of its regulatory capacity and the documented reality of labour conditions on the ground.
When an audience member asked how China can better help developing countries , the discussion unexpectedly revealed a disagreement between Mr. Dutz and Prof. Stückelberger about the nature of the China-related barrier. Mr. Dutz focused on the technical and platform dimension, arguing that Chinese communities use WeChat and other tools inaccessible to people outside China , making this a problem of communication platform exclusivity. Prof. Stückelberger, by contrast, focused on cultural and interpersonal dimensions, emphasising the need for intercultural sensitivity, mutual learning, and understanding of different cultural practices such as time management and funeral attendance . This was unexpected because both speakers were responding to the same question but diagnosed fundamentally different problems requiring different solutions.
The session opened with a broadly optimistic framing of AI as a tool to address Africa's development gap , and this optimism was echoed by Dr. Zhang and the Kenyan government representative who described AI as 'a new opportunity for Kenyans to actually assert themselves' . This made it unexpected when Dr. Robles Pelayo introduced a structural critique arguing that the very AI systems being promoted as opportunities for Africa are built on the exploited and invisible labour of African workers , and that the workers teaching AI to detect harmful content are simultaneously making their own jobs redundant . The disagreement was not merely about solutions but about whether the premise of AI as opportunity for Africa is accurate without first addressing the labour exploitation embedded in AI development.
Prof. Stückelberger raised an unexpected concern about energy prioritisation, warning that if new energy sources are directed primarily to data centres and AI infrastructure, communities will be left without power for basic needs such as cooking and school lighting, potentially leading to social unrest . This directly tensions with Prof. Krawczyk's advocacy for expanding AI infrastructure including data centres and geospatial data platforms in Africa , which would require significant energy resources. While Krawczyk did not explicitly address energy trade-offs, his enthusiasm for building open data infrastructure and AI capacity implicitly assumes energy availability that Stückelberger questions. This energy dimension was not anticipated as a point of contention in what was framed as a capacity building discussion.
The discussion reveals a multi-layered set of disagreements operating at different levels of analysis. At the surface level, speakers broadly agree that Africa faces a significant AI and digital capacity gap requiring urgent action. However, beneath this surface consensus lie substantial disagreements about: (1) whether capacity building should be framed as technical skills training or as structural co-design of governance ; (2) whether data sovereignty is best achieved through community rights and taxation or national infrastructure and legislation ; (3) whether foreign investment is primarily an opportunity or a neocolonial risk ; (4) whether infrastructure must precede or run in parallel with human capacity development ; (5) whether language or physical access is the primary barrier to global participation ; and (6) whether existing legal frameworks are adequate to address AI labour exploitation or whether new international instruments are needed . The most fundamental disagreement is between the predominantly technocratic and optimistic framing of most speakers and Dr. Robles Pelayo's structural critique that AI development is built on exploited African labour , which challenges the premise of the entire discussion.
All speakers agree that Africa faces a significant AI and digital capacity gap that requires urgent attention, and that capacity building is a necessary response . However, they disagree substantially on what capacity building should look like, who should design it, and what structural conditions must accompany it. Prof. Ke Gong and Dr. Xiao Zhang focus on technical and skills training , Prof. Krawczyk adds urban planning and AI integration , Prof. Stückelberger insists on addressing six interrelated structural factors , and Dr. Robles Pelayo argues the entire framing must be reconceived as co-design rather than training .
Africa is severely behind on SDGs, with less than 7% of measurable goals on track, necessitating a targeted engineering capacity building programme leveraging digital technology - Engineering capacity gap in Africa Capacity building should be demand-driven, addressing real challenges faced by local technicians and policymakers rather than offering pre-packaged solutions designed in Geneva or Washington - Start from real local needs AI capacity building should be expanded to include urban planning, integrating AI and geospatial data technologies to address Africa's rapid urbanisation challenges - Urban planning and AI integration Good governance and political will are the most critical factors for achieving sovereign AI capacities in Africa, and all six dimensions — people, minerals, capital, data, energy, and politics — must be addressed as interrelated factors - Six interrelated factors for AI sovereignty Kenya's national AI strategy 2025–2030 places capacity building at its core, investing in local talent, digital infrastructure, digital hubs, and public Wi-Fi to enable citizens to participate as equal partners in global AI development - Kenya's national AI strategy
Prof. Krawczyk, Prof. Stückelberger, and Dr. Robles Pelayo all agree that the current model of AI development extracts value from Africa and the Global South without adequate return . They share the goal of ensuring that African communities and workers receive fair value from their data and labour. However, they propose different mechanisms: Krawczyk focuses on community data rights and token taxation , Stückelberger on national data centres and AU legislation , and Robles Pelayo on labour audits, union representation, and governance co-design . The shared diagnosis of extraction without return does not translate into a shared remedy.
Communities and individuals should have a fundamental right to their data, giving them a negotiation position when AI companies collect and exploit community data - Fundamental right to data Africa urgently needs its own data centres to achieve data sovereignty, as current dependence on US, Chinese, and European data centres undermines African control over its own data - Need for African data centres No capacity building project should be funded without independent audits of working conditions throughout the data supply chain - Independent audits of labour conditions
Prof. Ke Gong and Dr. Xiao Zhang both strongly agree that capacity building must be grounded in local realities and real needs rather than externally designed packages . The Kenyan government representative echoes this by emphasising Kenya's own national strategy and local talent investment . However, they differ on the balance between local ownership and international collaboration: Prof. Ke Gong emphasises working with local engineering organisations , Dr. Zhang emphasises identifying exact local problems , and the Kenyan representative stresses that national efforts alone are insufficient and global solidarity is needed , suggesting different views on the appropriate locus of programme design authority.
Capacity building should be demand-driven, addressing real challenges faced by local technicians and policymakers rather than offering pre-packaged solutions designed in Geneva or Washington - Start from real local needs Capacity building must respond to real local needs and work closely with local organisers rather than being designed remotely - Local engagement is essential National efforts alone are insufficient and must be complemented by global solidarity and international collaboration in AI capacity building and infrastructure - Need for global solidarity alongside national efforts
Mr. Dutz, Prof. Krawczyk, and Dr. Zhang all agree that existing AI and digital technologies can be deployed immediately to address inclusion gaps without waiting for perfect conditions . They share an optimistic view of AI as a practical tool for bridging divides. However, they focus on different applications: Dutz on language translation for open-source participation , Krawczyk on geospatial AI for urban planning , and Zhang on cloud-based AI services and open-source models for capacity building . Their agreement on the principle of using available technology diverges when it comes to which technology to prioritise and for what purpose.
AI translation tools now make it technically feasible to make resources available simultaneously in multiple languages, removing language as a barrier to participation in global open-source and technology projects - AI as a tool for language inclusion AI capacity building should be expanded to include urban planning, integrating AI and geospatial data technologies to address Africa's rapid urbanisation challenges - Urban planning and AI integration Infrastructure investment must proceed in parallel with human capacity building, using existing mobile technology and cloud-based AI services rather than waiting for full infrastructure completion - Parallel infrastructure and human capacity development
- Africa is severely behind on Sustainable Development Goals, with less than 7% of measurable goals on track, making targeted engineering and AI capacity building an urgent priority.
- Capacity building programmes must be demand-driven and grounded in real local needs, designed in close collaboration with local organisers rather than being pre-packaged solutions developed remotely in Geneva or Washington.
- Training should be structured in layers tailored to different audiences — policymakers need AI governance understanding whilst engineers and technicians need practical tools — rather than adopting a one-size-fits-all approach.
- Infrastructure investment and human capacity development must proceed in parallel, leveraging existing mobile technology and cloud-based AI services rather than waiting for full infrastructure completion.
- AI capacity building should be expanded to include urban planning, integrating AI and open geospatial data technologies to address Africa's rapid urbanisation challenges in areas such as sanitation, street networks, and public transport.
- Communities and individuals should have a fundamental right to their data, providing a negotiation position when AI companies collect and exploit community data, and mechanisms such as token taxation or data levies should return value to the people whose data was used.
- Africa urgently needs its own data centres to achieve data sovereignty, as current dependence on US, Chinese, and European data centres undermines African control over its own information.
- AI development depends on millions of invisible data workers in the Global South who label data and review harmful content under conditions that frequently violate international labour law, including documented cases of Kenyan workers suffering PTSD and clinical depression whilst being paid as little as $2 per hour.
- No capacity building project should be funded without independent audits of working conditions throughout the data supply chain, and data workers and their unions must be included in global AI governance dialogues.
- Capacity building in the Global South must be redefined as co-design of governance and algorithmic auditing, not as a source of cheap labour.
- Investment capital is the most serious bottleneck for AI development in Africa, with billions of African private capital sitting in Geneva and London rather than being reinvested in Africa due to risk perception.
- All six interrelated dimensions — people and values, critical minerals, capital, data sovereignty, green energy, and good governance — must be addressed together to achieve sovereign AI capacities in Africa.
- Non-English speakers are effectively treated as second-class citizens in global technology communities, but AI translation tools now make it technically feasible to enable multilingual participation in open-source and technology projects.
- Chinese technology communities tend to use platforms such as WeChat that are inaccessible to people outside China, limiting broader global collaboration.
- National efforts such as Kenya's AI Strategy 2025–2030 are important but insufficient on their own; global solidarity and international collaboration are essential complements.
- China's engagement in Africa has significant positive impacts but requires greater intercultural sensitivity, mutual learning, and investment in vocational training to build genuine trust and avoid social conflict.
“Prof. Krawczyk proposed integrating AI and open geospatial data into urban planning capacity building for Africa, arguing that rapid urbanisation in Africa and Asia creates both a challenge and an opportunity. He suggested developing courses that attract urban planners to learn AI capacity and vice versa, and proposed creating an open public geospatial dataset for African countries as a 'tremendous open public good.'”
“Krawczyk raised the question of data ownership and community rights, arguing that individual data points have little value but community-level data does. He proposed three mechanisms: a fundamental right to data for communities, templates for data-sharing agreements that return value to communities, and a form of taxation on AI outputs (tokens) to redistribute value back to the people whose data trained the models.”
“Prof. Stückelberger presented a six-dimensional framework for AI sovereignty in Africa, encompassing: people and values (including corruption and nepotism), critical minerals (e.g., cobalt from DRC), capital sovereignty, data sovereignty, green energy, and politics/governance. He argued these six factors are deeply interrelated and that focusing only on technical or financial dimensions misses the full picture.”
“Dr. Robles Pelayo highlighted the invisible labour underpinning AI systems, specifically the data labellers and content moderators in the Global South who are paid exploitative wages (e.g., Kenyan workers paid $2/hour versus the $12.50/hour charged to OpenAI by the subcontractor), suffer post-traumatic stress disorder from reviewing harmful content, and are dismissed when they attempt to unionise. He argued this constitutes a violation of international labour law and called for independent audits of working conditions in data supply chains, inclusion of data workers' unions in AI governance, and a redefinition of capacity building as co-design rather than cheap labour provision.”
“Mr. Dutz pointed out that the language barrier — specifically the dominance of English in open-source communities and global AI development — creates a 'second-class citizenship' for non-English speakers. He noted the irony that subtitle requests during the very session being discussed were declined, and argued that AI translation tools could and should be used to make open-source participation accessible in any language, including African languages.”
“Dr. Zhang argued that capacity building should not wait for infrastructure to be complete, but should proceed in parallel using existing tools (cloud-based AI, open-source models on modest hardware, multilingual interfaces). She also warned against 'prepackaged supply' — programmes designed in Geneva or Washington that do not reflect local demand — and advocated for layered, targeted training that distinguishes between the needs of policymakers, software developers, and network operators.”
How can AI capacity building programmes in Africa be better integrated with urban planning, particularly leveraging open geospatial data and AI technologies?
Africa and Asia are experiencing rapid population growth with significant urban challenges including sanitation, street networks, and public transport. Developing courses that combine urban planning with AI capacity could provide a foundation for sustainable urban development, making this a critical area for further research and programme design.
Who owns the data collected from African communities by AI companies, and what frameworks should govern data ownership and benefit-sharing?
When AI providers collect community health or other data to develop solutions, the data and resulting intellectual property typically remain with the AI company. Establishing clear data ownership rights and benefit-sharing mechanisms is essential to ensure communities receive fair returns from their data contributions.
Should there be a fundamental right to data for communities, and how could such a right be legally established and enforced?
Individual data points have little value, but community-level data is significant. Establishing a fundamental right to data for communities would provide a negotiating position and legal basis for communities to benefit from their collective data, requiring further legal and policy research.
What templates or frameworks could enable communities to share their data with AI companies in exchange for tangible returns such as capacity building or revenue streams?
There is currently no standardised mechanism for communities to negotiate data-sharing agreements with AI companies. Developing such templates would help communities in the Global South protect their interests and benefit from AI development, warranting further research into legal and contractual models.
Could a taxation mechanism on AI outputs (e.g., tokens) be used to redistribute value back to communities whose data contributed to AI development?
AI companies derive enormous financial value from knowledge distilled from community data. Exploring taxation on data or AI output tokens as a redistribution mechanism could address global inequities in AI value capture, requiring further economic and policy research.
How can African countries assert greater sovereignty over their critical mineral resources, particularly in relation to AI hardware dependencies, and what legal frameworks govern long-term mining contracts?
Africa holds vast critical mineral resources essential for AI hardware, yet ownership and processing largely remain with foreign companies. The tension between long-term contracts and national industrialisation goals raises complex legal and geopolitical questions requiring further investigation.
How can private capital held by African investors in European and North American banks be redirected to invest in Africa's AI and digital infrastructure?
Significant African private capital sits in banks in cities like Geneva but is invested elsewhere due to perceived risk. Developing trust-building narratives and risk management frameworks to attract this capital back to Africa is a critical area for further research in development finance.
How can African nations collectively develop shared data centres to achieve data sovereignty, and what governance structures would be needed?
Currently, the vast majority of data centres are located in the US, China, and Europe, creating dependency for African nations. Researching how the African Union and member states could coordinate to build shared data infrastructure while balancing national interests is essential for digital sovereignty.
How can energy production in Africa be balanced between meeting household needs and powering data centres and AI infrastructure?
Data centres require enormous energy, yet many African households lack reliable electricity. Research is needed into energy policy frameworks that prioritise people-centred access while also enabling the digital infrastructure necessary for AI development.
Should AI capacity building programmes wait for full infrastructure development, or should human capacity building proceed in parallel using existing mobile and cloud-based tools?
Many African countries lack adequate digital infrastructure, yet delaying capacity building until infrastructure is complete may widen the intelligence divide further. Research into effective parallel strategies for infrastructure investment and human capacity development is needed.
How can AI capacity building programmes be better tailored to local needs rather than being designed in Geneva or Washington and applied uniformly across Africa?
Prepackaged capacity building programmes designed in high-income countries may not address the specific challenges faced by local technicians and policymakers in Africa. Research into demand-driven, locally co-designed training methodologies is needed to improve effectiveness.
How should AI capacity building be structured in layers to address the distinct needs of different stakeholder groups such as policymakers, software developers, and network operators?
A one-size-fits-all training approach is likely inefficient. Research into layered, role-specific capacity building frameworks that address the distinct knowledge requirements of different professional groups would improve outcomes and resource utilisation.
What labour standards and independent audit mechanisms should be required for data annotation and content moderation workers in the Global South, particularly in Africa?
Documented cases, particularly in Kenya, show that data workers are paid exploitative wages and suffer serious psychological harm from reviewing harmful content. Research into enforceable international labour standards and independent audit frameworks for AI supply chains is urgently needed.
How can data workers' unions and worker representatives be formally included in global AI governance dialogues and rule-making processes?
Those who perform the invisible labour that makes AI systems function are currently excluded from governance discussions. Research into mechanisms for meaningful worker representation in AI governance bodies would help address this democratic deficit.
How can AI capacity building in the Global South be redefined as co-design of governance and algorithmic auditing rather than provision of cheap labour?
Current models risk reducing the Global South to a source of low-cost labour for AI development. Research into alternative models that position Global South participants as co-designers of AI governance and auditing frameworks would promote more equitable outcomes.
How can AI translation tools be used to overcome language barriers in open source and international technical communities, enabling participation in languages other than English?
Non-English speakers are effectively second-class participants in global technical communities. Research into deploying AI translation tools to make mailing lists, documentation, and discussions accessible in multiple languages simultaneously could significantly broaden participation from Africa and other regions.
How can Chinese technical communities be encouraged to use communication platforms that are accessible to international collaborators, rather than tools such as WeChat that are not accessible outside China?
Chinese open source and technical communities frequently conduct discussions on platforms inaccessible to international participants, undermining global collaboration. Research into inclusive communication norms and platform choices for international technical communities is needed.
How can intercultural competency and mutual cultural learning be systematically embedded in China-Africa investment and technology transfer programmes?
Cultural misunderstandings and different working norms create tensions in China-Africa collaborations. Research into effective intercultural training components for investment programmes, including language learning and cultural sensitivity, could improve outcomes and reduce conflict.
How can research collaboration between Chinese and African higher education institutions be strengthened to support transdisciplinary AI development at the ground level?
Researcher-to-researcher collaboration between Chinese and African institutions was identified as a promising avenue for ground-level transdisciplinary work. Further research into frameworks and funding mechanisms for such collaboration is warranted.
How should AI policies in higher education institutions be developed to prevent students from being unfairly penalised for using AI tools, while ensuring ethical use?
As AI tools become widespread in academic settings, institutions need clear ethical use policies that distinguish legitimate AI-assisted work from plagiarism. Research into policy frameworks that balance academic integrity with the realities of AI tool adoption is needed.
Are digital divides creating new age-related inequities, and what rights frameworks should be developed to protect older people in the context of AI adoption?
The discussion touched on multiple dimensions of the intelligence divide, but age-related divides were identified as an underexplored area. Research into how AI adoption affects older populations and what policy protections are needed was flagged as a priority for further investigation.
How can the Engineering Capacity Building for Africa programme scale beyond pilot training in Kenya, Uganda, and Cabo Verde to establish sustainable training centres across all five African sub-regions?
Pilot trainings have yielded positive feedback, but the programme needs to address how to scale effectively while remaining responsive to local needs and working closely with local engineering organisations rather than operating remotely. Research into scalable, locally embedded capacity building models is needed.
How can open geospatial datasets, such as Microsoft's open building dataset, be combined with other open data sources to create a comprehensive open public good for urban planning in African countries, and what would this cost?
Open geospatial data exists but needs to be combined across sources to be truly useful for urban planning. Prof. Krawczyk estimated costs in the low millions range, suggesting this is feasible but requires further research into data integration, governance, and funding mechanisms.
