Meta Bursary Abstracts

Meta provided the African Observatory with funding to support the attendance of emerging AI researchers at the AfricAI Conference in Kigali, Rwanda on 12 - 14 June 2023

The winners of the award were selected for their commitment to working on AI ethics, human rights and policy/governance in African research. Beyond the AfricAI conference, the African Observatory provides the awardees with mentorship as well as support for the publication of their work. Read more about the selected candidates work in the abstracts below.

Smart townships in South Africa: A Double-edged sword

Mabona Machaba

South African townships are among the world’s most unequal and segregated. These townships are primarily impacted by a variety of social issues, including unemployment, housing shortages, inadequate infrastructure, and unregulated government. To address these challenges, President Cyril Ramaphosa alluded to his dream of building smart cities to prepare for the fourth industrial revolution in 2020, and has since approved the Greater Lanseria Master Plan, a smart city that would house millions of people. While smart cities have been a dominant discourse, little is said about the future of townships in the fourth industrial revolution. So, with the fourth industrial revolution in full swing, we must ask whether townships will be able to cope with the new technological changes given their infrastructural issues and digital divide.

This paper seeks to expand on research that has begun to investigate the sociological feasibility of smart townships in South Africa. It draws attention to the uniquely intertwined benefits and challenges of developing smart townships in South Africa. This paper draws its findings from implemented government policies to demonstrate the policies' attempts and failures to prioritize townships in terms of infrastructure development, training, and partnership opportunities. Townships must be included in South Africa's 4IR strategy to ensure that the majority of people have access to technologies such as the Internet of Things and virtual reality, among others. This paper argues, through policy analysis, that 4IR policies should prioritize the unique realities of townships, taking into account their marginality, in order to ensure that townships become new innovation hubs in the country. The paper recommends that policy prioritize smart townships in economic agendas while opening doors to informal sectors and entrepreneurship, as well as linking ICT policies in townships within this 4IR context.

The Place of Artificial Intelligence in African Education

Diana Nyakundi

Artificial Intelligence will have powerful impacts on learning not only through direct support for students, but also by empowering educators to be more adaptive to learner needs. In spite of the diverse range of uses of AI technologies, the use of AI in education to enhance learning and to foster skills for jobs and the adaptation of life in the age of AI has not been as prevalent. Since Aiken & Epstein published their ethical guidelines two decades ago (Aiken & Epstein, 2000), there has been limited published work that explicitly focuses on ethics and the potential impact of AIED designs and deployment methods. Relatively few initiatives focus on AI in education leading to a recent recommendation that policymakers should ‘provide an enabling policy environment and curricular spaces for exploring AI’ (UNESCO Releases Report on the Mapping of K-12 Artificial Intelligence Curricula | UNESCO, n.d.).

However, the use of Artificial Intelligence in Education (AIED) raises numerous questions and ethical concerns for students, educators, parents, policymakers, and other stakeholders. Some of these concerns include privacy issues and the need to ensure that consent is genuinely informed and given by learners, surveillance by teachers, the autonomy of learners, gaps in datasets, and the prevention of bias on the basis of gender, race, social status, income inequality, religion, and family status among others.

Ethical and Human Rights Concerns on the Use of Automated Decision Making in Migration Management: Questioning AI Governance

Halifet Aye

This paper provides the technical aspects of AI decision-making systems in migration management. It explains how they can considerably impact the lives of vulnerable groups from a human rights perspective. It also provides an ethical analysis of the AI application in the immigration and refugee arena and mainly focuses on automated decision systems. In doing so, it relies on reviewing various literature available in the area. The findings demonstrate that using automated decision-making and other AI technologies has opened windows of opportunities for advancing migration governance. However, without critical ethical analysis, the use of automated decision-making may result in moral issues and violations of a variety of rights, including the rights to equality and non-discrimination; freedom of movement, expression, religion, and association; privacy rights, and rights to life, liberty, and security of the person. 

Hence, it is crucial to address the drawbacks of automated decision-making, and people on the move should be part of the discussions shaping it. More importantly, the same protection the migrating community has offline in various human rights instruments must also be included online to ensure their security and enhance migration governance using AI technologies to provide a better ethical outcome.

AIoT in Africa: Opportunities and Challenges for Ethics and Society

Bridget Chipungu

This paper explores the ethical and societal implications of AIoT technologies in the African context, focusing on the intersection of artificial intelligence and the Internet of Things (IoT), and the ethical concerns that arise at this intersection. While AIoT technologies have the potential to bring significant benefits in areas such as healthcare, agriculture, transportation, and energy management, they also pose significant ethical and societal challenges. For example, there may be concerns related to data privacy and security, as well as algorithmic bias and transparency. Additionally, the adoption of AIoT technologies may lead to job displacement and exacerbate existing inequalities in access to technology and economic opportunities.

Drawing on existing literature and case studies, this paper explores these challenges and proposes strategies to address them, including working with policymakers, stakeholders, and communities to ensure that AIoT technologies are developed and deployed in ways that prioritize the needs and interests of individuals and communities, and that potential risks and harms are identified and mitigated.

Overall, this paper argues that by taking a proactive and responsible approach to the deployment of AIoT technologies in Africa, it may be possible to realize the significant benefits that these technologies have to offer while minimizing potential harms and ensuring that the needs of individuals and communities are prioritized. The paper will focus on the intersection of AI and IoT, and the ethical concerns that arise at this intersection.

Assessing the Implications of Foreign Technology Dependence on AI Development in Africa: An Analysis of Technology Transfer and Acquisition Strategies and their Socio-Economic and Political Impacts

Hanani Hlomani

The adoption and use of AI products has been promoted in many African countries due to their potential to improve lives, reduce poverty, and promote economic inclusion. These promises speak to the aspirations of most governments’ development plans with many African countries naming AI as a fundamental pillar of their national development strategies.

The problem, however, is that Africa's technological advancement has been largely reliant on foreign technology transfer and acquisition, particularly in the area of AI. This paper provides an analysis of the implications of foreign technology dependence on AI development in Africa.

The paper investigates African nations' various technology transfer and acquisition strategies, with the intention of demonstrating that, while most strategies facilitate technology adoption, they also pose significant risks of exploitation by foreign firms which may result in economic and political vulnerabilities, potentially jeopardising national security and sovereignty. Documents of interest include the 4 national AI strategies, national development plans and draft policy papers in selected African countries.

The paper points out that the social, economic, and political effects of these transfer and acquisition strategies are shaped by what African governments and societies choose to do. So, African countries cannot just let other countries impose their technology on them and on foreign terms; they need to take an active role in shaping their own technological futures. African countries need to take a strategic approach to acquiring and transferring technology. They can do this by making rules and policies that encourage local ownership and control of technology and intellectual property, as well as by investing in the development of their own workers and the transfer of skills. Also, African governments need to make it easier for local companies and foreign partners to work together to develop technology that fits local needs.

AI Ethics in Wildlife Conservation

Irene Nandutu

With the increased use of Artificial Intelligence (AI) in wildlife conservation, issues around whether AI-based monitoring tools that comply with standards regarding AI Ethics are on the rise. This study seeks to extend the literature on (1) current AI Ethics and AI Ethics issues in wildlife conservation, (2) Initiatives Stakeholders in AI for wildlife conservation should consider integrating AI Ethics in wildlife conservation.

The work follows an inductive method using a grounded theory to develop an AI ethically aligned framework. The proposed framework uses agile software life cycle methodology to implement guidelines towards the ethical upgrade of any existing AI system or the development of any new ethically aligned AI system. The guidelines enforce, among others, the minimization of intentional harm and bias, diversity in data collection, design compliance, auditing of all activities in the framework, and ease of code inspection. This framework will inform AI developers, users, conservationists, and policymakers on what to consider when integrating AI Ethics into AI-based systems for wildlife conservation.

Ethics by Design: AI Governance Imperatives within the African Context

Raymond Onuoha

In the practice domain, in spite of the proliferation of AI ethics principles, there are still tensions that arise in implementing these principles. A key knowledge gap within this AI development domain is the absence of an empirically established method to translate ethical principles into practice. This interdisciplinary research proposes to explore the existing approaches and propose new mechanisms in resolving ethical tensions within the African context as they impact both policy and practice in the design and use of AI systems. The research will be facilitated by means of AI design use cases in partnership with Data Science Nigeria, with whom the applicant has a substantive working relationship (Official Letter of Support has been secured).

Using Privacy-preserving Synthetic Mobile Money Transaction Datasets for Financial Fraud Detection

Denish Azamuke

Mobile money has accelerated the uptake of digital financial services in Sub-Saharan Africa. However, the implications on data privacy and security issues are least studied, which can be attributed to the limited availability of datasets on mobile money transactions and user behavior. Due to the intrinsically private nature of the mobile money financial datasets, we developed a simulation platform tailored to mobile money transactions from Sub-Saharan Africa to generate high-quality synthetic financial datasets that are as close as possible to a sample of real mobile money transaction data obtained from one of our partner mobile money service providers in the region.

We call our simulation platform MoMTSim (Mobile Money Transaction Simulator) which is multi-agent-based and architected following the ODD (Overview, Design Concepts, and Details) model of design using a generic MASON toolkit that is multi-platform, fast at execution, supports parallelism, and is capable of reinforcing large-custom simulations that are computationally intensive like our simulator. 

Using the privacy-preserving synthetic datasets from the simulation platform that bear a resemblance to the real data in terms of statistical properties, unique transaction dynamics, and casual behaviors in the real ecosystem, we injected researched fraudulent scenarios; split-deposit fraud, refund fraud, and direct deposit fraud among others that were gathered from a survey used to determine “ground truth”. The survey involved the major stakeholder of mobile money service; agents, clients, merchants, banks, telco officers, and the security unit. Therefore, this study presents a cutting-edge financial simulation platform in the realm of mobile money service in the Sub-Saharan context for generating privacy-serving synthetic financial datasets.