Chapter 34: Exercises — AI Ethics in Emerging Markets

Comprehension Exercises

Exercise 1: The Diversity Within "Emerging Markets" Select three countries from different regions (e.g., Nigeria, Vietnam, and Colombia) that are commonly classified as "emerging markets." For each country, research: (a) Internet penetration rate and trajectory (b) AI development ecosystem (universities, startups, research institutions) (c) Existing AI governance frameworks or digital governance laws (d) Primary economic sectors that AI could most impact (e) Major AI-related challenges specific to that country's context

Write a 500-word comparison that demonstrates why a single "emerging markets AI ethics" framework would be inadequate for all three countries.

Exercise 2: The Leapfrog Analysis The chapter uses M-Pesa as an example of technology leapfrogging traditional infrastructure. Research two additional examples of technology leapfrogging in the Global South (e.g., solar energy replacing grid infrastructure, mobile banking replacing traditional banking, telemedicine replacing specialist referral systems).

For each example: (a) What traditional infrastructure was leapfrogged? (b) Who benefited from the leapfrog and who was excluded? (c) What governance frameworks emerged around the technology? (d) What does the example suggest about the conditions under which AI leapfrogging could work?

Exercise 3: Agricultural AI Assessment Research one agricultural AI application that has been deployed in sub-Saharan Africa or South Asia (e.g., Plantix, Hello Tractor, Apollo Agriculture, Twiga Foods). For each application: (a) What agricultural problem does it address? (b) What data was used to train the AI system? (c) Was the AI trained on data relevant to local agricultural conditions? (d) What infrastructure does the application require? (e) Who owns the data generated by the application? (f) Is there evidence of genuine benefit to smallholder farmers?

Exercise 4: Language Representation Audit Using publicly available information about the training data composition of any major large language model (Llama, Gemma, or similar model with published data cards), analyze the language representation in its training data: (a) What proportion of training data is in English? (b) What is the representation of African languages collectively? (c) What is the representation of the ten most spoken African languages individually? (d) What performance differences has research documented between English and African language performance? (e) What would it cost to meaningfully improve African language representation in a large language model?

Exercise 5: Medical AI Bias Documentation Research the documented evidence of medical AI bias affecting non-Western patient populations. Identify at least three specific applications where medical AI has shown documented performance gaps for African, Asian, or Latin American patient populations. For each: (a) What is the specific performance gap? (b) What caused it? (c) What are the potential health consequences? (d) What has been done to address it?

Analysis Exercises

Exercise 6: Data Colonialism Mapping Apply the "data colonialism" framework to a specific AI company operating in emerging markets. Choose one of the following: Facebook/Meta in Africa, Google in India, Jumia in multiple African markets, or Grab in Southeast Asia.

(a) What data does the company collect from users in the region? (b) Where is this data processed and stored? (c) Who owns the AI systems trained on or using this data? (d) What economic benefits do local users and communities receive? (e) What proportion of the economic value generated by this data flows back to the region? (f) Is the term "data colonialism" an appropriate characterization of this relationship? Why or why not?

Exercise 7: Surveillance Export Analysis The chapter discusses Huawei's Safe City deployments in Africa. Research the Carnegie Endowment for International Peace's "AI Global Surveillance Index" and: (a) Identify the five African countries with the highest documented AI surveillance technology deployment (b) For each country, assess its Freedom House freedom rating (c) Identify the primary technology suppliers for surveillance infrastructure in each country (d) Document any specific cases of surveillance technology being used against political opposition, journalists, or civil society (e) Assess what governance frameworks, if any, govern the use of this surveillance infrastructure

Exercise 8: Safe City Due Diligence Framework A municipality in a country with a mixed democratic track record (Freedom House rating: "partly free") is considering deploying a Safe City AI surveillance system. As an independent AI ethics consultant: (a) What information would you need to conduct a responsible assessment? (b) What governance requirements would you consider minimum conditions for an ethically defensible deployment? (c) What red lines — conditions under which you would advise against deployment regardless of other factors — would you identify? (d) What ongoing monitoring and accountability framework would you recommend?

Exercise 9: Annotation Labor Analysis The Sama Group case documents annotation labor conditions for OpenAI in Kenya. Research the broader AI annotation labor industry: (a) What is the estimated global market size for AI data annotation labor? (b) What countries are major centers of annotation labor, and what wages are typical? (c) What psychological hazards have been documented for content moderation and annotation workers? (d) What labor rights frameworks protect (or fail to protect) these workers? (e) What "living wage" for annotation workers in Kenya, the Philippines, and Pakistan would look like, and how would it compare to current wages?

Exercise 10: Masakhane NLP — Community-Driven AI Research Research the Masakhane NLP initiative in detail: (a) What is Masakhane's governance model and how are research priorities set? (b) What African languages has it built NLP capabilities for? (c) How does it address data sovereignty in its research design? (d) What funding sources does it rely on, and what are the governance implications of those funding relationships? (e) What has it accomplished that large technology companies' internal African language research has not? (f) What would it need to scale its impact?

Applied Exercises

Exercise 11: Responsible Deployment Framework Design a responsible AI deployment framework for a company deploying an AI-powered agricultural advisory service to smallholder farmers in East Africa. Your framework should address:

(a) Pre-deployment community engagement — who must be consulted and what must be agreed? (b) Data collection and governance — who owns data generated by the service, and what are the use limitations? (c) Infrastructure requirements — what connectivity and device requirements does the service impose, and how will it serve farmers who don't meet them? (d) Local language support — what languages must the service support? (e) Cultural appropriateness — what cultural considerations should shape the service's design? (f) Benefit-sharing — what economic benefits will farmers receive beyond access to the service? (g) Accountability — what recourse do farmers have if the service gives harmful advice?

Exercise 12: Labor Supply Chain Audit Your company uses AI systems that incorporate third-party AI components, including components trained using annotation labor provided by outsourced annotation services.

Design a labor supply chain due diligence process for AI annotation labor that includes: (a) Supplier qualification criteria (minimum standards for working conditions, wages, psychological support) (b) Pre-contract audit requirements (c) Ongoing monitoring mechanisms (d) Remediation and escalation processes for identified violations (e) Transparency and public reporting commitments

Exercise 13: Digital Infrastructure Gap Analysis Your company is developing an AI-powered healthcare diagnostic tool for deployment in sub-Saharan Africa. The system was initially designed to operate in cloud-connected mode, requiring reliable 4G internet connectivity.

Conduct an infrastructure gap analysis for deployment in three target countries (Nigeria, Ethiopia, and Tanzania): (a) What proportion of your target users would be excluded by the 4G connectivity requirement? (b) What offline or low-connectivity operation modes would be technically feasible? (c) What performance trade-offs would offline mode entail? (d) What electricity supply constraints would affect deployment? (e) Redesign your deployment approach to maximize reach across the infrastructure constraints you identified.

Exercise 14: National AI Governance Comparison Compare the national AI governance frameworks of three Global South countries: India, Kenya, and Brazil.

Using publicly available sources: (a) What AI-specific governance frameworks has each country adopted or proposed? (b) What data protection frameworks apply? (c) How does each country's framework address AI bias and fairness? (d) How does each framework address AI surveillance and privacy? (e) What is each framework's approach to AI in high-risk applications (healthcare, employment, credit)? (f) What gaps do you identify in each framework?

Exercise 15: Financial Inclusion AI — Ethical Assessment A fintech company is using mobile phone behavioral data (airtime patterns, app usage, location data, social network size) to build credit scores for borrowers in Kenya who lack traditional credit histories.

Conduct an ethical assessment that addresses: (a) What benefits does this model offer to underbanked borrowers? (b) What proxy discrimination risks does the use of mobile phone behavioral data create? (c) What transparency and explainability obligations should the company maintain? (d) Who owns the mobile phone data used to build the credit model? (e) What data governance agreements should be in place with the mobile network operators who provide the data? (f) What recourse do borrowers have if the credit model produces inaccurate or discriminatory decisions?

Critical and Reflective Exercises

Exercise 16: Is Leapfrogging Real? The chapter presents leapfrogging — the idea that developing economies can bypass traditional technological infrastructure development stages — as a genuine possibility for AI. Critics argue that leapfrogging narratives are often used by technology companies to market products in developing markets rather than reflecting genuine development potential.

Write a 500-word critical analysis of the leapfrogging concept as applied to AI. What conditions make leapfrogging genuinely beneficial? What conditions make "leapfrogging" language a cover for extractive commercial deployment?

Exercise 17: The Representation Problem in AI Research The chapter notes that the global AI research community is overwhelmingly concentrated in a small number of wealthy countries and dramatically underrepresents researchers from the Global South.

(a) Research the geographic distribution of publications at major AI conferences (NeurIPS, ICML, ICLR, ACL) in 2023 or 2024. What proportion of papers came from sub-Saharan African institutions? From South Asian institutions (excluding US-educated researchers at US institutions)? (b) What are the structural barriers that prevent Global South researchers from publishing at these conferences? (c) What interventions have been most effective at increasing Global South participation in AI research? (d) What would a genuinely equitable global AI research ecosystem look like, and what would it take to build one?

Exercise 18: Ethics Washing in Development AI The "ethical AI" and "impact sourcing" marketing around AI development work in the Global South has been characterized as ethics washing — using ethical language to market services that do not meet ethical standards.

Identify two companies (besides Sama Group) that market their AI services using similar ethical or development impact language. For each company: (a) What specific ethical or impact claims do they make? (b) What evidence is publicly available about working conditions for their data annotation workers? (c) Does the evidence support or undermine the company's ethical claims? (d) What would genuine ethical AI annotation work require that these companies' marketing does not demonstrate?

Exercise 19: The Sovereignty Dilemma African governments face a genuine dilemma in AI governance: building effective AI regulatory capacity requires resources, expertise, and time that many African governments lack, while the cost of waiting for that capacity to develop is that AI systems from foreign companies operate in their jurisdictions without adequate governance.

Several response strategies have been proposed: - Adopt EU-style comprehensive AI regulation (accepting that governance framework was developed without African input) - Develop indigenous AI governance frameworks from scratch (accepting governance gaps during the development period) - Join regional governance frameworks through the African Union (accepting that regional consensus may be slower and less demanding than national regulation) - Participate actively in global governance processes (accepting limited influence given resource constraints)

Write a policy memo analyzing these options and recommending an approach for a specific African country of your choice.

Exercise 20: Responsible Technology Partnership Design Design a responsible technology partnership agreement between a major AI company (choose one: Google, Microsoft, Amazon, or Meta) and a government in sub-Saharan Africa for deployment of AI infrastructure.

Your partnership design should specify: (a) Data governance — who owns data generated, where is it stored, who has access? (b) Benefit-sharing — what economic benefits flow to the partner country and its communities? (c) Local capacity building — what training, employment, and institutional development commitments are required? (d) Governance — what oversight mechanisms exist over the partnership and the AI systems deployed? (e) Accountability — what remedies exist if the partnership produces harm? (f) Sovereignty protection — what provisions ensure that the partner country retains meaningful control over its digital infrastructure?

Scenario and Simulation Exercises

Exercise 21: Global South Perspective Simulation Working in groups of four to five, simulate a session of a Global Partnership on AI working group where representatives from Global South countries are presenting their AI governance priorities.

Assign the following roles: - Representative from Nigeria (AI development hub, fintech sector) - Representative from Ethiopia (large population, significant infrastructure gaps) - Representative from Kenya (advanced mobile money ecosystem, annotation labor concerns) - Representative from GPAI member from a wealthy country (US or EU) - Civil society representative from a Global South digital rights organization

Simulate a 30-minute discussion about the GPAI's research agenda priorities for the next year. What tensions emerge? What compromises are reached?

Exercise 22: AI Ethics in the Field — Community Engagement Simulation An AI company wants to deploy a healthcare AI diagnostic tool in rural Uganda. Before deployment, they are conducting a community engagement process in a village of 3,000 people, most of whom are subsistence farmers with limited formal education and no prior experience with AI.

Design and role-play a community engagement session that: - Explains what the AI system does in accessible, non-technical language - Identifies community members' concerns and priorities - Discusses data governance in terms the community can evaluate - Addresses questions about what happens if the system gives wrong medical advice - Explains the benefit-sharing arrangement honestly

Debrief on: What information was difficult to communicate? What concerns arose that you had not anticipated? What governance commitments would the community require?

Exercise 23: Supply Chain Transparency Audit Working in groups, conduct a supply chain transparency audit for a major AI company's AI annotation labor practices. Use publicly available sources: annual reports, sustainability reports, public press coverage, academic research, NGO investigations, and the company's public statements.

Your audit should assess: (a) What is publicly known about the company's annotation labor practices? (b) What information is notably absent from public disclosure? (c) How does the company's transparency about annotation labor compare to its transparency about other supply chain issues? (d) What specific disclosures would constitute adequate supply chain transparency for AI annotation labor? (e) Draft a public statement that the company should be able to make, and identify the practices that would need to change for the statement to be true.

Exercise 24: The Infrastructure-Inclusive AI Design Challenge Working in groups, you are asked to redesign a healthcare AI diagnostic tool (initially designed for well-resourced US hospitals) for deployment in rural sub-Saharan Africa. The original system requires: high-speed internet connectivity; a tablet or laptop for the clinical interface; cloud-based inference; regular model updates via internet; and EMR integration.

Your redesigned system must function in: - Facilities with intermittent electricity (4 hours per day on average) - Facilities with no reliable internet connectivity (occasional 2G access) - Contexts where the primary user is a community health worker with secondary school education - Local languages (Swahili, Amharic, or a language of your choice)

What design trade-offs are necessary? What performance trade-offs are acceptable? What does redesigning for this context reveal about the assumptions built into the original design?

Exercise 25: The Data Colonialism Negotiation A global AI company wants to build a comprehensive agricultural AI model for East Africa. To do so, it needs training data from smallholder farmers — images of their crops, yield records, soil test results, and geographic coordinates. The company is prepared to offer farmers free access to the resulting AI tool in exchange for their data.

Negotiate a data governance agreement that protects farmers' interests. On behalf of a coalition of smallholder farmer cooperatives: (a) What rights over the data must be retained? (b) What limitations on the company's use of the data are required? (c) What ongoing benefit-sharing provisions are needed? (d) What governance mechanisms ensure compliance? (e) What exit rights — including data deletion — must be preserved?

The company representative should argue for a simpler, more commercially flexible arrangement. Negotiate toward a compromise that both parties can defend publicly.