Algorithmic Tokenism vs. Digital Marema-Tlou: Decolonizing AI in Evaluation
Seminario web | En línea
-
Organizado por:
Mphatlalatsane Management Services
Sobre el evento
As the global M&E sector accelerates its adoption of Artificial Intelligence, a critical ethical blind spot is emerging: the linguistic and cultural bias of Large Language Models. Because mainstream AI is predominantly trained on Western datasets and Eurocentric epistemologies, deploying these tools in the Global South risks ushering in a new era of "algorithmic tokenism." Without rigorous safeguards, AI will misinterpret local realities, sanitize grassroots trauma, and erase the "Quiet Knowledge" embedded in indigenous communities.
This session challenges evaluators to look beyond Silicon Valley defaults and apply the Sesotho philosophy of Marema-Tlou (collective wisdom) to digital M&E infrastructure. Drawing on decolonial frameworks, the session will explore how evaluators can avoid digital extraction by co-creating localized AI prompts and guardrails directly with communities.
This session challenges evaluators to look beyond Silicon Valley defaults and apply the Sesotho philosophy of Marema-Tlou (collective wisdom) to digital M&E infrastructure. Drawing on decolonial frameworks, the session will explore how evaluators can avoid digital extraction by co-creating localized AI prompts and guardrails directly with communities.
Presentador/a
| Nombre | Título | Biografía |
|---|---|---|
| Thabiso Lakajoe | Mr. | Thabiso Lakajoe is an M&E executive, Wits School of Governance Master's candidate, and founder of Mphatlalatsane. With 15+ years of experience, he champions decolonial evaluation, merging indigenous knowledge systems with global AI and M&E standards to drive structural accountability and impact. |
Resumen
Thabiso discussed the importance of ensuring that evaluation evidence serves communities rather than just institutions or donors, emphasizing the need for human wisdom to collaborate with responsible AI in evaluation processes. When asked by Ana about validating interpretations, Thabiso suggested involving community members and beneficiaries through stakeholder workshops to validate data interpretation before finalizing evaluation reports. Thabiso also acknowledged Ana's observation about variations in stakeholder inclusion depending on who commissioned the evaluation, explaining that meaningful evaluations require involvement from the beginning to end and should be understandable to the communities from which data was collected. Thabiso discussed the challenges of using AI in community evaluations, highlighting how AI cannot capture complex social issues like domestic violence or neglect without proper context and community input. He explained the concept of algorithmic tokenism, where community voices become data points rather than being meaningfully included in evaluations.
Ana
• Incorporate final reflection or workshop sessions at the end of evaluations to allow stakeholders and community members to review and validate evaluation reports before finalization.
• Raise awareness and, where possible, influence commissioners of evaluations to support the inclusion of stakeholders in the validation of evaluation findings and reports, especially when organizational resources or preferences may limit participation.
Thabiso
• Develop and promote guidelines for evaluators to ensure evaluation questions are workshopped, piloted, and validated with community members, beneficiaries, implementing partners, and program staff before evaluations commence.
• Advocate for and implement feedback loops in evaluation processes, ensuring that data, analysis, and reports are shared with and validated by the communities from which data was collected.
Collaboration
• All participants (Thabiso, Ana, Manuela, Vuyiswa): Before using AI tools in evaluations, proactively assess whose voices, languages, and knowledge may be missing, and take steps to ensure linguistic and cultural inclusion in data collection and interpretation.
• All participants (Thabiso, Ana, Manuela): Encourage the use of Digital MaremaTlou principles, listening before measuring, community validation, linguistic inclusion, context preservation, shared interpretation, and ethical knowledge stewardship in the design and implementation of AI-enabled evaluations.