The IDEMS Podcast

Continuing their examination of the assumptions underlying today’s dominant AI narrative, David and Kate explore what makes data useful, trustworthy, and meaningful. They discuss the limitations of extraction-based approaches to AI, the importance of local context and data ownership, and the challenges of building systems that can learn across diverse communities without centralising control. The conversation highlights why better data—not just more data—may be key to building more effective and trustworthy AI systems.

Continuing their examination of the assumptions underlying today’s dominant AI narrative, David and Kate explore the distinction between AI as a product and AI as a sociotechnical system. They reflect on the often-invisible infrastructure, labour, resources, and governance structures that sit behind AI technologies, and discuss why understanding these systems is essential for making informed choices about technology, impact, and innovation. The conversation highlights how different assumptions about ownership, trust, and accountability shape the technologies we build and the societies they serve.

Continuing their discussion on the future of AI, David and Kate explore the economic and institutional forces shaping today’s dominant AI models. They discuss the roles of investment, monopoly power, research funding, and commercial incentives in driving ever-larger AI systems, and consider how these pressures influence both technological development and public narratives around AI. The conversation highlights why the current trajectory of AI is not inevitable and what alternative paths might look like.

Continuing their discussion on the future of AI, David and Kate explore how advances in large language models could enable a new generation of smaller, more specialised AI systems. They discuss why the next wave of innovation may come from building tools that are more efficient, focused, and responsive to real-world needs rather than simply pursuing ever-larger models.

David and Kate explore the historical divide between Symbolist and Connectionist approaches to AI, reflecting on how today’s dominant AI narratives emerged and what may have been lost along the way. They discuss the difference between expert systems built on structured human knowledge and data-driven learning systems based on neural networks, and consider the implications of each for governance, traceability, social impact, and responsible technology development. The conversation highlights how alternative approaches to AI may offer more practical and trustworthy pathways for addressing real-world challenges.

In the second part of their discussion, David and Kate reflect more deeply on the Earthkeepers versus AI Empires convening in Zambia, exploring the diverse perspectives and tensions that emerged during the event. They discuss questions of power, governance, indigenous knowledge, and technological futures, as well as the growing recognition that current AI trajectories are not inevitable. The conversation highlights alternative visions for AI and digital technologies built around community ownership, trusted data, local governance, and smaller-scale systems designed to serve real social needs rather than concentrated power.

In the first of a two-part discussion, David and Kate reflect on a recent convening in Zambia that brought together activists, technologists, researchers, and civil society groups concerned with the impacts of AI infrastructure and large-scale data centres. They discuss the influence of Karen Hao’s book Empire of AI, the emergence of global resistance movements around extractive AI development, and the distinction between AI as a useful tool and the broader systems of power shaping its deployment. The conversation highlights growing concerns around the resource demands and extractive dynamics associated with large-scale AI infrastructure.

Lily and David discuss the challenges of working with rainfall and climate data, exploring ideas of data quality, data rescue, and data accreditation. They reflect on different sources of climate data—from weather stations and satellites to reanalysis products—and examine how these can be evaluated for specific applications such as agriculture. The conversation also highlights ongoing research into rainfall intensity, satellite validation, and the importance of building evidence around which climate products are appropriate for different contexts and uses.

Lucie and David continue their discussion on Farmer Research Networks (FRNs), focusing on the idea of embedded scaling and its implications. They explore how scaling out, scaling up, and scaling deep each change the nature of the data and the research itself, and reflect on the challenge of designing systems where farmers collect and use data for their own benefit while also contributing to wider learning and research.

Lucie and David discuss the origins and evolution of Farmer Research Networks (FRNs) within the work of the Global Collaboration for Resilient Food Systems. They explore how FRNs were conceived as a way to combine participatory research with large-scale data and reflect on ongoing debates around embedded scaling, participation, and the distinction between FRNs and approaches such as Participatory Action Research (PAR). The conversation highlights both the promise and the practical challenges of building research systems that are deeply contextual while capable of generating broader learning and impact.

Lily and David discuss recent work on the ePICSA system, focusing on the development of a structured summaries database to support climate information for agriculture. They explore how moving from file-based systems to a database approach creates new opportunities for versioning, quality control, decentralised workflows, and accreditation of climate products. The conversation also reflects on the broader challenges of climate data quality, data rescue, and building sustainable systems that can support national meteorological services.

Santiago and David discuss how a specialised AI assistant is supporting their work authoring and reviewing STACK questions. They explore the balance between human expertise and AI support, reflecting on how the assistant improves efficiency, enhances question quality, and helps navigate complex documentation and legacy code. The conversation also highlights the broader potential of specialised AI assistants as collaborative tools that augment, rather than replace, expert human work.

Michele and David explore the ideas and design principles behind the Open App Builder, a system developed through IDEMS’ collaboration with Parenting for Lifelong Health (PLH). They discuss the challenges of creating flexible, reusable app infrastructures that support collaboration across technical and non-technical teams, and reflect on the long-term vision of enabling local organisations to build and adapt their own digital tools. The conversation highlights both the complexity of the system and the growing opportunities emerging as it matures and expands into new contexts.

Michele and David discuss the development of facilitator apps within the Parenting for Lifelong Health (PLH) ecosystem, exploring how these tools support facilitators delivering parenting programmes in diverse contexts. They reflect on the growing role of adaptable digital tools for reporting, monitoring, and programme delivery, and discuss the wider opportunities these technologies create for scalable implementations, embedded research, and locally owned digital ecosystems that could support many different types of interventions in the future.

Santiago and David discuss the emergence of STACK question authoring as a growing area of work within IDEMS. They reflect on the value of cross-institution collaboration, the role of formative assessment and feedback in mathematics education, and how expertise developed across diverse educational contexts can support universities internationally.

David and Digital Green CEO Rikin Gandhi discuss the intersection of farmer research networks, participatory agricultural research, and AI-enabled extension systems. They explore how tools like Farmer Chat could support large-scale, farmer-led experimentation by combining rich qualitative data with rigorous research design. The conversation highlights the potential for more collaborative, context-sensitive agricultural systems that place farmer agency at the centre of both research and technology development.

Lucie and David discuss the upcoming Pan-African workshop on community-owned digital technologies, exploring its origins in a broader call for more locally grounded tech development. They reflect on the challenges of bringing together diverse participants—from developers to community organisations—and the importance of designing technology that genuinely serves community needs.

Lily and David discuss the AIMS Rwanda doctoral course on problem solving in data science, reflecting on how participants engage with complex simulated datasets. They explore the challenges students face in identifying underlying models, the limitations of relying on AI tools, and the importance of interpretation and human insight in data analysis. The conversation highlights how the course evolves alongside new technologies while continuing to reveal fundamental lessons about working with data.

Lucie and David reflect on key moments from the 20th anniversary celebration of the West Africa Community of Practice (COP), focusing on the activities and discussions that shaped the week. They explore the innovation fair, engagement with policymakers, and participatory sessions reflecting on the past and future of the COP.

Lucie and David reflect on the 20th anniversary celebration of the West Africa Community of Practice (COP) in Niger, focusing on the broader context in which it took place. They discuss the practical and geopolitical challenges of working in the region, the importance of in-person engagement for building trust, and the contrasts between local and international ways of working. The conversation highlights the resilience of the COP and the deeper insights gained from engaging with complex, real-world environments.

Santiago and David discuss the organisation of the 2026 International STACK Conference, marking the first time the event is held outside Europe. They reflect on its expanded format, emphasis on workshops and inclusivity, and the growing global reach of the STACK community. The conversation also highlights capacity building, new collaborations, and the potential for the conference to act as a catalyst for future developments.

George and David explore how the PBDM modelling approach could extend beyond agroecology into public health and epidemiology. They discuss similarities and differences between modelling ecosystems and disease systems, the potential for compositional frameworks to connect models across scales, and the broader opportunity for interdisciplinary collaboration. The conversation highlights both the conceptual challenges and the long-term vision of using more holistic models to inform better public health decisions.

George and David discuss the next phase of work on Physiologically Based Demographic Modelling (PBDM), focusing on efforts to scale its application in agroecological systems in West Africa. They explore the challenges of building and deploying complex ecosystem models, the interdisciplinary collaboration required, and the long-term vision of integrating deterministic models with responsible AI to support decision-making, from policy to smallholder farmers.

Santiago and Kate explore the thinking behind organising the IDEMS podcast as it grows into a substantial body of work. They discuss the development of a microsite, including tagging and filtering systems to help listeners navigate diverse topics, as well as ideas like curated collections and favourite episodes. The conversation reflects on how to balance structure and openness, making the podcast more accessible while preserving its exploratory nature.

Santiago and David explore the gap in open GCSE and iGCSE textbooks, and the opportunity to build curriculum-aligned resources from existing open content. They discuss how tools like PreTeXt can enable “same content, different variants” across exam boards, with interactive technologies like STACK adding further value, and reflect on how current work is putting the structures in place for scalable, adaptable textbooks.

Lily and David continue their discussions on converting open textbooks into PreTeXt. They focus on the “Learning Statistics with …” ecosystem, where an original open book has spawned variants for R, JASP, Jamovi, CogStat, French, and potential new versions such as R-Instat. They explore how PreTeXt could better manage multiple independently maintained variants by identifying what differs, easing updates from a base text, and supporting responsible human ownership.

Lucie and David discuss the tension between two of IDEMS’ principles: being “collaborative by nature” and “enabling opportunity” for individuals. They contrast rising Western individualism with more community-focused African contexts where personal sacrifice can support collective coherence. They argue collaboration and individual success aren’t contradictory, citing Italian cooperatives, worker-owned factories, and the importance of compromise and recognising different needs to avoid extractive relationships. Can the same thinking be applied to technologies like AI?

Michele and David discuss humanoid robots, and argue the real question isn’t when they arrive, but how different societies will accept and use them. They push back on framing humanoids mainly as worker replacements, suggesting their biggest impact may be social—especially in contexts like Italy’s aging population and shrinking workforce, where care needs are growing. They explore whether “humanoid” matters at all versus simply being multipurpose and designed to complement human carers rather than replace them.

When does measuring the “wrong” thing produce better results than measuring the “right” one? Lily and David continue the mini-series on Research Methods Support for Climate Resilient Food Systems with a story about proxy variables from West African farmer trials: measuring millet head weight at harvest proved a more reliable proxy for grain yield than weighing separated grain later.

Learning doesn’t come through passive consumption of information, but through doing. In this episode, Lily and David discuss the integration of STACK exercises into the PreTeXt textbooks that Lily has been working on – take a look at some previous episodes for more details. What if open textbooks could share the same bank of deep, automated-feedback assessments across multiple versions and courses?

What if educational resources could be available in a limitless variety of variants, each adapted to the tools students actually use? In this episode, David talks to Lily about a project she has been working on to convert open statistics and data science textbooks into the PreTeXt format. The discussion highlights why PreTeXt’s semantic structure and separation of authoring from publishing enable systematic changes across a book, supporting making software-specific and software-agnostic variants to tailor the books to various contexts where different approaches are more valuable.

Following previous episodes on Kenyan maths textbooks, Santiago and David reflect on the project from IDEMS’ perspective. They discuss the rapid push to complete open Grade 10 mathematics resources, driven by urgent teacher needs under the new curriculum and growing interest from the Kenyan Ministry of Education and CEMASTEA. They outline the core tools: a PreTeXt textbook designed for multiple variants, minimal STACK integration for mastery-focused interactive questions and feedback, and Moodle courses that combine short teacher training with learning-objective-based topic courses and forums for peer exchange and certification.

How do textbooks function differently across educational systems, and what happens when those systems operate in low-resource environments? Following on from previous episodes on the CBC open textbooks project in Kenya, Lucie Hazelgrove-Planel and David Stern discuss textbooks more generally. They discuss the different roles textbooks play in different educational contexts, and how a lack of contextualisation can block learning. They consider how an open textbook model, where they can be adapted into different variants without requiring an entirely new publication, could revolutionise the way textbooks are produced and used globally.

Students are increasingly turning to LLMs (Large Language Models) to solve maths exercises and get feedback. In light of this, is there still a place for deterministic online assessment tools like STACK? In this episode, Michele and David argue that this problem is an opportunity for educators and developers to build better alternatives, potentially embedding generative AI features in STACK to provide a more interactive, conversational experience. They consider more generally how LLMs affect exams, curriculum decisions, and student motivation, noting uncertainty about future skills and warning against reactive curriculum changes while encouraging experimentation.

David interviews Michael Obiero on the challenges of implementing Kenya's new Grade 10 Competency-Based Curriculum, such as resource shortages and unprepared teachers. To address these issues, they are developing digital resources, online courses, and a teacher support platform. Their grassroots initiative aims to create community-driven, context-sensitive solutions with the help of volunteers and limited funding. By fostering peer-to-peer learning and collaboration among teachers, they seek to build a sustainable model for adapting educational resources to various contexts and improving math education outcomes.

Lily and David discuss the significant influence of data collectors on survey variability and data quality, using examples from West Africa. They highlight the importance of thorough enumerator training to address issues like inconsistent definitions of household size.

In this episode Lily and David delve into the ANOVA (Analysis of Variance) tables, focusing on the sum of squares. They discuss how it helps account for data variability, and the difference between sum of squares and mean squares. The episode also touches on the limitations of p-values and emphasizes the ANOVA table's value as a descriptive tool, particularly in enhancing research methods in West Africa.

Lily and David discuss the application of ANOVA in agroecology research, focusing on its historical roots and its use as a descriptive tool. They emphasize the importance of understanding the degrees of freedom in the ANOVA table, highlighting its impact on effective data analysis and model fitting. This episode is part of the celebration of 20 years of research support in the region, showcasing the value of statistical methods in enhancing research outcomes.

Lily talks with Roger about an agricultural experiment in West Africa that revealed the impact of termite mounds on crop yield data. The discussion focuses on handling unexplained variability and the importance of recognizing outliers. Roger explains the necessity of removing certain plots to reduce data variability and achieve clearer results.

In this episode, Lucie and David discuss the complexities of sampling in research. They explore common misconceptions, and introduce three levels of sampling complexity. The episode highlights the necessity of understanding population structure and the compromises involved in effective sampling.