(Part B: Modeling Human Behavior to Solve the Clean Cooking Crisis)
Editor’s Note (October 6, 2025). This post has been edited to align more tightly with Part A’s framing (governing world-model-class simulations) and sets up Part C’s integrated model platform. A few minor connective edits clarify the Afram Plains context, refine terminology (adoption dynamics, decision support under uncertainty), and add bridge sentences to Part C. The core analysis and recommendations are unchanged.
Part A argued for advanced, high-granularity AI models to manage the physical complexity and investment risk associated with decentralized energy infrastructure. Our existing toolkits need to be expanded beyond advanced algorithms and hardware to encompass a broader range of issues. As such, they can’t help us much in uncovering the more subtle barriers to energy transitions. Some of these are rooted in societal values, preferences, and habits that influence energy choices — acquired over thousands, if not hundreds of thousands, of years.
This second installment (Part B) sets the context for the discussion by focusing on one of Africa’s most persistent energy-development challenges: the lack of access to clean cooking solutions. The problem extends beyond techno-economic assessments of improved cookstoves. It is also very much about the intangible factors mentioned above. How can we leverage AI to improve answers to questions like when, why, and how a sufficiently large number of households might decide to abandon polluting cooking fuels? Answers will require a deeper understanding of the complex, non-linear dynamics of human choice. Enter Task 2.
Policy Task 2: Simulating Policy Impact on Adoption of Clean Cooking Solutions
And what is the context of Task 2? As of my last check, about one billion people on the continent still rely on polluting fuels, primarily charcoal, wood, and dung, often used over open fires. Every year, approximately 720,000 people are killed by household air pollution in the region. Women and their children are the primary victims. A massive level of ecological drain is another consequence. In many parts of East Africa, the daily harvesting of wood fuel accounts for over 50% of total national emissions, also contributing to rapid deforestation.
I’ve heard that the annual funding required to achieve universal access to clean cooking in Africa is roughly $2 billion, totaling $37 billion by 2040. That’s good news if only the necessary knowledge-to-action pathways are free of the usual hurdles. Simply throwing more money at the problem will not work unless it is precisely targeted at producing and deploying high-quality social intelligence to inform decisions at all levels. In Africa, the clear and present threat to success is deeply rooted in specific cultural norms, combined with pervasive consumer ignorance of health risks. This amounts to a “wicked problem” – a tough-to-define, value-laden challenge with no definitive formulation or stopping rule. Wicked problems tend to be highly complicated, if not impossible to solve. You can’t really tell whether one “solution” is better or worse than another until you actually test it. The cost of making the wrong choice can be hefty. That makes it highly resistant to conventional linear modeling. We need better tools to tackle such wickedness in the struggle for universal access to clean cooking.
Modeling Agent Interaction and Action Inference in the Household SES
Traditional modeling breaks down when historical data on consumer behavior is scarce—an everyday reality in rural Africa. We cannot simply plug in past purchasing habits when the future requires radical shifts, such as adopting Results-Based Financing (RBF) or transitioning to bottled bioethanol.
This is where the TinyWorlds Action Tokenizer (briefly explained in Part A) proves indispensable. When we simulate a household or community as a “tiny world,” policy variables (such as a new subsidy, the price of LPG relative to charcoal, or a community health campaign) are set as environmental conditions. The Action Tokenizer then infers the resulting latent decisions, namely, the unobserved actions of the simulated agents (in this case, households) concerning fuel switching, technology purchase, and sustained usage.
This capability lets policymakers test the effectiveness of nuanced financial and behavioral policies before costly real-world deployment. Questions that can be addressed while avoiding the costs of real-world/pilot experimentation include:
- Can a microfinance scheme succeed if community trust is low?
- Will a public awareness campaign, run through local health practitioners, actually trigger sustained fuel switching?
By enabling decision-makers to observe the emergent, non-linear consequences of these policy inputs, the TWA approach gives users a better chance to convert qualitative social dynamics into quantifiable adoption rates, which are crucial for attracting private investment in clean cooking solutions.
Simulating Nexus Outcomes (Health and Deforestation)
A significant socio-economic benefit can accrue from a successful clean cooking intervention, which is readily observed across the broader energy-environment-health nexus. Because it models the physical properties and dynamics of the environment, TinuyWorlds can accurately trace the secondary consequences of behavioral shifts.
If the simulation confirms increased adoption of clean solutions, the Dynamics Model can immediately predict the cascading benefits, including measurable outcomes such as:
- Reduction in household indoor air pollution and the corresponding quantifiable decrease in pollution-related mortality and illness.
- Decreased demand for solid biomass (wood/charcoal) and the corresponding impact on deforestation rates and carbon emissions (a vital metric for supporting climate commitments).
Policy-makers are increasingly using institutional demand as a market lever; transforming energy profiles of schools and hospitals (which collectively burn millions of tons of wood annually) requires large-scale, coordinated policy. By simulating the entire social-environmental system, Task 2 provides the necessary granular evidence to design market-creating policies, sorely needed to link investment decisions directly to measurable climate and health benefits in Africa.
Next in the Series: In Part C, my final installment, I will address Task 3: Granular Scenario Analysis for Decentralized Energy Resources (DER) and Grid Convergence. It will examine how TWA models can better simulate the complex trade-offs between centralized grid expansion, including the rapid SME-led growth of the off-grid sector. Bottom-line claim: A TWA approach to tackling wicked problems is needed to close the gaps in our capacity to tackle the wicked problem of sustainable energy for sustainable development in Africa.
