Louis Mondo
AI for Transformation
As Artificial Intelligence rapidly reshapes education, business, governance and economic development, a new form of inequality is emerging: the AI divide. This divide separates those who can access, understand and benefit from AI from those excluded by high technology costs, limited computing resources, inadequate connectivity and insufficient digital skills.
For Zimbabwe, bridging the AI divide is becoming as important as bridging the traditional digital divide. While expanding internet connectivity remains essential, the next frontier is ensuring that every citizen, regardless of geographical location or economic circumstances, can meaningfully participate in the AI revolution. Small Language Models (SLMs) offer a promising, affordable and practical pathway towards achieving this ambition.
Much of the global conversation about artificial intelligence has centred on powerful Large Language Models (LLMs), including those behind platforms such as ChatGPT and Gemini. Although these systems demonstrate remarkable capabilities, deploying large models independently can require substantial computing infrastructure, technical expertise and financial resources. For developing economies, these requirements can create barriers to widespread AI adoption.
Small Language Models present a different approach. These are compact artificial intelligence systems designed to perform useful language-related tasks with significantly fewer computational resources. Their greatest strength lies in making AI capabilities accessible through ordinary computing devices rather than depending exclusively on expensive data centres and cloud-based infrastructure.
Remarkably, appropriately optimised Small Language Models can be installed, hosted and operated directly on affordable smartphones, tablets and laptops. Depending on the model and device specifications, these systems can generate text, summarise documents, answer questions, translate information, support learning and process locally stored knowledge without requiring continuous internet connectivity.
This represents a fundamental shift in how artificial intelligence can be delivered. A low-cost tablet or laptop can become a self-contained AI platform, bringing useful intelligence directly to individuals and communities. Instead of sending every request to a remote server, the device processes information locally, reducing recurring data costs, dependence on connectivity and reliance on expensive cloud subscriptions.
For Zimbabwe, this offers an opportunity to democratise artificial intelligence through a frugal approach that prioritises affordability, accessibility and practical value.
Zimbabwe deserves recognition for its ongoing efforts to bridge the rural-urban digital divide. Investments in telecommunications infrastructure, expansion of mobile network coverage and the establishment of Community Information Centres have contributed to extending digital access beyond major urban centres. These initiatives demonstrate the importance of connectivity in advancing national development and creating opportunities for underserved communities.
However, connectivity alone does not guarantee meaningful participation in the AI economy. Citizens also require affordable devices, relevant applications and the skills to use artificial intelligence productively. Without these complementary capabilities, the digital divide risks evolving into an even wider AI divide.
Small Language Models offer a practical way to build upon Zimbabwe’s connectivity investments. Rather than replacing the need for internet infrastructure, locally deployed AI can maximise its value by enabling useful applications to operate even when connectivity is intermittent or expensive. Internet access can then be used for downloading models, updating information and accessing services that require external connectivity.
The technological advances making this possible are particularly significant. Techniques such as model quantisation, compression and knowledge distillation have enabled developers to reduce the memory and processing requirements of AI systems while retaining useful capabilities.
Quantisation, for instance, reduces the precision used to represent a model’s numerical parameters, allowing it to occupy less memory and operate more efficiently. Consequently, selected models that previously required powerful computing equipment can now run on consumer devices with relatively modest specifications.
This development challenges the assumption that meaningful AI adoption must begin with expensive infrastructure. While large models remain important for complex applications, smaller models demonstrate that appropriately designed AI can deliver substantial value at a fraction of the computational cost. The objective is not to replace larger AI systems, but to adopt a balanced approach that uses the most appropriate and affordable technology for each application.
The power of Small Language Models also lies in their adaptability. Rather than attempting to perform every conceivable task, smaller models can be configured for specific purposes, allowing institutions and developers to focus on applications that address clearly defined needs.
Through approaches such as fine-tuning and retrieval-augmented generation, these models can be adapted to work with specialised documents, educational materials, institutional knowledge and locally relevant information. This makes them particularly attractive for resource-constrained environments where efficiency and relevance are more important than model size.
For Zimbabwe’s education sector, this presents an opportunity to accelerate inclusive AI literacy within the Education 5.0 framework. Schools, universities, polytechnics and vocational training institutions could establish practical AI learning environments using affordable laptops, tablets and existing computers.
Learners would not need to depend exclusively on expensive international AI subscriptions to gain practical exposure to artificial intelligence. Locally hosted models could support experimentation, knowledge discovery and the development of AI-related competencies.
Importantly, AI literacy must extend beyond the ability to generate answers using intelligent systems. It should encompass understanding how AI works, evaluating the reliability of its outputs, recognising limitations and using the technology ethically and productively.
Small Language Models also contribute to the advancement of Zimbabwe’s indigenous languages. International AI systems often perform less effectively in languages with limited representation in global training datasets.
Developing locally relevant language resources and adapting smaller models could progressively strengthen AI capabilities in Zimbabwe’s indigenous languages. This would make AI more accessible while supporting the preservation and digital development of Zimbabwe’s linguistic and cultural heritage.
Zimbabwean software developers could leverage efficient, openly available models to build locally relevant applications, stimulating domestic innovation and reducing dependence on imported technology solutions.
These opportunities align with Zimbabwe’s Vision 2030 aspirations, the Smart Zimbabwe 2030 Master Plan and the National Artificial Intelligence Strategy 2026 – 2030, particularly the pursuit of inclusive national transformation through heritage-based innovation and human capital development.
Zimbabwe could advance this opportunity through a national frugal AI innovation initiative focused on resource-efficient models, affordable computing and locally relevant applications. Universities, research institutions, technology companies and innovation hubs could collaborate to evaluate suitable models, develop indigenous language datasets and establish practical AI learning programmes.
Existing Community Information Centres could progressively evolve into AI literacy hubs, allowing citizens to experience artificial intelligence through affordable devices and locally hosted applications. Such an approach would build upon existing national investments rather than require entirely new infrastructure.
Ultimately, Zimbabwe’s progress in artificial intelligence should not be measured solely by the sophistication of its systems or the size of its computing infrastructure. It should also be measured by how widely AI capabilities are accessible, affordable and beneficial to society.
Zimbabwe’s continued investment in rural connectivity is laying an important foundation for digital inclusion. Small Language Models offer a frugal pathway for transforming that connectivity into meaningful AI participation.
By combining expanding rural connectivity, affordable smartphones, tablets and laptops, and locally adapted Small Language Models, Zimbabwe can move beyond bridging the digital divide towards closing the AI divide, ensuring that artificial intelligence becomes a shared instrument for inclusive national development.
Louis Mondo is a Digital Transformation and AI Governance Expert
Email: [email protected]