As one of the most transformative technological advancements, artificial intelligence (AI) has the potential to improve the lives of millions of people around the world.
AI is already making a difference through diverse applications: the Rwandan company Zipline uses AI for advanced route planning to deliver blood and medical supplies autonomously through drones to hospitals. The Brazilian platform Traive uses AI to determine a more accurate credit score to help farmers get a fair credit rating.
However, AI development today is focused in the Global North, particularly in the U.S. Since AI models are always exposed to the assumptions and beliefs of their creators, this centralization leads to biases that are present in AI applications. Localizing AIcan ensure that benefits of AI are distributed more equitably across the globe and result in AI applications that are more context-specific and therefore more effective.
If we want localize AI systems in the Global South, we first need to understand which elements determine the success or failure of localized AI systems. Here, a systems map can be a useful tool to better understand key elements and actors involved in a system as well as the dynamics that unfold therein.
Key Elements of Local AI Systems
While each local AI system will be context-specific, the preliminary system map below provides initial insights into generic elements that can influence local AI systems.
Figure 1: Generic elements that influence local AI systems
Our preliminary research shows that four of these elements seem to be especially relevant for localizing AI: investments, infrastructure, computational resources and policies and regulations.
Investments: Training AI models is prohibitively expensive. Smaller companies in Low- and Middle-Income Countries (LMICs) often struggle to secure funding for computational resources and talent. Investing in AI companies in LMICs is increasing, and leveraging open-source datasets and local, community-owned pre-training models can drive local AI development, reduce costs and boost innovation for smaller actors.
Infrastructure: Infrastructure poses a challenge in handling massive datasets that can exceed several terabytes. Indeed, efficient data management throughout its lifecycle, from creation to deletion, is crucial. While data centers are today increasingly available in the Global South, ensuring adequate storage and network capabilities still remains a significant hurdle. Ideal data center locations require stable, renewable energy sources, minimal disaster risk, cool climates for energy efficiency, favorable markets and political stability. Achieving this balanced infrastructure is particularly challenging in LMICs.
Computational Resources: Access to sufficient computational power is a challenge for businesses venturing into AI innovation due to cost-intensive hardware like GPUs and TPUs. Systems using machine learning are often asymmetrical in their use of resources, demanding substantial computational power for training but less for actual usage. These high upfront costs often result in the exclusion of LMICs from AI development. By sharing computational resources across organizations or teams in different phases of an AI application lifecycle, the barriers to entry can be lowered while better utilizing resources.
Policies and Regulations: To ensure the safe and ethical development and use of AI, governments must regulate the field. Clear, transparent regulations level the playing field for AI research and development, fostering innovation and competition. Some countries in the Global South such as Colombia and Vietnam have a national AI strategy in place to shape the development of AI. Nigeria and Rwanda are working on similar strategies to take ownership of AI in their countries.
Getting here
Once you have a good understanding of the local AI system, you can look for leverage points and possible intervention areas. Where can you create the most impact with the least amount of effort and resources? Where can you catalyze change by altering system dynamics to drive the localization of AI?
With our Inclusive Innovation 2030 (ii2030) process, Endeva has developed a format that effectively maps complex and interdependent systems, identifies key levers for change and prototypes systems innovations. We do this by involving the right set of actors, including those who are directly impacted, from the start in the discovery, design and co-creation of prototype solutions that enable systems change. Since 2017, Endeva has successfully implemented over twenty ii2030s in LMICs, especially around tech-based opportunities that have facilitated breakthroughs in achieving the SDGs.
As one of the most transformative technological advancements, AI has the potential to significantly contribute to achieving the SDGs. However, to realize this potential, it is crucial to ensure equal participation from all. Impactful, action-oriented collaboration formats such as endeva’s ii2030 can lead the way in harnessing the full power of AI to create a better future for all.
Conclusion
Localizing AI is not just a call for fairness; it’s a strategy for enhancing the effectiveness of AI solutions and ensuring that they are tailored to the unique challenges faced by different regions. It’s about making AI a force for good that benefits all of humanity, regardless of where we live. As we navigate the AI revolution, it is important to remember that the unfolded power of AI lies in its ability to bring positive change to the lives of people everywhere in the world.
For those who wish to dive deeper into the topic, a working paper titled “Localizing Artificial Intelligence – A call for more effective AI solutions,” authored by Stefan Dehm & Christian Pirzer is available on our website.
This post has been written by Christian Pirzer, Managing Director at Endeva and Samantha Beekman, Analyst at Endeva .
