Jua Kali AI: Bottom-up algorithms for a Bottom-up economy

As artificial intelligence (AI) becomes a cornerstone of the global economy, AI’s foundations must be anchored in community-driven data, knowledge, and wisdom. ‘Bottom-up AI’ should grow from the grassroots of society in sustainable, transparent, and inclusive ways. 

Kenya, with its innovative bottom-up economic strategy, could play a pioneering role in new AI developments. Bottom-up AI could give farmers, traders, teachers, and the local and business communities the power to use and protect AI systems that contain their knowledge and skills that have been honed over generations. 

Kenya’s digital landscape is ripe for such innovation. It is home to a dynamic tech community. It has been the cradle of numerous technological breakthroughs, such as the widely known Mpesa mobile payment service and the Ushahidi crowdsourcing platforms. 

However, there is a prevailing notion, fuelled by media narratives, that AI development is the exclusive domain of big data, massive investments, and powerful computational centres. Is it possible for Kenya to circumvent these behemoths using its indigenous ‘Jua Kali’—an informal, resourceful approach—to cultivate Bottom-Up AI?

The answer is a resounding yes, as exemplified by the advent of open-source platforms and the strategic utilisation of small but high-quality datasets. 

Jua Kali micro-enterprises

Open-source Platforms: The pillars of Bottom-up AI

Open-source AI platforms are challenging the dominant paradigm that AI necessitates colossal AI systems  – as leveraged by prominent language models like ChatGPT and Bard. A purported internal document from Google candidly acknowledges this competitive edge: “They are doing things with $100 and 13 billion parameters that we struggle with at $10 million and 540 billion parameters. And they are doing so in weeks, not months.”

Names like Vicuna, Alpaca, LLama, and Falcon now appear alongside ChatGPT and Bard, demonstrating that open-source platforms can deliver comparable performance without extravagant costs. Moreover, they tend to be more adaptable and environmentally friendly – requiring significantly less energy for data processing.

Small and High-quality Data: The key resource of Bottom-up AI

As open-source algorithms become more accessible, the emphasis of bottom-up AI naturally shifts to data quality, which depends on data labelling, a human-intensive activity. A lot of data labelling for Chat GPT has been done in Kenya, which triggered numerous labour criticisms.

Alternative approaches are feasible. As a matter of fact, at Diplo we have pioneered integrating data labelling into our regular activities, from research to training to project development. This is akin to using digital highlighters and sticky notes within our interactive frameworks, thus organically fostering Bottom-Up AI.

Text is not the sole medium for knowledge codification. We can also digitally annotate videos and voice recordings. Imagine farmers sharing their insights on agriculture and market strategies through narratives, enhancing the AI’s knowledge base with lived experiences.

Beyond Technology: Embracing organisational and societal shifts

The primary hurdle for Bottom-up AI is not technological but organisational and revolves around societal and policy priorities. Building on its digital dynamism, Kenya has the potential to lead by marrying technological advances with practical, citizen-focused applications.

Kenya’s bottom-up AI could contribute to preserving our knowledge and wisdom as a global public good, which we should pass on to future generations as the common heritage of humanity.