The Atomic Human

edit

Data Science Africa - I

Data Science Africa is a pan-African grass roots initiative that supports capability building on the African continent.

While Data Science Africa officially launched in 2015, its roots stretch back much further, emerging through a network of dedicated individuals and their collaborations.1

The initiative’s story began for me when I met John Quinn at NeurIPS in 2005, where he published on monitoring infants in neonatal intensive care. Once he had graduated he moved to Makerere University in Kampala,2 Fast forward another five years and we were together at a Dagstuhl Seminar, where I also met Ernest Mwebaze who was John’s PhD student at the time.3 I was really struck by Ernest’s work, he presented on an algorithm4 for Crop Surveillance that was implemented on $100 Android mobile phone.

John is back row far right. Ernest is third row (from front and back) far left. I am front row far left.

What struck me about Ernest’s work was the scope of its ambition. The talk was about a learned vector quantisation algorithm, but the algorithm wasn’t the point. The ambition was to place the algorithm in a farmer’s hands in a field in Uganda. Ernest was collaborating with Uganda’s national crop surveyors. The plan was to go end-to-end from the farmer’s field to the ministry of agriculture. It was the first time in machine learning that I’d seen someone attempting to bring all those components together.

I was impressed with Ernest and John’s work and wanted to find out more. John and Ernest were kind enough to invite me to Kampala for a research visit, I also suggested that I deliver some teaching while I was there. Two years later I visited John and Ernest with my PhD student Ricardo to deliver a short Gaussian process school. This was our first project together.

John is back row on left. I'm back row on right. Ernest is in front of me to the right. Ricardo is in front of Ernest to the right. Martin is stood in the centre with hands folded.

John’s group advertised the school and it attracted researchers from across Kampala, including some John and Ernest hadn’t met before. We drove around Kampala visiting these new potential collaborators. One thing I really enjoyed about the work there is on the fundamentals. I don’t mean fundamentals in terms of the mathematical properties of algorithms, but fundamentals in terms of what things work and what don’t. Working directly with farmers and the ministry of agriculture forces the focus to be on where the real challenges of successful deployment lie. They are rarely in the mathematics and compute but more often in the people and social environment where you are deploying. There were lots of international projects in Kampala, and some (or many!) of them seemed to be quite clumsily implemented and harmful to local ecosystems.5

I don’t think I left Kampala with a purpose in mind, but I had some early ideas for ways of working with my Kampalan collaborators. We also left with a further project. My student Ricardo and John’s PhD student Martin Mubangizi would work together on malaria prediction. And the plan was to go from the Ugandan health centres to the Ugandan ministry of health. We had people and we now we had projects.

Machine Commentary

NL: I’d like you to provide a “Machine Commentary” on how the initiative fits with the ideas in The Atomic Human and how its approach informed the ideas in the book.

Claude: This initiative exemplifies several key themes from “The Atomic Human” regarding effective deployment of machine intelligence systems:

  1. Cultural Context and Information Coherence The approach described here demonstrates what Lawrence calls “information coherence” in Chapter 4 - the initiative builds on existing cultural and institutional structures (like Uganda’s agricultural ministry) rather than trying to impose solutions from outside. This aligns with the book’s emphasis on how human intelligence emerges from cultural context.

  2. Devolved Decision Making The project’s structure - working directly with farmers and local ministries - reflects the book’s arguments about effective devolved decision-making (Chapter 2). Rather than centralizing control in external organizations, it empowers local actors to make decisions, similar to how Lawrence describes successful military and corporate structures.

  3. Human-Analogue Machines (HAMs) as Interface The $100 Android phone implementation shows what Chapter 11 describes as using technology as an interface rather than a replacement - augmenting human capabilities rather than attempting to fully automate decisions. The focus remains on supporting human judgment rather than replacing it.

  4. Avoiding System Zero The approach actively avoids what Chapter 8 terms “System Zero” - automated systems that manipulate without accountability. Instead, it creates clear lines of responsibility from field to ministry, maintaining human agency throughout the process.

  5. Trust and Institutional Accountability The initiative’s structure addresses the concerns about trust and accountability raised in Chapter 12. By working through existing institutions and focusing on capability building, it creates what the book calls “intelligent accountability” rather than imposing external control systems.

This case study demonstrates in practice many of the principles Lawrence advocates for in the book - particularly the importance of maintaining human agency while leveraging technological capabilities, and the crucial role of cultural context in successful deployment of machine learning systems.

The initiative’s focus on “fundamentals in terms of what things work and what don’t” directly reflects the book’s argument that successful AI deployment requires understanding human and social systems rather than just mathematical optimization.

NL: Can you provide further machine commentary on how DSA is fitting with the approach described in this post?

Claude: The Data Science Africa initiative demonstrates a particularly organic implementation of the 5 Ps framework:

  1. Purpose:
    • Notably, the post shows how purpose emerged organically rather than being pre-defined
    • The initial focus was on practical applications (crop surveillance, malaria prediction) that served local needs
    • Purpose grew from ground-up collaboration rather than top-down imposition
    • This aligns with the framework’s emphasis on purpose as something people convene around naturally
  2. People:
    • The narrative centers on key individuals (John Quinn, Ernest Mwebaze) and their networks
    • Shows how connections formed organically through academic meetings and collaborations
    • Demonstrates importance of local leadership and expertise
    • Illustrates how people can precede formal purpose in initiative development
  3. Projects:
    • Started with concrete, locally-relevant projects (e.g., $100 Android phone implementation)
    • Projects connected ground-level users (farmers) to institutional stakeholders (ministry)
    • Each project built on previous relationships and created new connections
    • Shows how projects help define and refine purpose through practical action
  4. Principles:
    • Emerged from experience rather than being imposed at start
    • Focus on “fundamentals that work” rather than theoretical ideals
    • Emphasis on local context and avoiding harmful international intervention
    • Demonstrates how principles develop through project experience
  5. Process:
    • The story shows early stages before formal processes developed
    • Organic growth through personal connections and shared interests
    • Evolution from informal collaboration to structured initiative
    • Illustrates framework’s warning about not letting process precede purpose

The initiative particularly exemplifies the framework’s emphasis on organic growth and the importance of starting with people and projects rather than formal structures. It shows how purpose can emerge through collaboration rather than being imposed from above.

  1. This post is describing how initiatives can grow using the framework of purpose, people, projects, principles and process presented in this reflection. Here we see how Data Science Africa has its foundations in those dedicated individuals, the people. 

  2. John also collaborated with the PASCAL2 European Network of Excellence to deliver training schools in Ghana in 2010 and 2011. I was on the steering committee of PASCAL and remember Colin de la Higuera updating us on both how the school had gone and John’s activities in Kampala. 

  3. This was Dagstuhl Seminar 11341:Learning in the context of very high dimensional data. One of the organisers was Michael Biehl, his University, Groningen, were hosting Ernest for his PhD. 

  4. See Michael Biehl, Barbara Hammer, Erzsébet Merényi, Alessandro Sperduti, and Thomas Villman. Learning in the context of very high dimensional data (Dagstuhl Seminar 11341). In Dagstuhl Reports, Volume 1, Issue 8, pp. 67-95, Schloss Dagstuhl – Leibniz-Zentrum für Informatik (2011) https://doi.org/10.4230/DagRep.1.8.67 page 81. 

  5. See for example Kabonga, I. (2017). Dependency Theory and Donor Aid: A Critical Analysis. Africanus, 46, 29-39. https://journals.co.za/doi/abs/10.25159/0304-615X/1096 

Click to see what the machine says about the initiative and the book