<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Kossiso Royce on Medium]]></title>
        <description><![CDATA[Stories by Kossiso Royce on Medium]]></description>
        <link>https://medium.com/@kossisoroyce?source=rss-94b30029db94------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*Tv4ygZNazxZnQ10y3tbE1w.png</url>
            <title>Stories by Kossiso Royce on Medium</title>
            <link>https://medium.com/@kossisoroyce?source=rss-94b30029db94------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 23:45:34 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@kossisoroyce/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[The Story of AI and Machine Learning in Africa Begins Here]]></title>
            <link>https://kossisoroyce.medium.com/the-story-of-ai-and-machine-learning-in-africa-begins-here-f991f8028057?source=rss-94b30029db94------2</link>
            <guid isPermaLink="false">https://medium.com/p/f991f8028057</guid>
            <category><![CDATA[africa]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[development]]></category>
            <dc:creator><![CDATA[Kossiso Royce]]></dc:creator>
            <pubDate>Wed, 22 Oct 2025 00:25:07 GMT</pubDate>
            <atom:updated>2025-10-22T00:25:07.027Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Kyd7liRV_H5zQNoJ" /><figcaption>Photo by <a href="https://unsplash.com/@possessedphotography?utm_source=medium&amp;utm_medium=referral">Possessed Photography</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Everyone wants to build AI for Africa. But there’s a problem: the data doesn’t exist.</p><p>I don’t mean it’s expensive or hard to access. I mean, it was never collected. Decades of education records, healthcare measurements, and infrastructure data — just not recorded. And you can’t train models on data that doesn’t exist.</p><p>I realised this while working on <a href="https://electricsheep.africa/">Electric Sheep Africa</a>, curating datasets. At first, I thought: if we raise money, we can gather what we need. Then it hit me. You can’t gather historical data. That moment in 2015, that measurement from 2018 — it’s gone. Time doesn’t run backwards.</p><p>So I did something simple: I made the data.</p><h3>Not Making It Up</h3><p>This needs to be clear. I’m not inventing numbers. I scraped whatever real data exists — World Bank reports, UNICEF statistics, ministry publications. When UNICEF says Nigeria’s primary enrollment rate is 68%, that becomes my parameter. When the World Bank reports a 1:40 teacher-student ratio, that becomes my constraint.</p><p>Then I use basic statistics. Normal distributions for test scores, because grades cluster around averages. Exponential distributions for session durations, because most are short and some are long. Weighted sampling based on real frequencies I can find.</p><p>No machine learning. No neural networks. Just probability distributions that statisticians have used for a century.</p><p>For medical data, I collected existing African imaging datasets, augmented them, and generated synthetic health records using physiological ranges and epidemiological patterns. If diabetes prevalence is 8% in a region, that’s my parameter.</p><h3>The Useful Kind of Wrong</h3><p>Synthetic data isn’t real data. I know that. It’s based on assumptions and distributions I encoded. It might be wrong in ways I can’t predict.</p><p>But what’s the alternative? Without data, researchers either skip African contexts entirely or make assumptions anyway — just hidden ones, buried in their models. At least synthetic data makes the assumptions explicit. You can see my parameters, disagree, and generate your own version.</p><p>The absence of data isn’t neutral. Someone decided not to collect it, not to fund it, not to prioritise it. I’m making a different choice.</p><h3>What Foundations Look Like</h3><p>Dataset repositories aren’t exciting. Synthetic data sounds questionable. There’s no flashy demo, no benchmark to beat, no paper at a top conference.</p><p>But I’m building something specific: the ability for someone to test an education model on Nigerian school data. The ability to develop diagnostic tools for African medical imaging. <strong><em>The ability to research at all.</em></strong></p><p>Every self-driving car model is trained on datasets that someone collected. Every medical AI used imaging databases that someone compiled. The models get the attention. The datasets are footnotes.</p><p><em>For Africa, we don’t even have the footnotes yet.</em></p><h3>Why Continue</h3><p>Because someone has to build the baseline. Someone has to create the infrastructure that seems pointless until everyone’s using it.</p><p>Maybe in five years, real data collection advances enough that my synthetic datasets become obsolete. That would be perfect. It would mean I built something that made itself unnecessary, which is what good infrastructure does.</p><p>Or maybe researchers use these datasets to develop methods that work immediately when real data arrives. Maybe someone’s model performs badly on my synthetic data, and that reveals what’s missing. Maybe someone notices what I couldn’t simulate well and realises that’s the research opportunity.</p><p>The work is slow. I spend days scraping PDFs from ministry websites, parsing inconsistent formats, checking distributions, and writing documentation. Most people will think it’s either useless (because it’s synthetic) or trivial (because “anyone could generate random data”).</p><p>But it’s not random. Every distribution has a reason. Every parameter comes from somewhere real. Every relationship follows documented patterns.</p><h3>Where Stories Begin</h3><p>The story of AI in Africa doesn’t begin with a breakthrough model or a famous lab or a big investment. Those things need foundations.</p><p>It begins with probability distributions seeded with real statistics. With documentation that makes assumptions explicit. With someone building the parts that let other people build the interesting parts.</p><p>I don’t know if Electric Sheep Africa will matter. I don’t know if anyone will use these datasets. But I know the story couldn’t begin until someone did this work.</p><p>When someone eventually builds something remarkable on African data — real or synthetic or some combination we haven’t imagined — it will be possible because someone built the boring parts first.</p><p>The story begins in the infrastructure. In the datasets. In the foundation work that makes everything else possible.</p><p>Here.</p><p>You can follow my work here: <a href="https://huggingface.co/electricsheepafrica">https://huggingface.co/electricsheepafrica</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f991f8028057" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>