Edition 83

Published
3 November 2026
Reading time
3 min
Contents
5 stories · 3 sectors

AI's answers, and who checks them

South Africa's government favours giving existing regulators the job of overseeing AI rather than creating a new authority, and TechCabal asks whether those regulators will have the resources to investigate AI harms and give consumers redress. Rufaro Mafinyani sets out three shifts reaching local businesses faster than the rules: small AI models that run on phones in isiZulu and isiXhosa, AI memory that can outlive a deleted client file, and pay-per-result pricing that puts outsourcing jobs under pressure. Google has launched a free AI learning hub for Africa, designed for entry-level smartphones, with isiZulu and isiXhosa being added. Further afield, academics in Australia explain where AI translation hides errors, and how to use a chatbot safely before a medical appointment.

Every story headline links to its original source.

Policy & governance

  1. South Africa’s AI regulation problem is bigger than its delayed policy

    TechCabalPolicy

    The government favours giving existing regulators the job of overseeing AI rather than creating a dedicated AI authority, Jeanette Morwane, deputy director-general at the department of communications and digital technologies, indicated at a Bowmans event in October. TechCabal argues that the replacement AI policy's bigger test is whether those regulators get the resources and clear responsibilities to investigate AI-related harms and give consumers redress. Morwane cited one regulator where two staff members were tasked with handling more than 200 cases involving child sexual abuse material, and the National Consumer Commission's acting commissioner, Hardin Ratshisusu, said existing laws could potentially address AI-related fraud and misinformation.

Society & work

  1. AI translation sounds convincing, but it can be riddled with hidden errors

    Stuff South AfricaSociety

    AI translation tools are fluent, but fluency does not equal accuracy, and even proficient bilinguals can miss errors hidden beneath natural-sounding text, writes Alisa Tian of UNSW, who reviews AI translations between English and Chinese. Tian says mistakes range from wrong word choices and tenses to hallucinations (confident presentation of false information), and that AI may use outdated or insensitive terminology where newer expressions are underrepresented in its training data. Tian says AI translation is more reliable for familiar, unambiguous topics between widely used languages, that the problems could be more serious in less common languages, and that accuracy can never be guaranteed without human checking.
  2. Using ChatGPT before a medical appointment? How to get the best support for your health questions

    Stuff South AfricaSociety

    Many people now ask AI chatbots such as ChatGPT about their health before seeing a doctor, but studies cited by Kirsten McCaffery, Julie Ayre and Liliana Laranjo of the University of Sydney show the advice can be unsafe or inaccurate, especially when questions lack detail. An international team has released the SMART guide, five steps for safer use, including checking facts and speaking to a real person when needed. The authors also urge patients to tell their doctor about chatbot use, so inaccuracies can be corrected.

Education & skills

  1. Google launches free digital and AI learning hub for Africa

    Hypertext (htxt)Education

    Google has launched a free learning hub for Africa, Skilling Africa with Google, hosting more than 15 digital and AI training programmes designed to work on entry-level smartphones and slow connections, with lessons downloadable for offline completion. Google says isiZulu and isiXhosa translations are being added, and that an AI study companion called SCOUT answers questions and helps learners plan study schedules. Google Africa MD Alex Okosi says the company aims to equip 3 million people with digital and AI skills by 2030.

More in this edition

  1. Smaller AI, longer memories: the changing economics of intelligence

    Business Day / BusinessLIVE

    Rufaro Mafinyani of Intellica Analytics writes that three shifts in AI are reaching South Africa faster than the contracts and policies meant to govern them. Small models are one: a trimmed version of South African start-up Lelapa AI's InkubaLM, which reads isiZulu and isiXhosa, fits in about 20MB, the kind of phone-sized "small AI" the World Bank's chief knowledge officer says developing countries should build. Another is AI memory: deleting a client's file does not delete the copies AI systems derive from it, which Mafinyani says creates exposure under POPIA, the data protection law. The third is paying suppliers only for resolved queries, and Mafinyani says falling prices for them land hardest on South Africa's business process outsourcing sector, which has created about 180,000 jobs in 10 years, mostly for young people.

Get the next edition

Confirm by email Published when there’s real news Unsubscribe anytime