
EP 39 | Keeping African Agency in AI: Vukosi Marivate’s Call to Build
Loading player...
Keep African agency in AI - that’s the through-line of Prof. Vukosi Marivate’s talk. Don’t outsource the future, build it locally. He points to a decade of grassroots work (Data Science Africa, Deep Learning Indaba, Masakhane) and argues for a full ecosystem: fundamental researchers, model builders, product teams, and real markets to test and absorb what’s built. He stresses R&D investment (AU benchmark 1% of GDP; SA ~0.6% after a corporate pullback from ~0.8% pre-COVID), and the role both government and corporates must play.
Language is a priority - with 2000+ African languages, we need digital dictionaries, speech recognition, and NLP tools that actually work for our context, otherwise Africa stays a consumer, not a creator. He explains why bias can’t be “fine-tuned out” of large models after the fact; you need better representation from the ground up. He also outlines the technical and data hurdles for LLMs in morphologically rich languages like isiZulu.
Chapters / timestamps
00:00 – Intro & “latest iteration” of his message
00:36 – Key takeaway: Africans must keep agency in AI; grassroots orgs to engage
01:23 – What a healthy African AI ecosystem includes (research → product → market)
03:46 – R&D funding realities: AU 1% benchmark; SA numbers; corporate role
06:19 – Market potential vs perceptions; demographics and opportunity
06:46 – 2000+ languages: designing systems that meet people where they are
08:35 – Why language matters: NLP journey, lack of tools, activism for languages
10:38 – Build locally vs being only consumers; multinationals without R&D offices
11:07 – Tanzania example: low-power, low-connectivity edge ML for farmers
13:59 – LLM challenges for African languages (data quality, morphology, encodings)
17:36 – Skills roadmap: read papers, implement, solid data/computing fundamentals
20:09 – Dev example: TTS trained on religious texts—limits and pitfalls
20:56 – Practical advice: use general-purpose/open datasets; improve with limited data
22:25 – Resource-efficient AI; Lab by AI speech/translation; API access notes
23:57 – What’s next: building traction; new Institute for Data Science & AI (UVic)
24:46 – For AI skeptics: get literate; understand benefits/risks; ask better questions
25:21 – Close
#TheAnglePodcast #VukosiMarivate #AfricanAI #AIAgency #NLP #AfricanLanguages #BiasInAI #EdgeAI #RAndD #BuildInAfrica
Produced and Published by Submedia.co.za
Language is a priority - with 2000+ African languages, we need digital dictionaries, speech recognition, and NLP tools that actually work for our context, otherwise Africa stays a consumer, not a creator. He explains why bias can’t be “fine-tuned out” of large models after the fact; you need better representation from the ground up. He also outlines the technical and data hurdles for LLMs in morphologically rich languages like isiZulu.
Chapters / timestamps
00:00 – Intro & “latest iteration” of his message
00:36 – Key takeaway: Africans must keep agency in AI; grassroots orgs to engage
01:23 – What a healthy African AI ecosystem includes (research → product → market)
03:46 – R&D funding realities: AU 1% benchmark; SA numbers; corporate role
06:19 – Market potential vs perceptions; demographics and opportunity
06:46 – 2000+ languages: designing systems that meet people where they are
08:35 – Why language matters: NLP journey, lack of tools, activism for languages
10:38 – Build locally vs being only consumers; multinationals without R&D offices
11:07 – Tanzania example: low-power, low-connectivity edge ML for farmers
13:59 – LLM challenges for African languages (data quality, morphology, encodings)
17:36 – Skills roadmap: read papers, implement, solid data/computing fundamentals
20:09 – Dev example: TTS trained on religious texts—limits and pitfalls
20:56 – Practical advice: use general-purpose/open datasets; improve with limited data
22:25 – Resource-efficient AI; Lab by AI speech/translation; API access notes
23:57 – What’s next: building traction; new Institute for Data Science & AI (UVic)
24:46 – For AI skeptics: get literate; understand benefits/risks; ask better questions
25:21 – Close
#TheAnglePodcast #VukosiMarivate #AfricanAI #AIAgency #NLP #AfricanLanguages #BiasInAI #EdgeAI #RAndD #BuildInAfrica
Produced and Published by Submedia.co.za





