Africa's Triple Digital Divide: When AI Can't Even Pronounce 'Iqanda'
What happens when the technology meant to democratise communication might just fall short of digitally speaking your language in sectors like finance, health or education?
Recently I created a "video" demonstrating something to ponder: AI's challenged ability to pronounce isiXhosa. Not "struggles with" or "needs improvement", but fundamentally struggles. Watch the demonstration embedded below to see TTS engines treat click consonants as system errors, forcing me to apply a phonetic hack.
[Video: Watch AI Challenged by Basic isiXhosa Pronunciation]
Arthur Goldstuck recently opined of Africa's "double digital divide" - the gap between connectivity and computing capacity. At the Ministerial Forum on Artificial Intelligence, he revealed that Africa contributes only 1% of global computing capacity, whilst the intelligent economy races toward R322-trillion by 2030 - "Africa generates vast amounts of data," Goldstuck notes, "but most is processed overseas".
But there's a third divide Goldstuck may have missed.
The Linguistic Divide: The Click = Glitch
As my video demonstrates, standard commercial TTS engines fail to a large extent with transcribing isiXhosa. The reason? Clicks are non-pulmonic consonants - they don't use airflow from our lungs like most European or Western sounds. AI trained primarily on European languages interprets these fundamental phonemes as "non-verbal interruptions". Background noise. Potential system errors.
To make our HAIBO PHANDA [I had to remove the "H" to convert the audio TTS output to pronounce "PHANDA"] educational avatars even approximate isiXhosa, I must write phonetically incorrect text - deliberate misspellings that trick AI into producing something more familiar. Ergo, one might imagine having to misspell your own name in order TTS engines may more accurately pronounce it. This reality, but I'm rather optimistic. In this sense, Minister Malatsi proclaimed at the same forum: "AI must work in our languages, in our contexts, and for our use cases".
The Numbers Tell the Story
Goldstuck cites that 70% of sub-Saharan Africans are under 30, yet 75% lack digital skills. Here's what those statistics perhaps hide: young isiXhosa speakers face an exciting choice - challenge linguistic AI authenticity by being aware of the possible digital excluded - pursue TTS tech of the sovereign kind. You can be both AI-literate and speak authentic isiXhosa. The technology just needs a slight prompt.
Sixteen African countries have drafted AI strategies. Addressing this linguistic exclusion has only been recently ushered-in with focussed intent. Perhaps the focus on data centres, AI factory computing power, skills development should be as crucial as AI being more inclusive of 2,000+ African languages.
Beyond Tick-Box Inclusion
The solution isn't adding African languages to existing systems. As I demonstrate in the video, that produces "tick-box accessibility" - language options that mask profound phonetic challenges. Haibo Phanda envisions vernacular-oriented TTS tools, building "context-relevant AI, one vernac at a time" - AI rebuilt from African phonetics upward.
This requires more than proposed AI_factory Nvidia-chip-driven centres. We need linguistic factories. We need AI factories that understand clicks aren't glitches, but opportunities.
The G20 - let's make it click
This November, world leaders gather in South Africa to explore a more inclusive, pan-African AI future, touting policies from digital inclusion, accessibility to digital transformation. Given that AI can't quite pronounce "iqanda" without phonetic trickery or can't recognise click consonants, I am excited that perhaps, millions may not have to choose between their mother tongue and digital participation.
Goldstuck asks whether Africa will "own more than their data". But, we need to be sensitised to a possible compromise - our AI digital future need not be one of linguistic of abandon.
Our triple divide - connectivity, computing factories, and linguistic authenticity - reveals AI's promise of inclusion where we must avoid technological gaslighting. Let's build a digital future where African vernac is inclusively transcribed. An AI interface where every click, honours every tone, respects every vernacular - let's jointly cut that ribbon at the opening of our digital bridge.