AI in education in Namibia
THE AI FOCUS · EDUCATION · NBC-AIF-003
Watch the 10:10 source episode · Evidence record · Read the paper · Transcript
Joel Tiago | 16 September 2026 | Companion v1.0
AI can help us try another explanation, prepare an exercise or work across languages. For education, the important question is whether the learner understands more and can do the work independently. My education episode explored these possibilities alongside the risks of unequal access and over-reliance. This companion brings the broadcast and its ten-slide presentation into one source-linked account.
What the episode sets out to do
The three teaching routes are tutoring, support for teachers and local-language learning. They appear in slide 3 and at 01:35-01:59 in the transcript. The slides also add practical activities and a proposed school pilot. Those ideas deserve a place in the paper, with their status made clear: proposed activities are not measured results.
A more precise account of the school context
The statement “more than 700,000 learners” is a lower bound, not a current census total. The Ministry reports 927,647 learners for January 2025. The 1:38 national ratio needs correction: the 2024 official report gives 26.9 learners per teacher. Class size is a separate measure. The approximate Grade 12 failure figure remains unverified because no year, examination or threshold was specified. [R01, R02; E01-E03]
Learning with an AI tutor
A learner can ask for a smaller step, attempt it, then request feedback. This is more informative than copying a complete solution. Adapting pace should not become labelling a child with a fixed learning style. Evidence for matching teaching to such categories is insufficient. [R06]
A high-school mathematics trial in Turkey found that performance with AI help and performance after removing that help can differ. It gives a reason to test unaided understanding, not a forecast for Namibian learners or a rating of the products named here. [R08]
Useful work for teachers
Drafting a rubric, preparing questions and suggesting feedback are concrete uses. MagicSchool’s own guidance keeps the educator responsible for reviewing the result. A school should count that review time when deciding whether the tool saves work. Claims that AI removes all administration or always spots struggling learners need stronger evidence. [R07; E05]
Language access needs language review
The episode narrates a demonstration of science content requested in a local language. The educational aim is sound: learners need explanations in languages they understand. But generated text must still be checked for correct terminology, meaning and curriculum fit. This review did not score the language output or test children’s understanding. [R10; E06]
What can learners actually access
Khan Academy’s free learning resources are distinct from Khanmigo parent/learner access. The latter has payment and location conditions in the reviewed documentation, so the slides’ “completely free” guidance needs correction. Photomath’s current help also requires an Internet connection; the date of its change is unresolved. [R03-R05; S01, S02]
The record has not established whether Yyeni was operational in schools at broadcast time. The slides say it was in development. That question should be resolved with a dated release or deployment record before repeating an availability claim. [E09]
Activities from the presentation
Slide 8 proposes asking for a step-by-step simultaneous-equations explanation. A useful follow-up is to solve a similar problem without the tool and explain why each step is valid. The same slide proposes a five-question Grade 7 science quiz app. Before learners use it, a teacher should check each answer, the distractors and the reading level. The production draft’s timing and developer-price comparisons have not been verified. [S03]
Slide 5’s benefits-and-risks discussion belongs alongside these activities. Access outside class may help, but learners need suitable devices, connectivity, language support and adult guidance. The episode acknowledges that divide. A shared software service cannot on its own make rural and urban learning conditions equal. [E07]
The twenty-school pilot on slide 6 remains a proposal. A future study should define the learning question, necessary permissions, sample and scoring before recruitment. It should measure independent understanding, teacher effort and who is excluded. No such trial was conducted for this paper. [S04]
Disclosures and source record
I presented this episode and am AISOD’s founder and CEO. AISOD Researcher is an AISOD product, so its mention involves a commercial interest. This is not an independent product recommendation. My relationship to Yyeni and its historical deployment status remain to be clarified. [R11]
TurboScribe produced the English Whale transcript on 16 September 2026. The saved timestamped text matches the displayed transcript after whitespace normalization. A separate reading copy records context-supported spelling fixes, including Joel Tiago and AISOD Researcher. No full audio audit, independent coding or peer review is claimed. The paper reviews ten transcript claim groups and four additional slide claims, not every statement in the episode.
The video’s upload metadata is dated 11 June 2026; the air date is unconfirmed. The source deck is Episode2 Presentation Draft, slides 1-10. Its exact on-air version has not been established. The paper and evidence register retain those distinctions instead of turning draft content into broadcast facts.
Version history
Version 1.0, 16 September 2026: first education companion, preserved TurboScribe transcript, ten transcript assessments and four slide assessments. Working paper v0.1 is available with the source record.
Evidence record
Ten transcript groups (E01-E10) and four additional slide assessments (S01-S04). This is a selected review, not an accuracy score. Timestamps follow TurboScribe and are navigation aids.
E01Number of learnersSupported as a lower bound00:36-00:46
Source and propositionTranscript — Namibia has more than 700,000 learners.
AssessmentThe Ministry reports 927,647 learners in January 2025. The broadcast lower bound is compatible with that total but gives neither year nor coverage.
Editorial actionUse a dated total and identify the included school phases. Do not call the lower bound false.
Time scopeAssessed against dated sources where available; original air date unconfirmed
Slides 2 · Sources R01
E02Teacher-to-learner ratioCorrection needed for national-average framing00:36-00:53
Source and propositionTranscript — A ratio of one teacher to 38 learners is presented, rising to 50 in some places.
AssessmentThe official 2024 report gives 26.9 learners per teacher nationally. Its regional range is 23.9-31.7. Class size and learner-teacher ratio are different measures; a particular crowded class would not establish the national ratio.
Editorial actionReplace 1:38 as a national average with a dated official ratio; document any separate class-size example.
Time scopeAssessed against dated sources where available; original air date unconfirmed
Slides 2 · Sources R02
E03Grade 12 failure figureInsufficient evidence00:53-01:11
Source and propositionTranscript — Roughly 40 percent of learners fail Grade 12.
AssessmentThe passage supplies no examination year, qualification, threshold or denominator. Failure, subject grades and eligibility to progress are not interchangeable. A matching official statistic was not established in this review.
Editorial actionDo not repeat 40 percent as a verified national rate. Specify the examination, year and definition first.
Time scopeAssessed against dated sources where available; original air date unconfirmed
E04Learning pace and learning outcomesOvergeneralised causal claim01:49-02:44; 06:29-06:51
Source and propositionTranscript — Different learning speeds or methods explain most failure, and AI adapts to learners so they understand better and faster.
AssessmentNo causal evidence for most failure is supplied. Changing pace and explanations is a useful teaching option, but does not establish improved learning. EEF finds insufficient evidence for fixed learning-style matching.
Editorial actionDescribe adaptable practice as a possibility and test independent understanding; avoid fixed learner labels or a single explanation of failure.
Time scopeAssessed against dated sources where available; original air date unconfirmed
E05Teacher support and all administrative burdensSupported functions; universal claim unsupported03:14-04:18; 05:29-05:37
Source and propositionTranscript — AI supports marking and homework, detects struggling students and takes all administrative work away.
AssessmentMagicSchool documents draft rubrics, feedback and lesson plans with educator review. These features do not demonstrate removal of all work, guaranteed early detection or net time savings in Namibian schools.
Editorial actionRetain teacher-support examples; measure preparation, checking and rework together. Treat early-warning systems as a separate validated application.
Time scopeAssessed against dated sources where available; original air date unconfirmed
E06Local-language generation and teaching qualityEducational rationale supported; product quality unmeasured04:27-05:13; 08:28-09:31
Source and propositionTranscript — AISOD Researcher generates local-language content and a Grade 4 learner will understand mathematics more easily in Oshiwambo.
AssessmentTeaching in a language a learner understands is supported by UNESCO guidance. The narrated generation demonstration supplies no language-quality scoring, curriculum check or learner outcome. It cannot establish universal comprehension or the quality of every generated explanation.
Editorial actionRetain the demonstration as an illustration; have fluent educators check terminology and reasoning and evaluate learners separately.
Time scopeAssessed against dated sources where available; original air date unconfirmed
E07Rural equity and accessInternal qualification required06:51-07:23
Source and propositionTranscript — The episode acknowledges device and Internet gaps, then says rural and urban users get the same thing.
AssessmentThe adjacent qualifications matter: a common software service does not establish equal connectivity, affordability, language support, teacher help or learning outcomes. The equality claim is conditional even within the episode.
Editorial actionSay the tools may extend reach when access and support are adequate. Track who cannot participate.
Time scopeAssessed against dated sources where available; original air date unconfirmed
Slides 5 · Sources R13
E08Affordable tools and child accessProduct-specific qualification needed05:45-06:07
Source and propositionTranscript — The named tools can be tried today, free or at low cost, for education or children.
AssessmentA list of brands does not establish eligibility for every learner. Khan Academy’s free content must be distinguished from Khanmigo parent/learner access; the latter has payment and location restrictions. Other tools also need product-specific checks.
Editorial actionName the precise feature, eligible users, price and location. See S01 for the stronger slide claim.
Time scopeAssessed against dated sources where available; original air date unconfirmed
E09Yyeni deployment statusDeployment not established02:51-03:07; 05:53-05:58; 08:08-08:18
Source and propositionTranscript — Yyeni AI is described as bringing personalised teaching to school learners and included among tools to try.
AssessmentSlides 3, 4 and 6 describe Yyeni as in development; slide 4 also presents tools as usable today. No dated release record, school deployment evidence or evaluation was located. Development and operational use need distinct labels.
Editorial actionKeep deployment status unverified pending a dated source or author clarification. Do not infer ownership from co-mention.
Time scopeAssessed against dated sources where available; original air date unconfirmed
E10Over-reliance and independent learningRisk supported at a bounded scope07:53-08:01
Source and propositionTranscript — People can depend on AI rather than their own thinking.
AssessmentBastani et al. distinguish assisted performance from unaided learning in a Turkish high-school mathematics trial. This supports taking reliance seriously, but does not diagnose screen dependency or measure effects in Namibia.
Editorial actionUse hints and an unassisted follow-up task; evaluate learning rather than counting completed answers.
Time scopeAssessed against dated sources where available; original air date unconfirmed
Slides 5 · Sources R08
S01Khanmigo described as completely freeCorrection needed in supporting slidesSlides 3, 4 and 9
Source and propositionPresentation — Slides call Khanmigo a completely free tutor for all subjects and learners.
AssessmentKhan Academy separates free learning content and teacher tools from parent/learner Khanmigo access, which requires payment and US eligibility in the reviewed documentation. The stronger slide wording is not spoken verbatim in the transcript.
Editorial actionCorrect the slides’ availability guidance; retain the broader spoken recommendation only with product-specific conditions.
Time scopeAssessed against dated sources where available; original air date unconfirmed
S02Photomath offline useCurrent guidance needs correctionSlide 4
Source and propositionPresentation — The slide says Photomath works offline and solves any maths problem.
AssessmentCurrent Photomath help requires a stable Internet connection. It does not support the unrestricted any-problem claim. The help page does not date the change to cloud solving.
Editorial actionRemove the offline guarantee from current guidance. Do not infer when it became inaccurate.
Time scopeChange date unresolved; no retrospective verdict on offline availability
Slides 4 · Sources R05
S03Quiz creation time and developer-price comparisonProposed demonstration; comparison unverifiedSlide 8
Source and propositionPresentation — A quiz app is planned to appear in under two minutes, compared with N$5,000-N$15,000 and two weeks of developer work.
AssessmentThe transcript narrates a local-language content demonstration, not this timed quiz build. The draft provides no timing record, developer quotation or comparable specification.
Editorial actionUse the quiz prompt as a proposed classroom exercise; remove unmeasured time and price claims.
Time scopeAssessed against dated sources where available; original air date unconfirmed
S04Twenty-school pilotProposal, not a reported studySlide 6
Source and propositionPresentation — The slide proposes piloting AI tutoring in 20 schools during 2026, with AISOD available to support.
AssessmentThis is a proposal in production material. No approval, recruitment, funding, implementation or outcome data were supplied.
Editorial actionDescribe it as a proposed study only. Any future trial needs its own protocol, school permissions and appropriate review.
Time scopeAssessed against dated sources where available; original air date unconfirmed
Slides 6 · Sources R12
Sources and locations
R01 Ministry of Education, EMIS overview
2025 statistics; webpage publication date not displayed. Retrieved 2026-09-16.
January 2025 panel: 927,647 learners and 34,325 teachers; January 2024 panel: 893,415 learners.
Source page
R02 Ministry of Education, Fifteenth School Day Report 2024
2024. Retrieved 2026-09-16.
Report narrative: national learner-teacher ratio 26.9, regional range 23.9 to 31.7. Search-indexed official PDF text retrieved; direct PDF open failed.
Source page
R03 Khan Academy, How do I sign up for Khanmigo?
17 July 2024 page update. Retrieved 2026-09-16.
Age and location requirements; parent/learner features are not part of free teacher-tools subscription.
Source page
R04 Khan Academy, Khanmigo product FAQ
Current page; historical availability not reconstructed. Retrieved 2026-09-16.
Can anyone use Khanmigo? and Why is a payment required? Distinguishes free Khan Academy content, teacher tools and parent/learner access.
Source page
R05 Google, Does Photomath need the Internet?
Undated current help page. Retrieved 2026-09-16.
Opening answer requires stable Internet; describes change to cloud-based solving without a transition date.
Source page
R06 Education Endowment Foundation, Learning styles
Review updated October 2025. Retrieved 2026-09-16.
Key findings and evidence-security section: insufficient evidence for matching instruction to fixed learning-style categories.
Source page
R07 MagicSchool, FAQ
Current page. Retrieved 2026-09-16.
Grading and lesson-plan sections: rubrics and draft feedback, reviewed and finalized by teachers.
Source page
R08 Bastani et al., Generative AI without guardrails can harm learning
25 June 2025; publisher marks corrected article. Retrieved 2026-09-16.
Current corrected article, Significance, Experimental Design and Main Results: assisted practice versus unassisted exam; a high school in Turkey. Correction notice linked by publisher, dated 20 August 2025; notice body was not retrieved. No effect-size estimate is reproduced here.
Source page
R09 Miao and Holmes, Guidance for generative AI in education and research
2023 guidance; overview updated 16 January 2026. Retrieved 2026-09-16.
UNESCO publication overview: human agency, age-appropriate use, data privacy and pedagogical validation.
Source page
R10 UNESCO, Languages matter: Global guidance on multilingual education
2025 guidance; overview updated 22 July 2026. Retrieved 2026-09-16.
Guidance overview: quality education in languages learners understand; multilingual education policy. Does not evaluate AISOD products.
Source page
R11 AISOD, About us
Undated current page. Retrieved 2026-09-16.
Product list identifies AISOD Researcher; team identifies Joel Tiago as founder/CEO. A commercial source, not an independent performance evaluation.
Source page
R12 Joel Tiago, Episode2 Presentation Draft.pptx
Production draft; slide 1 says 8 June 2026. Retrieved 2026-09-16.
Slides 1-10; local text extraction preserved. Exact on-air version not confirmed.
R13 NBC Digital News, Education episode
Upload metadata 11 June 2026; air date unconfirmed. Retrieved 2026-09-16.
10:10 episode; TurboScribe timestamped transcript obtained 16 September 2026.
Source page
Transcript and supporting files
The raw machine transcript remains unchanged. The reading copy applies only the logged spelling corrections. Neither is certified as an audio-verified transcript.
Read the timestamped reading copy
(0:00) We'll explore how artificial intelligence is transforming education and what could become (0:05) the smartest classroom in Namibia. From personalized AI tutors and teacher support tools (0:12) to learning in local languages, we examine how technology could help bridge educational gaps (0:19) and unlock new opportunities for every Namibian. Let's dive into it right now. (0:28) Currently, our challenges are quite huge but not really complex because we can solve them. (0:36) We have more than 700,000 learners in our schools and the ratio between teachers and learners (0:46) between one to 38, which sometimes in certain places it can go up to 50. And (0:53) 40 percent or more or less 40 percent can be more than 40 percent or less than 40 percent of (1:00) our learners fail grade 12. And we have 13 recognized local languages but that does not (1:11) mean that all our content is in our languages. Most of them are just in English or in Afrikaans. (1:19) This is the reality that AI is working into right now but that also means that the opportunity is (1:27) enormous and it's going to be enormous if access is equal. So how will AI solve such kind of (1:35) problems? We have three ways that we can use to solve these problems using artificial intelligence (1:41) and one of the ways will be AI tutoring and the other will be AI for teachers and the last one (1:49) will be local language AI. So AI tutoring is personalized teaching at scale. It's bringing (1:59) or adapting to the learners or to the person's speed of learning and methodology of learning. (2:07) Not everyone learns at the same speed. That's the reason why most people fail. They do not fail (2:14) because they are stupid. They fail because they do not learn in the same speed as the others (2:19) and sometimes they do not learn in the same or they do not adapt to the methodologies (2:26) of teaching that we have. But artificial intelligence brings this adaptation to them (2:32) and artificial intelligence can teach or can explain things to them in the way that they (2:39) understand and in the speed that they may understand. And artificial intelligence is (2:44) available 24-7. It doesn't lose patience. It doesn't become angry or stuff like that. (2:51) We can see such solutions at Khan Academy and in Namibia we have some wonderful companies that (3:00) are working into this and we have Yyeni AI that is bringing this personalized teaching to learners (3:07) in our schools. And talking about AI for teachers, we are not trying to replace teachers. What we are (3:14) trying to do is to upgrade the teachers because AI is able to help the teachers mark tests (3:23) faster than they could and generate personalized homeworks. AI flags struggling students (3:31) before they fall behind so AI can tell the teacher that, okay, this student is actually (3:36) falling behind. We need to use another methodology of teaching so that he can understand better. (3:43) And AI takes all the admin work from the teacher so that the teacher can focus on more important (3:51) things which are mentorship. So the teacher will be focused on mentoring the students, the learners, (3:57) instead of focusing on admin work which most of the times take a lot of their time. But one thing (4:04) you must also understand is that AI will never replace the teacher because something great about (4:10) teachers is inspiration and care. These are things that AI cannot be able to give. AI can't give (4:18) that to a student because AI is not a human being. And we also have local language artificial (4:27) intelligence. So this is where AI is able to speak in Oshiwambo, is able to speak in Oshiwambo, (4:33) in our local languages. We can see that at AISOD Researcher. You can see contents being generated (4:39) by AI in our local languages. A student who is in Ondangwa, who is probably in grade four, (4:48) will understand math easier if you talk to him in Oshiwambo instead of you trying to (4:57) teach him in English. So when you tell him in English, he has to translate math to Oshiwambo (5:04) and then understand. But when you tell him straight in Oshiwambo, he doesn't need to do that. (5:07) He will understand the message straight. So we have different technologies like Google and other (5:13) companies that are trying their best to bring Bantu languages into their system. But in Namibia, (5:21) we have UNAM that could build this library through their language department. And this (5:29) is something that most companies are working at in bringing solutions in Namibia. AI does not (5:37) and will not replace teachers, but it will remove all the burdens that teachers usually have. (5:45) Right now, we have different tools available, which is MyResearcher, PhotoMath, Duolingo, (5:53) Yyeni AI, Khan Academy. These are platforms that you can use. You can go there today and try it (5:58) for free or at low cost for education or for your kids. Just try it out and see how AI is (6:07) transforming different solutions. But let me say, our education system is available for different (6:16) solutions. It's also available for many opportunities, both for companies and for (6:22) governments. Now, there are advantages and there are disadvantages of artificial intelligence. (6:29) And one of them is the personalized learning. AI brings that personalized learning to the learner (6:35) because it adapts to the learner's speed of learning and method or style of learning. And (6:44) it can teach him in the way that that person will understand better and faster. And it's available (6:51) 24-7. But now, there is a disadvantage to that because of the digital divide. Digital divide (6:58) is, it happens because people or not everyone has gadgets, not everyone has computers, and not (7:05) everyone has access to internet. And we also have another advantage of artificial intelligence, which (7:12) is rural equity. The same person in Windhoek gets the same, what people in Windhoek get is the same (7:23) that people in rural areas like in Rundu or Ondangwa get. So, we have teacher support as well. AI (7:34) helps teachers with admin work. While all these good things are there, there are also bad stuff (7:41) because some teachers think that AI will replace their jobs. So, if you do not address it correctly, (7:46) it may bring a problem, they may go angry with the artificial intelligence instead of receiving it (7:53) or embracing it. And also, AI can bring digital or screen dependency. People can depend on AI more (8:01) than depending on their own intelligence. And we also spoke about local languages like (8:08) Damara and other languages. So, to sum up, we have Yyeni AI and AISOD Researcher that are bringing great (8:18) innovations in Namibia. We have UNAM that's exploring different solutions. And we also have (8:22) the Ministry of Education, Arts and Culture that are bringing different opportunities. So, we'll (8:28) watch a quick demonstration here to show a bit on how artificial intelligence is actually (8:37) bringing transformations and talking or communicating in local languages in Namibia. (8:44) So, right now we can see we are trying to communicate to it to give us some content and (8:50) it's giving us some science content. But then we are also going to request it to give us that (8:56) same content in a local language. So, this is a platform. It's AISOD Researcher or my researcher (9:05) and it's able to give you content both in English and in Africa or in Oshuwambo. As you can see in (9:13) the screen, right now we are requesting it. And wow, as you can see, it's generating the content (9:21) in our local languages. So, AI is here and we are all set to transform Namibia because we are not (9:31) late. So, that is it for today. Thank you so much for watching and I want you to remember that (9:43) the future of education is not just about technology. It's about ensuring every learner (9:50) has access to it. So, join us next week as we explore AI, how it's reshaping healthcare and (10:00) bringing medical expertise closer to every community. (10:05) My name is Joel Tiago and thank you for watching. Stay tuned.