Ubongo Kids in 20 years: using AI to turn broadcast learning into meaningful feedback

Ubongo Kids’ success so far is not accidental. It comes from combining curriculum-based learning with strong storytelling and a clear focus on local languages and cultural relevance. Research on educational media suggests that learning is more effective when educational content is embedded in engaging narratives instead of delivered in a purely instructional way (Holmes, Bialik and Fadel, 2019). Using television, radio, and low-data digital platforms has helped Ubongo reach children who are often left out of more technology-heavy education programmes, especially in low-resource settings (Holmes, Bialik and Fadel, 2019).

While Ubongo has been very successful at reaching learners, it still faces a challenge common to broadcast education - limited opportunities for meaningful feedback. Children may attempt to answer questions while watching programmes, but the system itself usually cannot respond to those attempts. This limitation matters because a strong body of research shows that timely, specific feedback is one of the most important factors in improving learning outcomes, particularly for younger learners (VanLehn, 2011).

Designing an AI-supported feedback loop for Ubongo Kids

Rather than using artificial intelligence simply to create more content, a more meaningful long-term use would be to support personalised feedback at scale. Personalised learning is frequently highlighted as a key potential benefit of educational technology, but it is also recognised as difficult to achieve in low-resource and multilingual settings (Holmes, Bialik and Fadel, 2019). A multilingual, AI-supported learning companion could help Ubongo address this challenge without changing its existing model.

Figure 1 shows how this could work as a feedback loop rather than a one-way broadcast. Ubongo episodes could continue to include short moments where characters pause and invite children to answer a question out loud. Breaking learning into small steps and responding to learner input has been shown to support better understanding than passive exposure alone (Ma et al., 2014). After the episode, a simple mobile or voice-based system such as WhatsApp or other low-data apps could provide a few minutes of follow-up practice linked directly to the learning objective.

Children would respond by speaking, and the AI system would identify patterns in their answers. It could then provide short corrective feedback and select the next question based on areas of difficulty. Research comparing different tutoring approaches suggests that adaptive feedback systems can approach the effectiveness of human tutoring unded certain conditions (VanLehn, 2011). Improvements in speech recognition and multilingual technology mean that voice-based interaction is becoming more realistic in low-connectivity settings and for languages that have often been overlooked in digital systems (Besacier et al., 2014).

AI-Generated Image of Children Learning with AI System at Home

This system could also support adults around the child. Caregivers and teachers could receive short summaries explaining what a child appears confident with and where additional support may be needed. Research suggests that learning data is most useful when it is easy to interpret and linked to clear actions, instead of presented as abstract scores or complex dashboards (Holmes, Bialik and Fadel, 2019).

Another benefit of this approach is its potential to support impact evaluation. Broadcast education initiatives often rely on surveys or occasional assessments, which can be costly and provide limited insight into ongoing learning. Adaptive learning systems collect data from ongoing interactions, which can show how learners develop over time (Ma et al., 2014). When combined and anonymised, this data could help Ubongo improve content where learners commonly struggle.

Working through this idea has changed how I think about AI in education. I previously viewed AI mainly as a tool for efficiency or automation. Engaging with the research suggests that its greatest value in contexts like Ubongo lies in supporting feedback rather than replacing human or creative elements of learning (Holmes, Bialik and Fadel, 2019).

If Ubongo Kids develops AI in this way, it could move beyond one-way broadcasting and towards a learning system that remains accessible and culturally grounded, while still being effective to individual learners. This would support the view that advanced educational technology does not need to be expensive or exclusive in order to be effective (Ma et al., 2014; VanLehn, 2011).

Reference List

Besacier, L, Barnard, E, Karpov, A and Schultz, T (2014) ‘Automatic speech recognition for under-resourced languages: A survey’, Speech Communication, 56, pp. 85–100.

Holmes, W, Bialik, M and Fadel, C (2019) Artificial Intelligence in Education: Promises and Implications for Teaching and Learning. Boston: Center for Curriculum Redesign.

Ma, W, Adesope, OO, Nesbit, JC and Liu, Q (2014) ‘Intelligent tutoring systems and learning outcomes: A meta-analysis’, Journal of Educational Psychology, 106(4), pp. 901–918.

VanLehn, K (2011) ‘The relative effectiveness of human tutoring, intelligent tutoring systems, and other tutoring systems’, Educational Psychologist, 46(4), pp. 197–221.