Generative AI boosts contextualized STEAM learning across Global South

The study reports that contextual relevance was another major strength of the AI-generated plans. These plans incorporated community artifacts, indigenous knowledge, local analogies, and familiar socio-cultural experiences. Examples tied to mobile money transactions, seasonal farming, local festivals, calabash engravings, and bilingual communication patterns made abstract STEAM concepts more relatable.

A team of Ghanaian and international education researchers has reported that customized generative AI tools can significantly strengthen the cultural relevance and instructional quality of STEAM lesson plans used across basic and junior high schools. Their findings show that AI-generated lessons, when guided by culturally responsive prompts, outperform Ghana’s standards-based NaCCA curriculum materials in contextual grounding, student engagement, and linguistic inclusiveness.

The study, titled Human Experts’ Evaluation of Generative AI for Contextualizing STEAM Education in the Global South and published on arXiv, examines how well generative AI could support culturally responsive pedagogy when used with a customized Culturally Responsive Lesson Planner, a semi-automated AI tool developed to embed Ghanaian languages, cultural practices, and local examples into STEAM instruction.

The research team worked with four STEAM education experts who evaluated AI-generated lesson plans and official NaCCA plans across mathematics, science, creative arts, and computing. They applied a validated rubric measuring bias awareness, cultural representation, contextual relevance, linguistic responsiveness, and teacher agency.

Their findings point to a growing role for human-AI collaboration in designing equitable and culturally grounded teaching materials for the Global South.

AI generates more culturally grounded lessons than national curriculum materials

The experts’ evaluations revealed that lesson plans produced through the Culturally Responsive Lesson Planner consistently scored higher than NaCCA plans in almost every domain tied to meaningful cultural and contextual inclusion. The strongest performance appeared in teacher agency, where AI-generated lessons reached the highest overall mean score. Experts found that the CRLP tool made lesson plans easier to adapt, more flexible, and more aligned with the realities of multilingual classrooms.

The study reports that contextual relevance was another major strength of the AI-generated plans. These plans incorporated community artifacts, indigenous knowledge, local analogies, and familiar socio-cultural experiences. Examples tied to mobile money transactions, seasonal farming, local festivals, calabash engravings, and bilingual communication patterns made abstract STEAM concepts more relatable.

By contrast, NaCCA lesson plans were found to be structurally sound but largely generic, lacking references to the lived experiences of Ghanaian learners. The experts noted that NaCCA plans closely follow national standards but do not provide cultural specificity or local examples that could help bridge gaps between curriculum expectations and classroom realities.

The researchers state that the AI tool excelled because its interactive prompting system encouraged experts to feed local linguistic elements, community symbols, and contextual data into the generation process. This allowed the AI to produce lesson plans that felt familiar and relevant to learners across Ghana’s diverse cultural landscapes. In mathematics, science, creative arts, and computing, reviewers described the AI-generated content as more engaging, more student-centered, and more responsive to Ghana’s multilingual environment.

Cultural representation and translation accuracy remain major weaknesses for AI

Despite strong overall performance, the study identifies cultural representation as the weakest domain for the AI-generated plans. The experts found that generative AI struggled with Ghana’s cultural diversity, often producing shallow or incomplete references to local traditions, languages, historical narratives, or identity expressions. This weakness was particularly visible in mathematics and computing, where the AI failed to weave cultural nuance into more abstract topics.

The researchers attribute this gap to the limitations of the AI’s training data, which is shaped by dominant, global language sources and contains little indigenous African content. As a result, AI-generated references to Ghanaian languages or cultural expressions were sometimes inaccurate or oversimplified.

Linguistic responsiveness also showed mixed results. The CRLP tool produced bilingual lesson content by integrating English with local languages such as Dagbani and Dagaare. Experts praised this bilingual approach as a meaningful step toward inclusive education. They confirmed that allowing students to respond in either English or a local language promotes deeper understanding and accommodates multilingual learners.

However, several reviewers raised concerns over translation inaccuracies, inconsistent syntax, and the risk of excluding students who do not speak the selected local language. Some AI-generated translations did not accurately reflect meaning, while certain idiomatic expressions lacked cultural authenticity.

The researchers warn that these weaknesses illustrate why teacher oversight remains essential. AI can generate materials quickly, but it cannot yet navigate the full complexity of Ghana’s cultural and linguistic landscape. Teachers must review, refine, and contextualize AI-produced lesson plans before they reach the classroom.

Experts call for stronger teacher oversight and local data integration as AI adoption grows

According to the authors, generative AI should not replace teachers but instead act as a co-designer in lesson development. The strongest endorsement from experts came from their observations that AI tools significantly reduce teacher workload by automating time-consuming lesson planning tasks. Automation, they said, frees educators to focus on refining content, adapting lessons, and addressing diverse student needs.

The CRLP-generated plans were also praised for their practical feasibility. The use of locally available materials, leaves, sticks, baskets, report cards, and community artifacts, made the lessons more sustainable and cost-effective. In contrast, some NaCCA plans required materials not easily accessible in low-resource schools.

However, the researchers point up that teachers must remain central to the process, especially in filtering out bias, correcting inaccurate translations, strengthening cultural representation, and ensuring alignment with national cross-cutting priorities like gender equity, sustainability, and digital literacy. While the NaCCA lesson plans address these cross-cutting areas more consistently, the AI-generated plans depend entirely on the quality and specificity of prompts.

The study’s mixed-methods approach, combining numerical scoring with expert reflections, revealed that AI tools are most effective when guided by educators with deep knowledge of Ghanaian cultural and linguistic contexts. The authors propose that future teacher training programs should integrate GenAI literacy to help educators adopt, adapt, and ethically apply AI tools in the classroom.

They also recommend that future research include fine-tuning generative AI models using indigenous Ghanaian corpora to improve accuracy in language and cultural representation. Expanding the expert pool, testing classroom implementation, and exploring local data sets could further strengthen the effectiveness of AI-supported lesson design across the country.