evidoria

← Back to browse

Good practice

RobotsMali — AI-Generated Bambara-Language Children's Books After Mali's Language Policy Shift

Mali · Bamako · See the Mali profile

After Mali dropped French as its sole official language in 2023, RobotsMali used ChatGPT, Google Translate and an AI image generator to produce 107 culturally grounded Bambara-language children's books in under a year, reaching over 300 elementary pupils.

107 books
Bambara-language children's books produced (under one year, from 2023 pivot)
60 draft textbooks
Draft textbooks produced in proof-of-concept sprint (8-week sprint)
300+ elementary pupils
Pupils reached in first year of distribution (first year of distribution)
9,000+ students
Students trained in STEM, robotics and AI since 2017 (organisation-wide, not specific to book programme) (2017-2024)
RobotsMali — AI-Generated Bambara-Language Children's Books After Mali's Language Policy Shift

Details

Maturity
Scaling
Promoter
RobotsMali
Period
2023–ongoing
Keywords
generative AI, local-language publishing, literacy, machine translation, primary education

Context

RobotsMali, a Bamako-based education non-profit co-founded by Michael Leventhal and Seydou Katilé, began as a robotics and AI-literacy programme in 2017 and has since trained more than 9,000 Malian students in STEM, robotics and AI. When Mali's military government ended French's 63-year status as the country's sole official language in June 2023, replacing it with 13 national languages including Bambara, the organisation pivoted to address an acute shortage of reading material in local languages in a country where illiteracy stands near 64%.

Objectives

To rapidly produce culturally grounded, Bambara-language children's books and draft textbooks to fill the local-language reading-material gap created by Mali's 2023 language-policy shift.

Activities

Using a pipeline of ChatGPT for text drafting, Google Translate and human translators for Bambara conversion, and the Playground AI tool for illustrations, RobotsMali produced 107 culturally grounded children's books in under a year, and separately generated 60 draft textbooks in an eight-week sprint as a proof of concept. The project is supported by Mali's Ministry of Education alongside UNESCO, the World Bank, Google and the Bill & Melinda Gates Foundation.

Results

The 107 books were distributed to elementary schools and had reached over 300 pupils within the first year.

Conclusions

RobotsMali's own leadership is candid about the technology's limits: co-founder Michael Leventhal has publicly noted that generative AI models frequently produce 'hypersexualized' depictions of African people and 'Eurocentric' imagery unless heavily corrected with negative prompts, and that African languages remain severely underrepresented in the datasets underlying these models, introducing grammar, idiom and cultural-nuance errors that require rigorous human review before books reach classrooms. There is no independent evaluation yet of reading-outcome gains — the evidence to date is output volume and distribution reach, not measured literacy impact.

Implementation
Indicative cost
Medium (€50k–€500k) — No specific budget figures are published, but the project draws on backing from multiple institutional funders (UNESCO, the World Bank, Google, the Bill & Melinda Gates Foundation) alongside Mali's Ministry of Education, suggesting a moderate multi-donor funding base; treated as a conservative 'medium' estimate pending disclosed costs.
Time to results
Short (< 1 year) — The book-production pivot itself moved fast — 107 books in under a year and 60 draft textbooks in an eight-week sprint — even though the parent organisation has operated since 2017.
Staffing & skills
RobotsMali co-founders (Michael Leventhal, Seydou Katilé) and core team, Human translators for Bambara conversion, Support from Mali's Ministry of Education, UNESCO, the World Bank, Google and the Bill & Melinda Gates Foundation

Conditions for success

  • An AI production pipeline (ChatGPT for drafting, Google Translate plus human translators for Bambara, Playground AI for illustrations) that pairs automation with mandatory human review
  • Timing the pivot to a specific policy trigger (Mali's June 2023 shift to 13 national languages) that created acute institutional demand
  • Multi-donor backing (government ministry plus UNESCO, World Bank, Google, Gates Foundation) rather than reliance on a single funder

Common failure modes

  • Generative AI models frequently produce 'hypersexualized' or 'Eurocentric' depictions of African people unless heavily corrected with negative prompts, per the co-founder's own public statements
  • African languages remain severely underrepresented in underlying model training data, introducing grammar, idiom and cultural-nuance errors that require rigorous human review before publication
  • No independent evaluation of reading-outcome or literacy gains has been published — evidence is limited to output volume and distribution reach

Where it fits

Governance type
NGO with government ministry and international-agency partnership
Scale
national, early-stage distribution
Income level
low income

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful