On the Use of Artificial Intelligence.

1. Why this guidance exists

Isomo's mission is to unlock Rwanda's human potential by using English learning as a vehicle for critical thinking, confidence, and 21st-century skills. Artificial Intelligence (AI) --- through tools such as IsomoBot (an AI learning companion), the VoiceBot - Ijwi (a speaking and listening coach), the AI Hour, and AI Bootcamps — offers a genuine opportunity to amplify the reach and quality of that work. Yet, AI also introduces real risks: inaccuracy, bias, privacy and data harms, over-reliance, and the danger of widening - rather than closing - the digital divide.

This document sets out how everyone at Isomo should think about and use AI responsibly. It is guidance, not a rigid rulebook: it states principles, gives practical do's and don'ts, and points to where to seek advice. It is inspired by responsible AI frameworks and evidence on AI in education, adapted to Rwanda's real-world context.

It sits between Isomo's AI Strategy (what we are building and why) and any future binding policy (enforceable rules). Where this guidance and a future formal policy or the law within applicable jurisdiction conflict, jurisprudence and executive order prevail.

2. Grounding frameworks

This guidance draws on:

1. the Government of Rwanda's National Artificial Intelligence Policy (MINICT, 2022), which sets Rwanda's framework for ethical, inclusive, and skills-led AI adoption - and against which Isomo intentionally aligns its AI work, particularly in AI literacy, equitable access, and locally-grounded content;

2. the U.S. Department of Defense AI Ethical Principles (2020): Responsible, Equitable, Traceable, Reliable, Governable - a widely cited, plain-language backbone for responsible AI;

3. the University of London AI Policy (UOL04, 2024) and the Russell Group principles on generative AI in education, which balance benefit-maximization with ethics, transparency, and academic integrity;

4. the Texas Christian University AI Policy (2026), for concrete operational controls on data, disclosure, and oversight;

5. UNESCO's guidance on generative AI in education and research (2023), which centers a human-centered, equity-first approach for low-resource settings;

6. the learning-science evidence that AI tutors and chatbots can improve language-learning outcomes, engagement, and confidence when well designed and supervised.

(Full references in Section X.)

Who this applies to

This guidance applies to everyone acting on behalf of Isomo when they use AI for Isomo's educational, operational, research, or administrative work:

7. Staff and facilitators - teachers, program staff, technical team, volunteers, and contractors who build, configure, deploy, teach with, or administer AI tools.

8. Students and learners - participants in the Academy, Circles, AI Hour, AI Bootcamps, and users of IsomoBot and the Ijwi.

Shared principles (Section 3) apply to all. Audience-specific guidance follows in Sections V (staff/facilitators) and VI (students/learners).

3. Our foundational principles

Isomo adopts the following principles for any use of AI. They adapt the DoD's five principles and the University of London / Russell Group principles to our mission and context.

1. Human-centered and accountable (Responsible). AI supports people; it does not replace human judgment. A human is always accountable for decisions and outputs. Final decisions that affect a person - a student's grade, progression, or opportunity; a hiring or disciplinary decision - must be made by a person, not by an AI.

2. Fair and equitable (Equitable). We take deliberate steps to minimize bias and to avoid creating or widening inequity. In the Rwandan context, this includes equity of access: AI must not advantage only those with devices, data, and connectivity.

3. Transparent and traceable (Traceable). We are open about when and how AI is used. Where AI materially contributes to content, analysis, or a decision, that use is disclosed. The data, configuration, and reasoning behind our tools should be documentable and auditable.

4. Reliable and well-scoped (Reliable). Each AI tool has an explicit, well-defined purpose. We test its safety, accuracy, and effectiveness for that purpose before relying on it, and keep testing across its life - especially for accuracy and cultural fit in the Rwandan curriculum.

5. Controllable and supervised (Governable). Every deployed AI tool has human oversight, the ability to detect unintended behavior, and a way to disengage or shut it down. We do not deploy AI we cannot monitor or stop.

6. AI-literate. We invest in the skills of staff and students to understand what AI can and cannot do, to evaluate its outputs critically, and to use it appropriately - treating AI literacy as a core learning outcome rather than an add-on.

7. Mindful of wider impact. We consider the social, environmental, and cost implications of AI - including the energy and money it consumes - and act proportionately, especially given the resource constraints of the communities we serve.

4. Responsible-use commitments (everyone)

These practical commitments apply to all staff, facilitators, and learners.

Keep a human in the loop

9. Use AI to assist - to draft, explain, practice, summarise, or give feedback - not to make final, consequential decisions on its own.

10. The person using AI remains responsible for the result, exactly as if they had produced it unaided.

Verify before you trust

11. AI can produce confident, fluent answers that are wrong, outdated, or invented (“hallucinations”). Check important information against a reliable source before using or sharing it.

12. Be especially careful with facts about Rwanda, the local curriculum, dates, names, statistics, and anything a student will be assessed on.

Be transparent

13. Disclose when AI has materially shaped your work (e.g., “drafted with AI assistance and then reviewed and edited”).

14. Facilitators should make AI's role visible to learners; learners should acknowledge AI use as their facilitator instructs.

Protect people's data and privacy

15. Do not enter personal, confidential, or sensitive information into public AI tools - this includes student names linked to records, health information, exam data, household details, financial information, staff HR data, or anything shared with Isomo in confidence.

16. Treat anything you would not post publicly as off-limits to a public AI tool.

17. Prefer Isomo-approved or contracted tools (with no-training and data-protection terms) for any work touching real Isomo data. See Section 7.

Watch for bias and harm

18. AI reflects biases in its training data and may produce stereotyped, culturally inappropriate, or unfair outputs. Review outputs for fairness, especially anything that affects or describes people.

19. Never use AI to create discriminatory, harassing, deceptive, or harmful content.

Mind the digital divide

20. When choosing or designing how AI is used in a program, assume limited connectivity and devices. Favour/Include low-bandwidth, offline-capable, or voice/SMS-accessible options so that AI extends opportunity to the many rather than concentrating it among the few.

Stay alert to AI-enabled security threats

21. AI makes phishing, fake voice/video, and scam messages more convincing. Be cautious with unexpected messages and report anything suspicious to Isomo IT/security contact.

5. Guidance for staff and facilitators

In addition to Section IV:

Building and deploying AI tools (technical team)

22. Define the purpose and the limits of each tool before launch (Reliable). Write down what it is for, what it is not for, and who is accountable for it.

23. Ground the AI in vetted, local content. Prefer retrieval-augmented generation (RAG) over fine-tuning where possible: it constrains the model to approved, Rwanda-relevant material, reduces hallucination, and is cheaper and easier to control. Where fine-tuning is used, still apply guardrails - no model is hallucination-free.

24. Design for the infrastructure that exists, not the one we wish we had: low bandwidth, intermittent connectivity, basic devices, high data costs. Realtime voice should use appropriate technology (not basic IVR), and we should actively pursue zero-rated data, telephony/SMS access, and offline modes.

25. Build in monitoring and an off-switch (Governable). Every tool needs a way to observe its behaviour, catch problems, and disable it. Consider a dedicated review function - human and/or a supervisory AI agent - to evaluate outputs, particularly during pilots.

26. Plan data storage and sovereignty deliberately. Decide where Isomo's data lives, who can access it, and how it is protected; document this. Treat long-run options (e.g., local compute / data hosting) as a strategic question, not an afterthought.

27. Document data sources, design choices, and configuration (Traceable) so others can understand, audit, and improve the system.

Teaching and assessing with AI

28. Set clear, course-level expectations. Tell learners, for each activity, whether and how AI may be used, and how to acknowledge it. (This mirrors how leading universities handle it - the institution sets the framework; the facilitator sets the rule for the task.)

29. Teach AI literacy explicitly. Cover how AI works, its limits, hallucination, bias, privacy, and responsible use - in the AI Hour and across programs. Decide as a team whether “AI literacy” means knowing about AI, creating value with AI, or both, and design activities and assessment to match.

30. Protect genuine learning. Guard against over-reliance: AI should increase practice and feedback, not replace the student's own thinking and production. Consider structured approaches, such as AI-free activities or terms alongside AI-allowed ones, so that foundational skills are built before tools are relied on.

31. Keep oversight of AI feedback to students. AI tutoring (IsomoBot, Ijwi) should be supervised by teaching staff; spot-check its feedback for accuracy and tone, especially early on.

Using AI for operations, admin, and proposals

32. Use AI freely for productivity (drafting, summarising, brainstorming) with non-sensitive information, applying Section IV throughout.

33. Do not paste donor-confidential, partner-confidential, staff, or student data into public tools.

34. Review AI-assisted external documents (proposals, reports) for accuracy and for any unsupported claims before they go out.

6. Guidance for students and learners

In addition to Section IV, in language for learners:

35. AI is a tutor and a practice partner, not a substitute for your effort. Tools like IsomoBot and Ijwi help you practice more, get feedback faster, and learn at your own pace. The learning still has to happen in your head.

36. Do your own thinking first. Use AI to check, explain, or extend your work - not to do the work for you. Copying AI answers without understanding them is cheating yourself, and may break your facilitator's rules.

37. Follow your facilitator's instructions on when AI is allowed and how to indicate that you used it. When in doubt, ask.

38. Don't believe everything the AI says. It can sound sure and still be wrong. Check important facts, especially for assignments and exams.

39. Protect your privacy and others'. Don't share your full name, home details, contact information, ID numbers, or any private information about classmates with a public AI tool.

40. Be respectful. Never use AI to bully, deceive, cheat, or create harmful or inappropriate content.

41. Ask questions openly. It's good to ask whether and how to use AI. You won't be penalized for asking - you're expected to learn this.

7. Choosing and approving AI tools

42. Prefer Isomo-approved tools for any work involving real Isomo, student, staff, or partner data. Approved tools are those reviewed for data protection, no-training-on-our-data terms, and fitness for purpose.

43. Public AI tools (those with no Isomo contract) may be used for general, non-sensitive productivity and learning - never for confidential or personal data.

44. Before adopting or buying a new AI tool or service (even a free one, and including new AI features added to tools we already use), check it against this guidance and raise it with the team AI lead for review.

45. When you can, opt out of letting tools train on your inputs.

As Isomo grows, we'll consider maintaining a simple published list of Approved / In-review / Not-approved tools and what data each may handle.

Interim arrangement. Until an AI lead is nominated (see Section VIII), tool review requests and approval questions should be routed to the Isomo IT so that no decision is blocked while the role is being filled.

8. When to seek advice, and how we govern this

Seek advice before you proceed if you are about to:

46. adopt, buy, or build a new AI tool or service;

47. use AI with personal data, or with any confidential Isomo/partner/donor information;

48. introduce a material change to how AI is used in teaching or assessment;

49. use AI in research or anything that may be published externally.

Oversight (to be confirmed by Isomo leadership). We recommend a small, named AI oversight function - e.g., a designated AI lead plus a lightweight review group drawing on teaching, technical, and leadership perspectives - responsible for:

50. maintaining and interpreting this guidance and reviewing it at least annually;

51. approving tools and reviewing new AI use cases;

52. monitoring outcomes, accuracy, equity of access, and incidents;

53. reporting to Isomo leadership.

Nominating the AI lead. The AI lead will be nominated by the Isomo team within 60 days of this guidance being approved, for an initial one-year term, reviewable annually. The nomination is open to any staff member or facilitator with demonstrated engagement in Isomo's AI work; the team will agree the nomination process (self-nomination, peer nomination, or both) before opening it. Until the nomination concludes, the interim routing in Section VII sustains.

If something goes wrong (e.g., personal data entered into the wrong tool, harmful output, a tool misbehaving): stop, and report it promptly to the AI Lead so it can be contained and learned from.

9. How we'll know it's working (monitoring)

We will review our use of AI against these questions:

54. Learning: Are AI tools measurably improving English proficiency, confidence, and digital/AI literacy? (Define metrics - e.g., test-score gains, speaking practice hours, AI-literacy assessments.)

55. Equity: Are the tools reaching learners with limited devices and connectivity, or only those already connected?

56. Quality & safety: Rates of inaccurate or inappropriate outputs caught; incidents and how they were handled.

57. Responsible use: Are disclosure, human-oversight, and data-protection commitments being followed?

Open questions to resolve as we pilot (from Isomo's strategy work): how exactly to define and assess AI literacy; whether to run AI-free vs AI-allowed terms; and the right metrics for each.

10. References

58. Government of Rwanda, Ministry of ICT and Innovation (MINICT). The National Artificial Intelligence Policy (April 2022; published April 2023). https://www.minict.gov.rw/fileadmin/user_upload/minict_user_upload/Documents/Policies/Artificial_Intelligence_Policy.pdf

59. U.S. Department of Defense, "DOD Adopts 5 Principles of Artificial Intelligence Ethics" (2020). https://www.defense.gov/News/News-Stories/Article/Article/2094085/ - and the underlying Defense Innovation Board, AI Principles: Recommendations on the Ethical Use of AI by the DoD (Oct 2019).

60. University of London, Artificial Intelligence (AI) Policy (UOL04, v1.2, 2024). https://www.london.ac.uk/sites/default/files/artificial-intelligence-policy-uol.pdf

61. Russell Group, Principles on the use of generative AI tools in education (2023).

62. Texas Christian University, Artificial Intelligence Policy (approved 14 April 2026).

63. Harvard University, Guidelines for Using ChatGPT and other Generative AI tools (Office of the Provost / HUIT). https://provost.harvard.edu/guidelines-using-chatgpt-and-other-generative-ai-tools-harvard

64. UNESCO, Guidance for generative AI in education and research (2023). https://www.unesco.org/en/articles/guidance-generative-ai-education-and-research

65. Serge, G., Oyebimpe, A., & Andala, H. O. (2021). Relationship between Teachers' Competency Level in Teaching English Language and Students' English Language Proficiency in Secondary Schools in Rwanda. Journal of Education, 4(7), 104–122. https://doi.org/10.53819/81018102t5030

66. Evidence on AI chatbots/tutors and language-learning outcomes (see ScienceDirect reviews cited in the Isomo AI Strategy, e.g., S2666920X24000316 and S2215039025000086).

Annex A - Alignment with Rwanda's National AI Policy (MINICT, 2022)

This guidance is designed to operate within, and contribute to, Rwanda's National AI Policy. Isomo's AI work delivers against the following activities in the policy's Implementation Plan:

Priority Area 1 - 21st Century Skills & High AI Literacy

67. Activity 12 (develop AI/ML/data modules for integration into the national curriculum): Isomo's AI Hour curriculum and AI Bootcamp materials shall be designed to be compatible with this integration path.

68. Activity 13 (Train-the-Trainer Program in AI skills): Isomo's facilitator training in AI tools and AI literacy shall contribute directly to upskilling teachers in both IT and non-IT subjects.

69. Activity 14 (align primary/secondary curriculum with the '21st-century curriculum'): Isomo's Academy and Circles programs shall embed AI literacy as a 21st-century skill alongside English, in line with this activity.

70. Activity 21 (public awareness campaign on AI - benefits and risks): Isomo's AI Hour, AI Bootcamps, and public communications shall extend AI literacy beyond enrolled learners into the wider Rwandan and global community.

Priority Area 6 - Practical AI Ethics Guidelines

71. Activity 67 (promote and advertise Rwanda's Guidelines on the Ethical Development and Implementation of Artificial Intelligence): Isomo will align this guidance with Rwanda's national AI ethics guidelines as they are finalized and published by RURA/MINICT. Where the national guidelines and this document conflict, the national guidelines prevail.

72. Activity 71 (annual participatory industry & society consultation forum): Isomo will participate where invited, and feed back lessons from its programs.

Isomo will review this alignment annually alongside this guidance.

Bracketed items (approval body, AI lead, interim contact, nomination deadline, incident contact)