Return to the DJAIRI portal
Mustanef

Ideas, conversation & futures

Return to the question. Open a horizon of understanding.

Menu

Readings & analysis

العربية — المادة المقابلة

24 original readings, each with an example, exercise and review rubric. We distinguish analysis from evidence and retain the earlier opening articles.

Why a Country Needs an AI StrategyBuying disconnected tools does not determine who owns decisions or funds maintenance. A useful strategy connects public needs with delivery capacity and accountability rather than counting mentions of AI.Small AI: Smaller Size or a Clearer Task?Small can describe parameter count, a narrow task or a local deployment. These differ. A smaller model is not automatically better, cheaper or independent of a supplier.Rwanda: Learning from Policy without Declaring SuperiorityA national policy lets us examine choices. It does not prove a country ranking or the success of every application. Separating ambition from delivery makes comparison useful.From National Readiness to a Small Organisation’s DecisionA national indicator does not automatically describe a school or association. Connectivity may be strong while permission to use a task’s data is missing. Local decisions need local evidence.Reading an Indicator before Ranking CountriesIndicators can have similar names but different units, years and samples. Combining them into a ranking can hide gaps or penalise missing data.Missing Data Are Not Evidence of Missing CapacityAn absent published number may reflect collection practices or delay. Filling the gap with a plausible value turns uncertainty into artificial certainty.Data Governance Starts with the Minimum NecessaryMore collection does not always improve understanding. Unnecessary data increase protection burdens and risks without helping the task.Human Review Needs Authority, Not Just a SignatureSomeone may sign an automated result without time or authority to change it. That does not make review effective or provide meaningful appeal.AI Skills: From Attendance to Demonstrated PracticeAttendance counts do not show what learners can do. Capability requires a task and rubric distinguishing polished responses from sound decisions.Local Language Is a Test Criterion, Not DecorationA translated interface does not prove understanding of dialect or local terminology. Errors may involve place names, units or respectful phrasing.A Low-Connectivity School: Start with What Remains AvailableA plan requiring constant connectivity may exclude the very school it serves. Availability is part of learning value, not an afterthought.Health: Keep Educational Trials Separate from DiagnosisA confident model response lacks patient examination and clinical responsibility. An information-organising exercise is not diagnosis.Water and Agriculture: Calculate before Claiming PredictionA learning equation can explain water quantity without proving field prediction or the suitability of an agricultural intervention. Inputs and limits are part of the result.Women and Access: Ask Who Cannot Reach the ServiceMore users do not prove that a gap has closed. The sample may change while connection cost, time and language still block some groups.Procurement: Buy the Ability to Exit TooAn entry price can hide migration, training, updates and shutdown costs. Without an exit plan, replacement becomes difficult even when benefits fail to appear.Open Source: A Right to Use That Still Needs ReadingAvailable code or weights do not make every dataset, mark or use unrestricted. Local deployment can still involve legal and technical responsibilities.Total Cost: Saved Time Is Not the Whole BenefitIf a system saves drafting time but increases correction time, its value changes. Honest calculation includes review, errors, setup and maintenance.A Local Test before ScalingScaling a small trial before examining failures spreads errors with the service. Case count alone does not establish representativeness.Fair Partnership: What Remains after Support Ends?A partnership may begin with equipment and training but leave unaffordable costs or dependencies. Local value depends on lasting capability, rights and responsibilities.Funding: A Fund’s Name Is Not Proof of EligibilityA climate or development fund does not fund every AI proposal. Eligibility, sector, country, partners and application route are separate conditions.Foresight: Scenarios, Not PropheciesPredicting one future can make an organisation fragile when cost, law or connectivity changes. A scenario helps when it leads to a revisable decision.Climate and Energy: Do Not Generalise One Model’s CostTraining and operation resources depend on size, hardware, task and usage. A figure from one experiment cannot describe every AI system.A Ninety-Day Plan: A Working Template, Not a Results PromiseA useful schedule assigns responsibilities and checks. It does not guarantee that every organisation succeeds in the same period; capability and authority differ.How to Read an AI News ClaimNews may announce a model, policy or trial; these are not equivalent events. A success headline may rely on a statement rather than published evaluation.

Earlier readings