a laptop is not a lesson

Problem statement

For two decades, the headline answer to "how do we help marginalised children learn?" has been to ship them devices — one laptop or tablet per child. The randomized evidence is now unusually clear and unusually disappointing: hardware distribution on its own does not raise learning. In Peru's One Laptop per Child rollout across 318 rural primary schools, the program drove computers-per-student from 0.12 to 1.18 and sharply increased usage — yet produced no effect on math or reading test scores after 15 months. A separate randomized home-laptop trial in Lima found children got much better at using the laptop, but showed lower academic effort (as reported by teachers) and no gains in achievement or cognitive skills. Long-run follow-up found negative effects on completing primary and secondary school on time. The pattern repeats across Romania, Uruguay, Honduras, Costa Rica, and China: device-led programs cluster around zero or negative learning impact.

Why this matters

Roughly 250+ million children are out of school and hundreds of millions more are in school but not learning to read or do basic math — a "learning crisis" concentrated among exactly the marginalised learners EdTech is sold to serve. Device programs are enormous line items: ministries and donors have spent billions on tablets and laptops on the implicit theory that access equals learning. When that theory fails, the cost is not only wasted money but opportunity cost — the same budget could have funded interventions with proven effects — and a credibility hit that makes the next (possibly good) EdTech proposal harder to fund. Getting the design principle right is therefore high-leverage: it governs how the next wave of investment, including AI tutors, is spent.

What’s been tried and why it hasn’t worked

The dominant model — procure devices, distribute them, hope learning follows — fails because it treats hardware as a learning input rather than as a delivery channel for instruction. The randomized trials isolate the mechanism: usage went up, but there was no instructional content tied to the curriculum, no structured pedagogy, and no integration with what teachers were doing — so children used machines without learning more. Worse, in several settings the device displaced instructional time or effort (the Lima home-laptop "lower academic effort" finding), making outcomes neutral-to-negative. By contrast, the EdTech approaches that do show gains are the ones built around an instructional model: computer-assisted learning that adapts to a child's level ("teaching at the right level"), structured-pedagogy programs, and lightweight channels (SMS, audio, interactive radio/IVR) that support rather than replace teachers — and even these only deliver after iterative adaptation. The lesson is precise and counterintuitive: the constraint is pedagogical integration, not connectivity or device count.

What would unlock progress

Progress comes from inverting the design: start from a validated instructional model — adaptive practice at the learner's level, teacher-complementary content, formative feedback — and treat the device as the cheapest channel that delivers it, often offline and on basic hardware. The open challenges are (1) making adaptive, curriculum-aligned content that works offline on the low-end phones marginalised families actually own, (2) designing it to augment a teacher's lesson rather than pull children into solitary screen time, and (3) measuring real learning gains in low-connectivity settings where assessment data is hard to collect. Newly tractable because cheap smartphones, offline-first platforms (e.g., Kolibri), and low-cost adaptive engines now exist where they didn't during OLPC.

Entry points for student teams

A team should pick one narrow learning outcome (e.g., early-grade numeracy or decoding) and prototype an offline, teacher-complementary adaptive tool for a low-end phone — building on open offline-first infrastructure like Kolibri (learningequality.org) and openly published early-grade assessment tools like ASER's reading and arithmetic instruments (asercentre.org) — explicitly designing with teachers so the tool fits a real lesson rather than replacing it. The classroom learning-gain trial itself is out of semester reach: it takes a school partnership, an IRB covering minors, deployment, and an outcome window longer than a term, all owned by the partner school or NGO — so the semester deliverables are the co-designed prototype with teacher usability evidence, plus the trial's design: a pre-registered A-B evaluation plan with instrument and protocol the partner could run unchanged. Relevant disciplines: learning science / pedagogy, mobile app development (offline-first), data/measurement and experimental design, and human-centered design with teachers and students. The win condition is a tool a teacher has actually rehearsed inside a real lesson plus a trial package ready to measure the gain — demonstrating the "instruction-first, device-as-channel" principle the evidence demands.