Document Type
Working Paper
Publication Date
10-29-2026
Abstract
Solar microgrids in sub-Saharan Africa routinely fail to generate the productive demand needed to service their investment, because the capability to convert electricity into economic activity does not exist at electrification. We report a 15-month deployment in Oloibiri, Nigeria, a community whose productive base was displaced by oil extraction, testing whether AI-facilitated learning can build that capability at a cost compatible with microgrid investment.
The Davidson AI Innovation Center enrolled 104 learners (136 accounts) with zero prior computer access. They recorded 4,134 learning sessions, and 25 individuals earned 90 certifications on unscaffolded work, for just over $20,000 in cumulative investment, about $220 per certification. Four findings bear on electrification investment. First, capability formation is visible in the interaction data: across 1,984 transcripts from 48 learners with fixed effects, the AI’s share of conversation words falls 0.079 per 90 days (p < 0.0001), not because the AI writes less but because learners write far more. AI output grows 117 words per session per 90 days; learner output grows 344, three times faster, in fewer turns. Across cumulative session bands, requests for AI clarification fall 70%. Second, capability converted into service: 34 learners delivered 332 advisory consultations (229 fully documented) to 191 community members across eight communities, and learners moved into paid technical work and paid educator roles. Third, deep users account for a disproportionate share of compute cost, a success signal rather than a sustainability problem, and model tiering cut per-session compute cost by 95%. Fourth, facilitator dependency is real but bounded: engaged learning fell to zero during the facilitator’s January–February 2026 absence but continued through a later absence once successors existed. The evidence also has limits: assessed skill scores rose only among the ten learners with 100 or more sessions, and a repeated task showed no change, so retention is the binding constraint.
The minimum viable investment is not a grid connection. It is four to six computers, a satellite subscription, and a community-embedded facilitator committed to producing their own successors.
Keywords
productive use of energy, rural electrification, capability approach, AI-facilitated learning, off-grid communities
Disciplines
Engineering
eCommons Citation
Hallinan, Kevin; Hao, Lu; Davidson, Bennywhite; Morris, Divinegift; Onuahu, Confidence; and Clergy, Silas, "Education Before Electricity: AI-Facilitated Capability Formation as a Pre-Condition for Productive Energy Use in Off-Grid Communities" (2026). Mechanical and Aerospace Engineering Working Papers. 2.
https://ecommons.udayton.edu/mee_wk_papers/2
