Tool Transfer Is Not Capacity: When a Partnership Becomes a Bridge

In a randomised trial across 319 rural primary schools in Peru, the number of computers per pupil rose from 0.12 to 1.18 after the One Laptop per Child programme, and their use grew at school and at home. After 15 months the researchers found no effect on enrolment or on mathematics and language scores, though they did find positive effects on some general cognitive measures. The Inter-American Development Bank published this in 2012. The device had arrived; the hoped-for result had not, within the period measured.

The tool and the capacity

What happened in Peru is not a verdict on technology but a reminder of the difference between transferring a tool and transferring capacity. A tool is a ready-made device or model. Capacity is the ability of the people on the ground to understand what they use, adapt it to their setting and maintain it when support ends. This is close to what the World Bank concluded in its 2016 report on "digital dividends": digital technologies spread quickly, but their broader benefits, in growth, opportunity and services, lagged behind. Closing the connectivity gap is necessary but not sufficient, for countries need "analog complements": regulation that encourages competition, skills suited to the new economy, and accountable institutions. The article "Beyond Connection: How Does AI Widen an Old Divide?" sets out the layers of the divide.

When people from the setting take part

At the other pole stands a research paper on machine translation for low-resource African languages. Its authors proposed "participatory research": involving everyone the development process needs, since translation researchers cannot alone solve the shortage of resources. The approach produced new datasets and performance benchmarks for more than 30 languages, with human evaluation for about a third of them, and scientific contributions from participants with no formal training. The data, models and code were published in an open repository, which bears the name of the Masakhane research community. This is one experience in one field and should not be generalised, but it shows that someone who lives a language may build what an outsider cannot.

Who owns the data

UNESCO's Recommendation on the Ethics of Artificial Intelligence, adopted by 193 states in November 2021, provides for protecting privacy throughout the life cycle of a system, and for data use that accords with international law and the sovereignty of states. We have no evidence of bad faith in partnerships between technology companies and nonprofits. But long reliance on a single platform, and the fate of beneficiaries' data, are legitimate questions whose answers belong in the contract before the reports. The article "Board the AI Train While the Tickets Are Free Today" examined the offers available to organisations and their hidden price.

Clauses worth writing into the agreement

This is an editorial distillation of the evidence above, not requirements stated in it:

• A written goal of building local capacity, measured by the number of local people able to run and modify the system, not by the number of devices distributed.
• Participation of people on the ground in the design from the first day, not at handover.
• An explicit clause on data ownership and the organisation's right to understand the models applied, with a preference for open solutions where possible; see "Open Models: Genuine Openness or a Slogan?".
• Funding that extends beyond the first year, and a timetable for transferring operation to the organisation.

A bridge whose key is not handed to those on the far bank remains the property of whoever built it. So ask of any partnership: what will remain when the partner leaves? Keeping faith in what we receive and what we give is among the causes of blessing, God willing.