In a short open access article in the journal AI & Society, Muhammad Surajo Ibrahim critiques “The literacy mirage: how generative AI bypasses rather than builds informational capacity in the Global South“. The piece is published as a letter in the journal’s intriguingly titled Curmudgeon Corner: “a short, opinionated, column on trends in technology, arts, science and society, commenting on issues of concern to the research community and wider society”.
Ibrahim sets out to criticise the idea that generative AI will provide easy access to information and knowledge for the “information marginalised”, in those parts of the worlds lacking adequate educational systems, library services, and information and IT infrastructures. Ibrahim comments that this is “a seductive vision, one that donors, ministries, and development agencies have found easy to embrace. I think it is dangerously wrong.”
He points out an ability to deal effectively with information and knowledge requires time consuming practice in dealing with sources. Information skills are “forged through the friction of searching, failing, questioning, and revising. Remove the friction, and you remove the pedagogy”. Generative AI, by design, removes this friction, providing easy answers, and avoiding the need for understanding of, and critical engagement with, the information environment.
This is a common enough argument in the developed world, but it has a particular resonance for the Global South. In the developed world, an uncritical reliance on AI generated material is likely to challenged by teachers, lecturers, librarians, and others; in less well-resourced parts of the world, such checks are less likely to happen. Furthermore, Ibrahim points out, much of the Global South relies on a rather traditional style of education, based on canonical texts, memorization, and a slow synthesis of knowledge. This intellectual tradition, whatever its limitations, can provide a valuable check on an over-reliance on AI, but is vulnerable to being displaced by a “shallow, algorithmic” approach. AI tools themselves are not the problem; rather the problem is that they are being claimed as a substitute both for an existing intellectual capacity, and also for investment in wider information infrastructures and in information literacy development.
Ibrahim states the issues bluntly: “the ‘AI for development’ community would do well to attend carefully to what it is actually proposing. It is proposing to insert a highly confident, frequently hallucinating, culturally Anglophone, and epistemologically opaque system between populations and the very cognitive processes through which those populations might build the capacity to interrogate such systems”.
An insightful piece, raising issues which should be contemplated anyone concerned about the implications of wider take-up of generative AI worldwide.
The Occasional Informationist 