After looking at all the ethical issues inherent with the creation of the leading LLMs, this page represents one alternative: grassroots AI.

Watch the short presentation from Professor Vukosi Marivate from the University of Pretoria and Lelapi AI, a South African company focusing on AI in African languages.

I learned about Professor Marivate when I was preparing to give a workshop at Indiana University for the African Language Teacher's Association annual conference. The organizers were interested in the question of what instructors of less commonly taught languages can do to improve tools like ChatGPT's performance in their language. (Note that in the world of machine learning they use the term low-resourced languages, which focuses on the global availability of digital resources, particularly high-quality resources that are needed to train language models). In the immediate term the answer to that question is "not much," for reasons that should be apparent by now: most of the LLMs' "learning" takes place during their pre-training, so we can't teach them new tricks.

Improvement in AI Performance in Low-resourced Languages

Two longer-term answers to that question: 1) AI models are likely to continue getting better over time, including in low-resourced languages, and 2) smaller, more grassroots organizations can work together on specific languages, as illustrated in Professor Marivate's video. This could mean either developing their own AI tools or creating high-quality data sets that can be shared with existing AI companies.

I'm desperately trying to get Professor Marivate to give a webinar on this question, which would be open to everyone, but he's very busy and in high demand. For now, please take a look at at least the first four pages of Participatory Research for Low-resourced Machine Translation.pdf, specifically the Introduction and 3.0 The Machine Translation Process. This section describes the various roles needed to develop or improve machine translation tools (very similar to that of generative AI) and highlights the role people knowledgeable and passionate about particular languages (like language teachers) could play. But that leads to several more challenges--finding the right group, learning how exactly you could contribute, and volunteering a lot of work where you may or may not see much tangible benefit. These are the types of questions I'm hoping Dr. Marivate could help us learn about.

The document above was referenced in the video above and can be found on the Publications page of the Masakhane website.