African language processing startups lost investor funding after Meta announced No Language Left Behind in 2022, a translation model covering 200 languages including 55 African languages. Investors concluded Meta had solved the problem and small startups couldn't compete, said Timnit Gebru, director of Distributed AI Research Institute.
OpenAI representatives then approached similar organizations claiming OpenAI would make them obsolete in their languages, offering minimal payment for their data. "They basically threaten them by saying, 'OpenAI is going to put you out of business soon,'" Gebru said. "'You're better off collaborating with us and supplying us data for which we're going to pay you peanuts.'"
The pattern extends beyond Africa. Big Tech's universal models systematically displace specialized tools built by marginalized communities worldwide, despite local organizations having deeper linguistic expertise and lower computational costs. When companies announce models covering hundreds of languages, investors withdraw from regional startups.
Abeba Birhane, also at Distributed AI Research, argues 'AI for good' framing serves as corporate deflection. "It allows companies to say 'Look, we're doing something good! Everything about AI is not bad. And you can't criticize us,'" she said, referencing grassroots movements resisting or refusing AI systems.
Gebru characterizes mainstream AI development as "stealing data, killing the environment, exploiting labor" to build universal models. Training large models across hundreds of languages requires far more resources than targeted tools serving specific communities.
Distributed AI Research advocates empirically-grounded policy over corporate promises. Their framework prioritizes direct funding for grassroots AI communities rather than universal solutions that concentrate power in tech giants while claiming to help underserved populations.
The critique challenges corporate-dominated AI ethics discourse. Instead of accepting that universal models benefit everyone, researchers document structural harms: displacement of local expertise, extraction of community data for minimal compensation, and environmental costs of massive models. They call for resource-efficient specialized models serving communities directly.