Today, ARTICLE 19 launches Techtonic, a brand new podcast series looking at freedom of expression online and the human impact of Big Tech and social media.
Over the course of the series, leading tech reporter Chris Stokel-Walker will speak to experts around the world, including former insiders at Big Tech companies , digital rights activists, scholars and journalists at the sharp end of artificial intelligence, deepfake technologies, and internet shutdowns.
For the first episode, ‘AI’s English language problem’, we look at the language biases inherent in AI technologies with guests Aliya Bhatia and Gabriel Nicholas from the Center for Democracy and Technology.
Aliya and Gabriel recently authored Lost in Translation, an analysis of the use of large language models in the context of non-English content.
Chris, Aliya and Gabriel discuss how large language models work, the underrepresentation of non-English-language nations in training data and the effects of an AI trained largely in English.
As large language models are mainly trained in English, what impact does this have on communities in India that speak Telugu or Malayalam, or on Catalan speakers in Spain, with translations often suffering from anglo-centric bias, and prone to mistakes? How might companies redress the balance to ensure a diversity and quality of data that cover different cultural contexts?
Techtonic is produced by Christopher Hooton and Nicola Kelly.
Subscribe to Techtonic wherever you get your podcasts to hear new episodes, released fortnightly on Mondays.