The Linguistic Colonialism of Default Languages
The English language has long been a weapon of colonialism. But in our digital era, where the embrace of decolonial thinking in shaping ongoing advances in artificial intelligence is championed, exclusionary language and the practices of linguistic colonialism are often still present (Mohamed et al., 2020). ChatGPT4 boasts availability in fifty languages, yet none of them are the predominant indigenous languages of African nations (Botpress, 2023). As a remnant of the colonial era, English is still the most common language on the continent, yet Swahili, Yoruba and Igbo are not supported. MidJourney is almost exclusively English-language based, and struggles even with European languages such as German (Stelzel, 2023).
Generative AI platforms largely reflect the personal identities and languages of their creators, and still primarily default to English, embedding existing culture and hierarchies of power into their systems. They often exclude to the point of discrimination as models begin to train themselves on their own outputs over time (Franzen, 2023). They magnify mistakes, but also misrepresent and misunderstand less popular or common data. The irreversible scaling of discriminatory defects inside of these systems transcends the problems of simply defaulting to English.
However, if we motivate Agre’s vision of a critical technical practice of AI (Agre, 1997), we afford ourselves the foresight and opportunity to examine and catch these practices in the models before they scale beyond our control. It provides a necessary explicability (Floridi & Cowls, 2019) and encourages decolonialist practices towards a more inclusive view of the future.
References:
Agre, P. (1997). Toward a critical technical practice: lessons learned in trying to reform AI. In Bowker, G., Star, S., Gasser, L., Turner, W. (Eds.) Social science, technical systems and cooperative work: beyond the great divide, psychology press. pp. 131–157.
Berberich, N., Nishida, T & Suzuki, S. (2020). Harmonizing Artificial Intelligence for Social Good. Philosophy & Technology. [Digital File]. Retrieved from: https://doi.org/10.1007/s13347-020-00421-8.
Botpress Community. (2023). List of languages supported by ChatGPT. Botpress.com. Retrieved from: https://botpress.com/blog/list-of-languages-supported-by-chatgpt.
Floridi, L., Cowls, J., Beltrametti, M., Chatila, R., Chazerand, P., Dignum, V., Luetge, C., Madelin, R., Pagallo, U., Rossi, F., et al. (2018). AI4People—an ethical framework for a good AI society: opportunities, risks, principles, and recommendations. Minds and Machines, 28(4), 689–707.
Franzen, C. (2023). The AI feedback loop: Researchers warn of ‘model collapse’ as AI trains on AI-generated content. Venturebeat.com. Retrieved from: https://venturebeat.com/ai/the-ai-feedback-loop-researchers-warn-of-model-collapse-as-ai-trains-on-ai-generated-content/.
Mohamed, S., Png, M. T. & Isaac, W. (2020). Decolonial AI: Decolonial Theory as Sociotechnical Foresight in Artificial Intelligence. Philosophy & Technology. [Digital File]. Retrieved from: https://doi.org/10.1007/s13347-020-00405-8 .
Stelzel, P. (2023). Midjourney tested in foreign languages. Medium.com. Retrieved from: https://philippstelzel.medium.com/midjourney-tested-in-foreign-languages-ac60053bcadb.
