Anirudh Week 1: An Analysis of the Modern Extension of Colonial Speech

Image credit: The Penn Review

In his 1986 collection of essays Decolonising The Mind, Ngũgĩ wa Thiong'o posits that the greatest weapon of colonialism wasn’t the rifle. It wasn’t the cannon nor the gunboat. It was the weaponization of language. He vividly describes the systemic tactics colonial powers used to strip down the native tongues of colonized children in favor of English, severing the children’s connections to their cultures, identities, and even self-worth. And even though these colonial empires have collapsed, the inherent framework of identity erasure has been fundamentally upgraded. Today, the maxim of colonial erasure has been repackaged by the algorithms of Artificial Intelligence.

Throughout the history of AI—whether we see it as a tool of objectivity or question it with suspicion—we cannot ignore that these systems are at their core built on data. Specifically however, they are built with massive datasets scraped from the Western, English-speaking internet. It is fundamentally tied to the biases of the Western canon. Thus, Artificial Intelligence models don’t simply process data, but amplify these existing systemic biases. 

So when we consult an LLM to proofread our work, or utilize generative AI in drafting an email, it prioritizes a standardized flavor of Westernized syntax. Regional dialects, cultural idioms, and non-Western styles of rhetoric are flagged as being “incorrect” or being of “low quality”. 

Just like how Thiong'o describes the punishments students faced when they wrote in their mother tongue, the modern digital world subtly coerces us to flatten our linguistic identities, as to be deemed “visible and valued” by the algorithm, one must write like the algorithm. Thus, we aren’t teaching machines to talk like each one of us, but rather teaching them to dictate who we are allowed to be.