TLDR: A recent Aeon essay highlights that generative AI models currently access only a narrow portion of human knowledge, predominantly Western and institutional epistemologies. This limitation risks amplifying existing knowledge hierarchies, marginalizing diverse ways of knowing, and potentially disconnecting future generations from vast bodies of traditional wisdom. The article suggests that while AI developers might view this as a solvable data problem, the complexities of data sourcing and representation are far greater.
Generative Artificial Intelligence (GenAI) models, despite their advanced capabilities, operate with access to merely a ‘small slice of human knowledge,’ according to a recent essay published in Aeon. This fundamental limitation carries significant implications for how future generations will interact with and perceive information, potentially leading to a profound disconnection from diverse forms of human understanding.
The essay argues that current GenAI systems are far from neutral, exhibiting a strong bias towards ‘dominant epistemologies,’ which are typically Western and institutional. This inherent bias marginalizes alternative ways of knowing, particularly those embedded in oral traditions, embodied practices, and ‘low-resource’ languages such as Hindi or Swahili, spoken by hundreds of millions globally. By inadvertently amplifying these existing knowledge hierarchies, GenAI risks contributing to the erasure of centuries-old systems of understanding.
Furthermore, the article points out that GenAI not only reflects but also has the capacity to intensify these hierarchies as human behavior adapts to its pervasive influence. The increasing integration of AI overviews into search engines and the rising popularity of AI-powered search platforms like Perplexity exemplify this shift. A September 2025 study, analyzing ChatGPT usage since its November 2022 launch, revealed that approximately half of user queries sought practical guidance or information. This study also indicated a change in user behavior, with U.S. Google users becoming less likely to click on search results when an AI summary was present. Previously, users would navigate multiple links to compare viewpoints and gather comprehensive information, a practice now being supplanted by AI-generated summaries.
The author illustrates this concern with a personal anecdote involving a family medical decision. When faced with a father’s tumor diagnosis, the author, relying on internet research, initially sided with Western allopathic medicine, dismissing traditional Siddha medicine. However, the father’s secret adherence to herbal concoctions led to the tumor’s eventual disappearance, prompting the author to reconsider the authority of digitally dominant sources and acknowledge the vast gaps in the ‘seemingly all-knowing internet.’
Also Read:
- AI-Generated ‘Workslop’ Erodes Employee Trust in US Workplaces, Gene Marks Argues
- Generative AI’s Emerging Role in Shaping Climate Change Perceptions and Decisions
This trend suggests that in an increasingly AI-mediated world, future generations might lose touch with extensive bodies of experience, insight, and wisdom. While AI developers might contend that this is merely a ‘data problem’ solvable by incorporating more diverse training datasets, the essay posits that the challenges associated with data sourcing, prioritization, and representation are far more intricate than such a simplistic solution implies. The core issue at stake is not just representation, but the very resilience and diversity of human knowledge itself.
