Imagine a world where the wisdom of centuries is slowly fading, replaced by a narrow, homogenized view of knowledge. This is not a dystopian fantasy but a reality we're creating with every advancement in AI. But here's where it gets controversial: as we train AI on the vast digital corpus, we risk amplifying existing biases and erasing alternative ways of knowing, especially those rooted in oral traditions, local practices, and 'low-resource' languages. And this is the part most people miss: the very systems we trust to guide us—from healthcare to education—are increasingly influenced by AI, which itself is shaped by these biases. This isn’t just about representation; it’s about the resilience and diversity of knowledge itself.
A few years ago, my family faced a medical dilemma when my father was diagnosed with a potentially malignant tumor on his tongue. My older sister, a Western-trained doctor, and my parents, who favored traditional remedies, clashed over the treatment. As the family mediator, I turned to the internet for guidance and, unsurprisingly, sided with my sister. We pushed for surgery, but my dad, using my sister’s pregnancy as a distraction, secretly continued his herbal treatments. Months later, the tumor shrank and disappeared. This left me questioning: had I been too quick to dismiss traditional knowledge in favor of digitally dominant sources?
This personal experience mirrors a global issue. At Cornell University, my research on responsible AI systems has revealed how digital platforms and generative AI (GenAI) amplify power imbalances in knowledge. The early internet was dominated by English and Western institutions, a bias that has hardened over time. Now, GenAI, trained on this skewed data, threatens to entrench these inequalities further. For instance, a 2025 study found that around half of ChatGPT queries sought practical guidance or information, yet these systems privilege Western, institutional knowledge while marginalizing alternatives.
Consider languages: they are vessels of knowledge, carrying rituals, customs, and ecological insights. Yet, 97% of the world’s languages are classified as 'low-resource,' often underrepresented or absent in digital spaces. For example, Hindi, spoken by 7.5% of the world’s population, accounts for only 0.2% of Common Crawl’s data. Tamil, my mother tongue, spoken by over 86 million people, represents just 0.04%. This digital exclusion risks erasing centuries of wisdom, from Indigenous architectural techniques to local ecological knowledge.
Take Dharan Ashok, an architect reviving natural building methods in India. He highlights how traditional construction, rooted in local materials, is being lost due to a lack of documentation and oral transmission. Similarly, Bengaluru’s water crisis reflects the erosion of community-led water management systems, replaced by centralized, Western-inspired models. These examples illustrate how knowledge homogenization undermines sustainability and resilience.
The issue extends beyond data gaps. GenAI models, by design, amplify dominant ideas, creating a feedback loop that narrows accessible knowledge. This 'mode amplification' means less frequent or niche knowledge—often from marginalized communities—is further sidelined. Commercial pressures exacerbate this, as AI models cater to lucrative, English-speaking users, reinforcing Western cultural values and epistemologies.
This isn’t just a technical problem but a structural one. Organizations often prioritize institutionally validated knowledge to avoid liability, sidelining local practices. Nonprofits like Seva, which document Indigenous agricultural methods, struggle to gain recognition and funding, caught in a cycle of needing validation to secure support but lacking resources to validate their work.
The loss of local knowledge isn’t just a tragedy for specific communities; it’s a global loss. As ecological crises intensify, the wisdom embedded in local systems becomes increasingly vital. Yet, AI’s role in education and knowledge dissemination risks further disconnecting future generations from this wisdom.
So, here’s the thought-provoking question: can we reconcile our pursuit of technological advancement with a genuine appreciation for diverse knowledge systems? Or will we continue to erase them, only to realize their value too late? The intelligence we need most might not be in AI but in our ability to see beyond the hierarchies that define what knowledge matters. Acknowledging our uncertainty, as I do with my dad’s herbal treatments, might be the honest starting point we need.
