AI’s Threat to Global Knowledge, and How AI Governance Can be Rewired Beyond the West

AI systems are trained on data and shaped by assumptions about language, social organization, law, economics and the individual. They also exhibit a clear Western bias because their foundation, made of training data, safety guardrails and development teams are overwhelmingly rooted in Western, English-speaking societies.[1] This is particularly concerning as these AI systems could alienate users from their own cultures and communities as worldwide usage of AI for personal and professional advice skyrockets. Companies’ efforts to make their model feel more local by integrating regional data are reduced to a facade when the model’s fundamental engine is a Western creation. These biases have led to real-world problems, with users from non-western countries reporting dangerous medical misdiagnoses and mistranslations leading to the marginalization of centuries of knowledge from across cultures, substituting communal values and wisdom with a monolithic, standardized perspective.[2]

The starting point of ensuring that these biases do not occur is the distribution of the production of AI to non-western companies and institutions that can train respective AI systems with their own cultural knowledge. This is where governance comes in. Who should decide how it is distributed? What steps must be taken to allocate resources as evenly as possible? Who has the power or capacity to distribute? Companies like TSMC and NVIDIA have little to no incentive to distribute their chip supply to smaller companies or public institutions, prioritizing large tech giants because of extremely limited manufacturing capacity, massive bulk order sizes, and the need to protect its business relationships and high profit margins during periods of high global demand. Chip factories like TSMC can only make a certain number of advanced processors at one time and building new factories takes years and billions of dollars.[3] Hence, working with giant buyers is safer and easier than managing thousands of small orders. So, even though there is a buyer (smaller companies and public institutions), and there is a seller (chip making companies), the transaction is still not happening, due to economic and political implications, and more importantly, due to the continued satisfaction of self-interest.

Hence, policies need to be put in place to diversify the production of AI away from these big tech companies; one that would make the biggest difference, and is already in motion in a few nations, is the increased funding for university research centres, public labs and startups in non-western countries to develop their own AI systems with local knowledge frameworks. These would prioritise data diversity focused on reducing cultural and racial blind spots by deliberately collecting images, texts, and speech patterns from underrepresented demographics. To supplement this would be to also prioritise local contexts in workforces, with teams embedded in local communities understanding regional nuances, idioms, and social sensitivities that outsiders often miss. Engaging local communities through participatory audits to test AI outputs would also hold developers accountable for real-world impact outside Western markets. An example of this policy’s performance is Nigeria’s artificial intelligence software model, N-ATLAS (Nigerian Atlas for Languages & AI at Scale) built by the Lagos-based tech company Awarri Technologies in partnership with Nigerian government agencies, including the National Centre for Artificial Intelligence and Robotics (NCAIR) and the National Information Technology Development Agency (NITDA). Instead of designing a brand-new AI from scratch, developers built the N-ATLAS-LLM by fine-tuning Meta's Llama-3 8B open-source model. The team trained and customized this architecture using over 400 million tokens of locally gathered multilingual instruction data and speech recognition systems (built on Whisper Small architecture) to accurately capture Nigerian voices and linguistic diversity.[4] These localised efforts have shown results in Egypt, Indonesia, Thailand and Sweden. Governments collaborating with entrepreneurs and tech startups can hence help mitigate these Western biases.

But not every country can afford to fund AI research the way these nations do. This raises a key question: why should limited public money go into AI at all? The answer lies in the long-term socioeconomic benefits. Studies show that AI applications in healthcare can reduce diagnostic errors by up to 30%,[5] while AI-powered agricultural advisories can increase crop yields by 15-20% in developing economies.[6] In education, localized AI tutors in regional languages have improved learning outcomes by 25% in pilot programmes across South Asia.[7] For countries like Sri Lanka, which have limited budgets but strong human capital, the goal should not be to build giant foundation models from scratch, but to adapt existing open-source models to local needs. This is far cheaper and faster, and can still deliver high impact in areas like tourism, agriculture, and public services.

Public-private partnerships (PPPs) offer a practical way forward. Governments can provide policy support, data access, and partial funding, while private investors bring capital, technical expertise, and market discipline. There is already a clear market logic for this. In India, AI startups raised $1.22 billion across 188 deals in 2025 alone, a 58% increase from the previous year. AI's share of total venture capital funding in India has nearly tripled in five years, climbing from 4.5% in 2020 to 12.3% in 2025.[8] Investors see profitable products and companies that can scale across South Asia and beyond. Local-context AI creates unique advantages: better language models, domain-specific tools, and data-compliant solutions that Western models cannot easily replicate. For smaller economies, pooling resources can make even more sense. State governments could partner with private investors to launch AI hubs focused on local industries, such as apparel AI in Sri Lanka or agritech AI in Indian states. Governments can also use incentives to shape the market: tax breaks, grants, or procurement preferences for companies that invest in local data and models, and conversely, stricter data-localisation or transparency rules for foreign models that do not adapt to local contexts. Such measures would encourage big tech and chip companies to negotiate fairer terms, expand local infrastructure, and share capacity with public institutions.

In short, rewiring AI governance is not just about ethics, it is also about economics. When local knowledge becomes the foundation of AI systems, it creates products people actually need, and companies that can grow. That is why funding local AI is not a cost, but an investment in a country's future.

[1] University of Puget Sound, “University of Puget Sound Professor Uncovers AI’s Western Bias,” March 2, 2026,

[2] Rest of World, “AI Safety Bias,” September 1, 2026,

[3] Law & Economics Center, “From Moore’s Law to Market Rivalry: The Economic Forces That Shape the Semiconductor Manufacturing Industry,” November 12, 2025,

[4] Data Science Nigeria, “N-ATLAS V1: A Milestone for AI in Nigeria and the Future of Artificial Intelligence in Africa,” 2026,

[5] Algarni, Ali. “CareAssist GPT Improves Patient User Experience With a Patient Centered Approach to Computer Aided Diagnosis.” Scientific Reports 15 (1): 22727. 2025

[6] Tyagi et al. “Advances in Climate-Smart Agriculture: Strategies for Sustainable Crop Production.” Curr. Rese. Agri. Far. 7–7 (1): 17 26. 2026

[7] ResearchGate, “AI Tutors and the Future of Personalized Education in Developing Nations,” March, 2026,

[8] Venture Intelligence, “India AI & Deeptech Investments Landscape Report 2026,” 2026