The WEIRDness of Generative AI
Education Inequality Exacerbated?
Carla Aerts
Taking organisations, leadership & startups into AI & Futures of Education | AIEd Strategy | EdTech | Thought Leadership | Policy | Strategic Innovation & Research | Innovation | Transformation | Facilitator | Speaker | Mentor |
March 15, 2025
Introduction
The world many of us live in is pretty WEIRD. Academic research, societal structures, medical interventions and education systems are all profoundly shaped by this WEIRDness. The acronym WEIRD[1] stands for:
W: Western E: Educated I: Industrialized R: Rich D: Democratic
It represents how the Global North and especially the West has disproportionately influenced world order, scientific progress, and economic development, ignoring the non-WEIRD people, their cultures, endeavours and societal structures. Despite evolving global dynamics, WEIRD perspectives remain dominant in technological innovation. The generative AI revolution is no exception.
LLMs underpinning GenAI, are predominantly built on Western Internet data and data models, algorithms and computational paradigms, heavily influenced by the “BigAI” players in Silicon Valley. Even as we witness a significant emergence of Chinese players, their systems still largely perpetuate WEIRD world views, potentially amplifying existing cultural, racial and societal biases.
The WEIRDness of GenAI presents significant challenges, and opportunities, for Education and equity and equality. While tackling these issues isn’t straightforward, this discussion will examine the WEIRD paradigm in a global context, highlight its cultural biases and implication for AI development and the LLMs, and explore the consequences for education as these technologies increasingly enter learning and teaching environments. We’ll consider how the introduction of these technologies can create new divides that could threaten education as a fundamental human right, but equally bring new opportunities that need actionable solution to help ensure AI advances educational equity rather than undermine it.
The WEIRD Paradigm in Global Context
WEIRD (Western, Educated, Industrialised, Rich and Demographic), a framework developed by Joseph Heinrich, has shaped our research shaped and knowledge, while “representing a rather thin slice of humanity’s cultural diversity”. The West and the Global North have been the drivers of this trend, too often ignoring cultural diversity and representation. It suffices to say that this results in considerable bias.
WEIRD dominance has significantly impacted cultural representation and diversity. A good example of this is in pharmacology and medicine, which only recently have started to consider ethnic diversity in the development of medicines and medical interventions. This example is only the tip of the iceberg in how WEIRD has had profound societal and cultural impact. The argument that our WEIRD view of the world, has led to a social colonization most certainly isn’t far-fetched.
Whilst we are witnessing slow change, the WEIRD frameworks and ensuing trends has ignored humanity’s rich cultural diversity and expression for far too long. The developments AI and, especially, GenAI sadly are only a reflection and continuation of this trend. After all, LLMs to date rely on Internet content, reflecting languages and cultures that are predominant on the World Wide Web. Several minority languages are supported by LLMs’, yet far too many local languages and cultures are not represented, perpetuating WEIRD in GenAI.
Cultural Bias and Its Implications
Most of you reading this article, need little introduction to the speed of evolution we have witnessed in the development of GenAI. Daily news on new versions of ChatGPT, Claude, Gemini, Perplexity, Midjourney and an increasing set of LLMs is challenging to keep up with. The past years of Internet and Social Media developments and ensuing social evolution, are set to pale into insignificance compared to what we are witnessing in AI and GenAI developments.
It is no secret that Silicon Valley has been the most significant driver of these technology developments and their societal impact. Although China is a significant player in GenAI developments and Europe is a small entrant, it is fair to see that the WEIRD dominance in technology development remains overwhelmingly US-influenced, resulting and Anglo-American cultural bias and dominance that reinforces inequality and unequal representation. The Internet may be vast, the slice of humanity’s cultural diversity representation is getting considerably thinner.
Bias in AI and algorithmic amplification of bias are no secret. It is subject of significant concern and discussion, the detail of which is not considered in this article. What is considered in this discussion, is its impact on cultural representation and the absence of indigenous, tribal and native cultures in GenAI support.
This absence presents us with a feedback loop between biased systems and societal outcomes, that is only set to grow and reinforce WEIRD-ness, creating a new ‘digital divide’.
Generative AI and LLMs: Examining WEIRD Origins
To understand LLMs’ weird origin, let’s have a short glimpse at their history. Large Language Models are the result of decades of natural language processing research. The field witnessed a fundamental breakthrough with Google’s development of the Transformer architecture in 2017. This allowed models to process language much faster. This innovation led to the birth of OpenAI’s GPT (Generative Pre-trained Transformer), that led to an explosive arrival of ChatGPT in 2022, bringing a paradigm shift, introducing conversational AI to the public. Other models such as Claude, Llama, Gemini (predecessor BARD), and Mistral were quick to follow. The continuing developments and evolutions of these technologies, the training of language models and refining of alignment techniques, have turned these technologies into powerful tools. They can generate human-like text and are even improving their reasoning capabilities across multiple domains.
LLMs rely on training data, typically harvested from the Internet and therefore representing its demographics and cultures. The Internet provides trillions of data points needed to train the models. But what about the demographics and cultures that aren’t represented in the data? They will be absent from the training data and subsequently from the Language Models.
The corporate concentration of AI development (BigGenAI) doesn’t preoccupy itself absence of cultural and language representation. They are responsible to their shareholders the growth of their solutions, so minority representation is not really their concern. Their argument: “There just isn’t data available to do this”.
Add to this problem the technical architecture decisions that rely on and encode the WEIRD assumptions, and little further explanation is needed. The problem is not going to be solved quickly.
The Educational Divide: Problems and Outcomes
Amplified WEIRD dominance and bias will continue and create societal impact and division to which education is not immune.
Educational AI applications will reflect this bias; minority, indigenous and native language and dialects will be left wanting. The same applies to cultural relevance and reference points in contextual examples and curricula support. Even where infrastructure and technology access allow for the introduction of GenAI in learning environments – too often still a fundamental issue - the crucial importance of context and cultural relevance won’t be reflected. The cultural bias and WEIRD dominance in data models and technical architectures will be amplified, affecting contextualised curricula delivery and assessment.
Both teachers and learners will lack resources and relevant support to develop the skills needed to be the citizens the AI era will need. This will lead to an AI-divide, not only in education but in future opportunity as these teachers and learners will be left behind.
What can we do to hold BigGenAI to account? The challenges are considerable. The ‘black box’ makes AI transparency hard and the lack of insight in algorithmic and neural network behaviours is increasingly limited. Add to that the above-mentioned profit motives of these tech giants, educational equity gets quickly demoted on their agenda.
Regulatory frameworks – even with the EU AI act and Data legislation, equally are still also reliant on the WEIRD views of humanity and don’t tend to take these considerations into account, leaving the BigGenAI players at liberty not to engage with these issues.
Impact on Education and the Right to It
United Nations’ SDG 4[2] stipulates ensuring inclusive and equitable quality education and promotion of lifelong learning opportunities for all by 2030. The achievement of this goal is increasingly unlikely.
This is not solely dependent on access to infrastructure and resources but is also due to an increasing and endemic shortage of qualified teachers. Most acute in the Global South, it is no longer solely a Global South problem. Adoption of AI in Education, not as a teacher replacement but supporting teaching and learning challenges is becoming more urgent. Yet, it is exactly the areas where teacher shortages are most severe that are most likely to be unable to reap the benefits of AI integration into education. This will exacerbate inequality and inequity and increase the competency- and skill-gaps teachers and learners face.
Several local, ethnic minority and indigenous initiatives have begun[3]. Yet it is unlikely they can move quickly enough or gain sufficient data access to make their impact felt in education. The need for diversity of knowledge can no longer be ignored, yet it will require considerable mobilisation and pressuring to secure BigGenAI support and expertise to provide solutions for equity in education and opportunity, leaving WEIRD frameworks behind.
From Challenges to Opportunities
To move to solutions, considerable changes in technical approaches need to be considered. These include data diversification strategies and alternative model training methodologies. These strategies and approaches need to be informed by cultural enquiry and auditing frameworks for GenAI. Without these, the above can’t happen.
Solving the challenge requires more than BigGenAI support and mindset change. Global regulatory and governance frameworks will need adapting and will be instrumental for solutions. These will require not only educational technology standards and regulation but will also demand investment for diverse AI development. Where the latter will come from, remains a very open question.
Current educational institutions’ response to AI varies significantly and, in many cases, sees levels of anxiety around AI integration into school and education institutions’ life and modus operandi. AI literacy is one of the main challenges that keep raising its head and needs solving. Educators are often not ready, nor are appropriate pedagogical approaches and the affordances of GenAI sufficiently considered.
We can’t just bring AI into the classroom, without changing what happens in it and how. Retrofitting to today’s classroom paradigm will not work and will result in failure. New pedagogies are needed as this is not solely a question of AI literacy and technology-led integration, Supporting and developing agency and ownership for educators, allowing them to encourage learner-agency and develop empowerment strategies to create a sense of AI ownership that relies on cultural relevance will be needed.
When it comes to local knowledge and its preservation, an additional challenge as well as opportunity is introduced that will need addressing. This requires relevant skillsets and teacher education, that not only focuses on curriculum, but the ethical and safe use of AI, focus on care and emotional wellbeing supported by cultural relevance. It will also require insights and wisdom of communities and their elders.
Realising this will aid the transition from problem to opportunity.
Conclusion
The explosion of GenAI will create new societal paradigms and profoundly change the world and the role of humanity in it. Education can no longer sit on the sidelines. Yet, when it comes to addressing equity and equality in education, these powerful tools and technologies are lacking. Support for minority languages and cultural relevance is lacking as current reliance of data and training of language models rely on and are informed by WEIRD frameworks favouring a representation of humanity, that prioritise its majority.
This reliance exacerbates the challenge of integration of GenAI and AI in education to promote equity and equality. It favours the WEIRD demographic, whilst ignoring indigenous people, ethnicities, native cultures, languages and dialects. Unless we can address this challenge comprehensively in co-creation and collaboration with and the support of the BigGenAI players, it is unlikely that the challenge can turn into an opportunity for education to be the catalyst for an equitable society. A society in which diversity can thrive and benefit the whole of humanity with equal access and opportunity, serving global community.
Very short reading List
Henrich, J., THE WEIRDEST PEOPLE IN THE WORLD, How the West Became Psychologically Peculiar and Particularly Prosperous, Harvard University, accessed 2025, https://weirdpeople.fas.harvard.edu/qa-weird
Indigenous Representation in AI: 21st Century Implication, A view of diversity and discrimination in AI Systems, accessed Feb 2025, https://www.theindegenous.org/indigenous-representation-in-ai-21st-century-implication
United Nations, The 17 Goals, Sustainable Development Goals, accessed Feb. 2025, https://sdgs.un.org/goals
In addition, many an AI newsletter, Andrew Maynards – The Future of Being Human, Ethan Mollick, et all.
Note: Proofread by Claude Sonnet 3.7
[1] Acronym coined by the anthropologist Jospeh Heinrich of Harvard University. He developed the WEIRD framework to raise people’s consciousness about psychological differences and highlight that WEIRD frameworks represent a rather thin slice of humanity’s cultural diversity. For Henrich WEIRD accentuates the sampling bias present in studies conducted in cognitive science, behavioural economics, and psychology.
[2] One of the 17 UN Sustainable Development Goals to be implemented by 2030, but unlikely to be realized. https://sdgs.un.org/goals
[3]nbsp;Example found here:nbsp;Indigenous Representation in AI: 21st Century Implication,nbsp;A view of diversity and discrimination in AI Systemshttps://www.theindegenous.org/indigenous-representation-in-ai-21st-century-implication