Talal Abu-Ghazaleh
AI systems are now able interact with us in the native
fluency of almost any language, yet beneath this multilingual surface lies a
single dominant worldview. While the fluency is certainly impressive, it is not
a true understanding and rather a linguistic performance built on foundations
that are overwhelmingly shaped by English language data, as much of it produced
in the US. Recent research by Indonesian researchers on AI shows that while
users may converse with it in their local language, the advice given reflects
American cultural assumptions that prioritize individual autonomy, direct
confrontation, and personal boundaries over the consensus building and social
harmony that define Indonesian life. It seem that while the language is local,
the logic is distinctly foreign.
This illusion of cultural alignment is one of the most
dangerous misconceptions surrounding AI today. They are trained on datasets
where English dominates to an extraordinary degree, nearly 90% in the case of
certain AI models, while Arabic, despite being one of the world’s major
languages, accounts for less than 1% of the data used to train many models.
Even when these systems respond in another language, studies show they often
conduct their internal reasoning in English before translating the output back
into the user’s language. The result is a subtle but persistent form of
cultural influence. The worldview of the training data remains intact, even
when the words appear to belong to someone else.
The consequences become clear when AI is asked to
interpret culturally specific concepts. In the Arab context, concepts such as
‘tarbiya’, which refers to the holistic moral, social, and spiritual
cultivation of a child within the family and community, are often reduced by
Western trained AI systems to simple parenting techniques or discipline
strategies. The AI transforms culturally embedded social values into Western
psychological categories that emphasize personal autonomy and individual
emotion rather than collective identity and social obligation. Its apparent
fluency hides its inability to grasp the social fabric that gives these
concepts meaning, as it has not been trained on them.
What makes this especially concerning is that the bias
arrives disguised as empathy. When AI speaks to you in your own language, with
warmth and attentiveness, you assume it understands your world. You assume its
advice is grounded in your values. Yet the worldview embedded in these systems
is structurally Western because the infrastructure, data, and economic
incentives behind them are Western. This is the predictable outcome of who
builds the systems and whose knowledge dominates the digital sphere. The
effect, however, is profound. AI becomes a quiet force of cultural
normalization that shapes how people think about family, education,
responsibility, and identity.
China recognized this risk early and chose a different
path. It built its own AI ecosystem, including models such as DeepSeek and
Qwen, which reflect Chinese cultural logic rather than American norms. When
asked about workplace conflict, Chinese models recommend indirect communication
and harmony preserving strategies that align with Chinese social values, rather
than the direct confrontation favored by Western models. This is not only
technological independence. It is cultural sovereignty. Other regions are
attempting similar efforts, but many of these rely on Western models as their
foundation. They add local vocabulary, yet the underlying logic remains
Western. It is similar to repainting a house while the foundation still belongs
to someone else.
For the Arab world, this moment is decisive. If we rely
on Western AI systems, we will inherit Western assumptions about what is right,
normal, or desirable. Our children will receive guidance shaped by values that
may not reflect our own. AI is no longer just a tool. It is becoming a teacher,
a counselor, a mediator, and soon, a decision maker. If its worldview is not
ours, then our future will not be ours.
When I reflect on my earlier roles, whether building
TAG.Global from the ground up or advising the highest ranks of global
government, I am reminded of the simple truth that if we do not create our own
systems, others will define our reality for us. I learned early in my career
that sovereignty is built and not granted, and in the age of AI, cultural
sovereignty is as vital as economic or political independence. We cannot afford
to let our identity be translated for us by systems that were never trained on
our history, culture or values.
Democratizing AI does not mean giving everyone access to
the same Western built systems. It means that we move our best foot forward,
empowering nations and cultures to build models that reflect their own
intellectual traditions, social structures, and aspirations. For the Arab
world, this requires more than adding Arabic vocabulary to foreign models. It
demands Arab owned datasets, Arab designed architectures, Arab governed AI
institutions, and Arab digital infrastructure. It requires a unified regional
strategy that treats AI as a pillar of sovereignty, as essential as energy,
education, or national security.