AI in Central Asia: How to Build Sovereign Intelligence Without Losing Cultural Heritage
In the first two issues of our "AI in Central Asia" series, we explored how Central Asia is taking its first systemic steps in the era of artificial intelligence. We examined the region's heterogeneous institutional landscape—from the development and adoption of national strategies and startup ecosystems to large-scale AI infrastructure development.
It has become clear that the algorithmic race in our region is not simply a tribute to a global trend, but a question of the long-term competitiveness of all of Central Asia. Some Central Asian states have reached an important milestone—the creation of their own, sovereign language models. And here a much more subtle and fundamental challenge emerges...
Let's look at the creation of national artificial intelligence not as a technical report, but as a living chronicle of our times, where digital independence is becoming the key to preserving cultural identity.
When we talk about the emergence of sovereign artificial intelligence, the average person often has the illusion that it's just another computer program downloaded from the internet by clever engineers and launched on a powerful server. In reality, colossal labor of thousands of people lies behind the beautiful interfaces.
In practice, the emergence of sovereign AI is a large-scale "construction project," where words, meanings, historical chronicles, scientific treatises, and much more are used instead of bricks. The countries of Central Asia clearly understand today that if artificial intelligence is not taught to think, joke, empathize, and work in our native language, then tomorrow, citizens could find themselves culturally captive to foreign algorithms.
It could happen that AI, trained exclusively on foreign datasets, will think in categories alien to us, replacing subtle national meanings with crude machine translations, and true spiritual values with philosophically and morally unacceptable ones.
Let's consider how Eastern countries, similar to us in national character, developed sovereign artificial intelligence models (for example, Falcon from the Abu Dhabi Institute of Technology Innovation), which are based on a combination of advanced Western technologies and deep adaptation of systems to local sociocultural norms.
Using basic architectures created by global industry leaders, IT specialists from foreign countries implemented comprehensive control mechanisms into sovereign AI systems that protected the national model from the influence of alien ethical principles while simultaneously imbuing it with traditional values.
How did this work in practice?
The process of protecting the system from inappropriate norms begins at the training data preparation stage, when content related to socio-political concepts that conflict with national moral norms, gambling, alcohol, erotica, etc., is excluded from the initial text corpora.
The model then undergoes a fine-tuning stage with the help of experts, during which scientists, linguists, and cultural experts evaluate the response options and adjust the algorithm's behavior, reducing the likelihood of generating inappropriate material.
An additional layer of protection is provided by external software filters that intercept sensitive user queries and format responses in strict compliance with national legislation and public ethics.
On the one hand, this is happening, while on the other, a parallel process of integrating traditional values is underway by incorporating a rich reservoir of classical literature, spiritually oriented works of past thinkers, historical and legal documents, and poetry into the training curriculum.
In this way, the model is trained to consider the peculiarities of the traditional way of life, to show respect for family institutions and the older generation, and to observe norms of politeness. The introduction of specialized language processing algorithms allows the system to recognize the complex semantic nuances of native speech, ensuring a natural perception of national values and cultural context.
Thanks to this approach, foreign countries are successfully adopting technological advances from global industry while maintaining their own digital and cultural sovereignty in the creation of national intelligent systems.
Our closest neighbors, Kazakhstan and Uzbekistan, were the first in the Central Asian region to accept this challenge and have embarked on a journey that can be called digital baptism. The main lesson of their experience is that the most advanced technologies are ineffective without a union of science and government.
For example, in Kazakhstan, the creation of the Kaz-LLM model brought together academics, linguists, and programmers. It turned out that simply collecting millions of pages of text from the internet isn't enough. The internet absorbs terabytes of information garbage, errors, and rudeness. It required the delicate work of philologists, who manually annotated folklore and explained the intricacies of proverbs and sayings to algorithms.
And if AI isn't trained to recognize proverbs and sayings with their imagery and metaphors, distinguishing them from ordinary phrases and sentences, then the machine might automatically translate the proverb "The further into the forest, the more firewood" as "As you move deeper into the forest, the total amount of timber available for kindling will steadily increase."
Uzbekistan has also gone far in matters of language verification, creating rigorous linguistic testing systems where every word and phrase is verified by independent native speakers to purge the database of artificial cripples from other languages.
Therefore, when discussing the use of international experience, we shouldn't limit ourselves to the topic of modern technologies. We need to identify the pitfalls countries have faced when beginning to build their own national artificial intelligence systems. The main risk here is the deceptive ease of achieving quick results, and this factor must be taken into account from the outset.
For Turkmenistan, as well as for Tajikistan and Kyrgyzstan, which have yet to follow this difficult path, the experience of their neighbors is a valuable guide. It's like a map to follow without repeating the mistakes of others.
Of course, it's tempting to take an existing foreign model, superficially translate it, and declare the creation of a national system, but this is a dead end. Such an AI will resemble a visiting foreigner who has simply learned the local vocabulary but lacks understanding of our national worldview, philosophy, and spirituality.
For Turkmenistan, with its rich reservoir of classical poetry, distinctive cultural heritage, and unique government structure, it is vital to develop its own "golden dataset." It's like the first, purest draft of a future book—if you start writing it yourself, the text will flow naturally. It will take on a life of its own, like a child growing up before their parents' eyes.
But even the purest draft has its legal implications. Another pitfall encountered by pioneers is legal entanglements. Specifically, the issue of using works of contemporary national prose and poetry, academic articles, monographs, and press materials without infringing copyright.
AI developers in the region need to create flexible legal mechanisms at the parliamentary, justice (adalat), and culture ministry levels so that authors of published materials become co-authors of the national AI, rather than resorting to lawsuits, as has been the case in a number of foreign countries, including the United States, Great Britain, Germany, Canada, Italy, India, and others.
We have already written that sovereign AI is not just a toy for generating images. In Uzbekistan and Kazakhstan, for example, it is already successfully operating within closed state boundaries. It helps identify shady tax schemes, analyzes X-rays in regional hospitals, balances the load on power grids during periods of abnormal heat, and precisely calculates the need for irrigation water for farmers' fields.
Not to mention that in schools and universities, artificial intelligence is becoming a personal tutor, addressing students' weaknesses and relieving teachers of the burden of reviewing mountains of routine reports.
For Turkmenistan, this path offers its own opportunities. Drawing on the experience of its regional neighbors, it's possible to avoid the pitfalls of "dirty data" and build a clean, verified model from the start.
The main thing to remember is that AI is merely a mirror reflecting what people have input into it. And for it to reveal the true soul of the nation, the country's best minds—from experienced school teachers to academics—must infuse it with meaning.
What practical steps should ministries and agencies take to begin building a sovereign database? Where should the collection and digitization of language materials begin?
What should be the strategy for training national personnel for this large-scale project, without the need to attract expensive foreign specialists?
How can we coordinate the work of ministries, agencies, academic institutions, educators, and all those who will be selecting language material for AI? This should be done in a way that avoids terminological inconsistencies due to the fact that different groups of specialists from different industries and fields will be working with the texts?
All of these issues are undoubtedly extremely relevant for the future, enabling the potential of all Central Asian states to be combined into a single regional artificial intelligence system. Resolving these issues will make it possible to transform the creation of national AI from local projects into a powerful geopolitical and infrastructural foundation for all of Central Asia.
If, given the experience of regional leaders, we're talking about the first practical steps for ministries, we need to begin not only with purchasing powerful servers (although this is a crucial prerequisite for AI development), but with taking stock of the national heritage. This begins with creating a "golden dataset," which has been mentioned many times before. This work should primarily involve qualified specialists from the Ministry of Communications, academia, and the highly-potential humanities and technical universities of Turkmenistan.
The first step here should be a large-scale revision of already digitized text collections. This includes the archives of the Institute of Language, Literature, and National Manuscripts, state libraries, digitized legal acts, and general literary and terminological dictionaries. All of this forms the foundation stone from which the first digital imprint of the language is taken.
Particular attention should be paid to the creation of a Unified State Register of Terms, which will clearly record official translations of scientific, legal, and technical concepts into Turkmen. This will protect the future neural network from being cluttered with random and incorrect translations from the internet. This is a very complex and time-consuming issue, so I believe we will discuss its details in a separate article.
It is important to immediately involve universities in this work, making the digitization and initial cleaning of texts part of coursework and theses, as well as the research projects of young philologists and translators.
However, collecting language data is only half the battle; it also requires effective management. Therefore, the strategy for training national specialists should be built at the intersection of two disciplines – computational linguistics and data engineering.
To avoid dependence on expensive foreign expatriate specialists, it's important to launch targeted interdisciplinary programs at the country's leading specialized universities. Philologists need to be trained in the basics of data tagging and text annotation, turning them into digital editors who will "teach" (or rather, "train" or "explain") machines context and cultural codes. At the same time, engineers need to specialize in working with big data and language model architectures.
The experience of Kazakhstan and Uzbekistan shows that the most effective approach is to create joint research labs with regional IT companies, where young specialists can complete internships on their colleagues' real supercomputers, gaining practical experience without interrupting their national work. Once this foundation is established, enormous opportunities will open up for the more active application of AI in key sectors of the Turkmenistan economy.
Thus, in the arid climate, water resource management is becoming a matter of survival not only for Turkmenistan, but also for Uzbekistan and the southern regions of Kazakhstan. Sovereign AI, integrated with sensors on the Amu Darya, the Karakum Canal, and other water management facilities, is capable of calculating evaporation, filtration, and the optimal balance of water distribution for field irrigation in real time, preventing colossal losses.
Incidentally, good news: in the coming days, water management agencies of the Central Asian states, in accordance with the decision of the Interstate Coordination Water Commission adopted on August 7, will form a working group to create an automated flow metering system in the Syr Darya basin. Thus, the Syr Darya will become the first "digitized" river in our region, and at the same time, a harbinger of the expansion of digital monitoring practices to the entire Amu Darya basin.
In the oil and gas sector, neural networks can become a key tool for predictive analysis. AI can process seismic data to more accurately identify new formations at the Galkynysh field, optimize drilling modes, and, crucially, continuously monitor the pressure and integrity of main gas pipelines.
Algorithms can detect micro-gas leaks or predict pipe wear long before dispatchers even see a drop in pressure on their instruments, dramatically increasing the reliability and safety of the entire gas export infrastructure in Turkmenistan and the Central Asian countries through which the three branches of the Turkmenistan-China gas pipeline pass. A fourth branch, for which construction is already underway, will join them in the foreseeable future.
But the most exciting horizon opens up at the intersection of the interests of all Central Asian states. Combining AI capabilities into a single system has the potential to transform our region into a cohesive economic hub. In the electric power sector, artificial intelligence could become a digital dispatcher for the Unified Energy System of Central Asia—the aforementioned Central Asian Energy Ring.
Integrating the hydropower capacities of Tajikistan and Kyrgyzstan with the thermal power plants of Turkmenistan, Uzbekistan, and Kazakhstan, under the control of a unified artificial intelligence system, will automatically balance energy flows during peak loads or seasonal low water levels, eliminating rolling blackouts.
In the transport and transit sector, a unified regional AI system could link the customs and logistics systems of all five countries, becoming a true digital dispatcher. It would be able to optimize cargo flows along the East-West and North-South corridors, predict checkpoint congestion, automatically verify transit declarations, and coordinate rail and road schedules—thus literally "removing" invisible bureaucratic barriers, dramatically reducing delivery times.
Ultimately, the synergy of national AI systems will create a large-scale regional precedent, in which the digital sovereignty of each individual country will not divide, but rather firmly unite Central Asia into a unified, high-tech, and independent ecosystem capable of competing on equal terms with any global tech giant.
Clearly, these ambitious goals and practical steps to achieve them involve more than just technical specifications; they require the consolidated views of the entire expert community. The creation of a "golden dataset," the legal protection of intellectual property, the ethics of algorithms, and interdisciplinary education—all these topics urgently require in-depth professional discussions involving scientists, developers, linguists, and government officials.
(To be continued)
Bekdurdy AMANSARYEV,
Center for Strategic Studies, Institute of International Relations, Ministry of Foreign Affairs of Turkmenistan
AI in Central Asia: So who is this multifaceted artificial intelligence?
