How are the countries of Central Asia – Kazakhstan, Kyrgyzstan, Tajikistan, and Uzbekistan – integrating artificial intelligence (AI) into their economies and public administration?

Globally, the hype around AI is subsiding: by some estimates, up to 80% of investments have not delivered the expected returns, as noted for example by the RAND Corporation. This has led to a certain cooling of interest in the topic. In Kazakhstan, however, we are seeing the opposite – there is currently a surge, and AI is being actively promoted despite parallel statements about the need to cut budget expenditures. This kind of contrast – “a feast during the plague” – is cause for concern. However, development cannot be stopped; it is necessary to take advantage of the situation to modernize and/or build new infrastructure that is ready to work with AI projects.

Developing AI requires large-scale investments, because it necessitates creating appropriate infrastructure: data centers, servers, high-performance processors. Today, such resources are concentrated in only a few countries, including the USA, China, Russia, and possibly some European states. Kazakhstan is not yet among them. We are only at the beginning of this journey. Plans are in place to build two data centers, which, according to preliminary estimates, will require about one and a half billion dollars. Meanwhile, at the moment even IPv4 addresses in the country are insufficient – they have to be purchased on the black market. A phased transition to the new IPv6 addressing should be undertaken, which would allow a “leap forward” in the development of AI and all digital projects.

If we speak about the region as a whole, Uzbekistan is moving forward most actively – practically every week brings news of new AI initiatives. They are actively attracting Chinese and Russian specialists and carrying out concrete projects. Kazakhstan has concentrated efforts on developing conceptual documents and legislation. Kazakhstan is also pursuing the creation of its own National AI Platform. Kyrgyzstan so far has neither the infrastructure nor the necessary services – often even official email is absent in government institutions. Kyrgyzstan is only now beginning to move forward – it needs the appropriate development of infrastructure and essential services between the state and society. Digitalization is not only the use of email, but first and foremost the provision of digital services to the population. In Tajikistan, the situation is even less favorable. Thus, in the context of AI in Central Asia, we are primarily talking about Kazakhstan and Uzbekistan, which are rightfully the catalysts of this entire process. One should not exclude the factor of competition among the countries themselves.

In which sectors is the adoption of AI most active? How is AI being used in government administration and public service delivery?

In April 2025, Stanford University conducted a study in the field of AI, “AI: numbers, trends, reality,” across 8 categories (“Research and Development,” “Technical Performance,” “AI Responsibility,” “Economy,” “Science and Medicine,” “Policy and Governance,” “Education,” and “Public Opinion”). It shows the following:

At the moment, the most active use of AI in Kazakhstan is observed in the media sector – text, video, and image generation has already become part of everyday practice, especially in the creative industries. However, in other sectors, the adoption of AI remains extremely limited. For example, the level of digitalization in Kazakhstan’s industry is only about 3%, and this is not even about implementing AI, but just basic digitization of processes – mostly using paper documents, Word, or Excel. In such conditions, one can say that the only widespread use of AI is among journalists, students, and schoolchildren, who use existing models to prepare essays and project work.

In medicine, there are isolated attempts to use AI for image analysis, for example in reading X-rays, but these are rare, one-off cases. Full-fledged AI solutions are not yet used in healthcare. As for government administration, the situation is similar: due to the absence of data centers capable of processing sensitive government data within the country, the use of AI is practically nonexistent. The issue is not a lack of desire, but a lack of technical capability. Moreover, today big data is held exclusively by the state, as is funding for AI projects. According to Kazakhstan’s legislation, data can be used only within the country, which has led to the initiative of creating its own AI platform.

How would you assess the current state of AI strategy and legislation in the region?

In Kazakhstan, a concept for AI development is being drafted, as well as a bill that is expected to include issues of copyright, data transfer, and security. Work on it is being overseen by Kuanyshbek Yesekeyev – a person who was involved from the beginning in creating the country’s e-government, which inspires a certain confidence in the process. However, as of now the law has not been adopted, and its prospects of coming into force remain unclearalthough its prospects of coming into force have become realistic. Even after it is passed, one should not expect immediate results – that will take at least two or three years. Draft legislation is being adopted very quickly, since the implementation of a major project – the creation of a National AI Platform – is anticipated.

In Uzbekistan, the process is moving more dynamically: decrees have already been issued, projects launched, and international partners actively engaged. Unlike Kazakhstan, which has placed emphasis on a conceptual framework, Uzbekistan is focused on practical implementation. Which of them has chosen the better strategy will become clear over time, but both countries are demonstrating a desire to move forward in this sphere.

In Kyrgyzstan, at present there are no official documents or regulatory initiatives in AI – discussions are ongoing, but it has not yet reached practical steps. In Tajikistan, the situation is similar or even less developed.

What role do governments, businesses, and academic institutions play in developing AI ecosystems in Central Asia? How developed is cross-sector collaboration?

As of today, governments remain the main drivers of AI development in the region. Businesses show interest, but their involvement is limited to separate initiatives. The role of universities is mainly to train personnel. However, we cannot expect them to generate ideas and developments on the level of world centers like MIT or Stanford – local universities are primarily focused on education, and even that is often carried out with outdated programs.

As a result, the formation of a full-fledged ecosystem that includes interconnected efforts of the government, private sector, and academia has not yet occurred. Collaboration between sectors is sporadic. Each participant operates autonomously: government bodies implement their own programs, private companies solve applied problems, universities work in isolation from practice. These elements are very poorly integrated with each other. These problems are not unique to Kazakhstan; they are common across all countries in the region.

How evident are regional cooperation or competition trends in the AI field? Are there examples of cooperation among Central Asian countries, and which country is taking the lead?

At the moment, one can hardly speak of genuine cooperation between the countries of the region. Rather, competition is observed. Both Kazakhstan and Uzbekistan have declared their intention to become regional AI hubs. Uzbekistan, for example, is actively collaborating with China and Russia – recent visits by high-ranking officials confirm the interest of outside players. Kazakhstan is also developing partnerships, including with Arab countries and China.

Kyrgyzstan also proclaims ambitions to become an AI hub, but without basic infrastructure this seems more like a bid to stay on trend. Thus, instead of uniting efforts, we see a pursuit of leadership – the countries are acting in parallel, not jointly.

How do you evaluate the scale of educational initiatives such as training one million people in Kazakhstan or one million developers in Uzbekistan? Are these real steps or just PR? To what extent do these programs meet the labor market’s needs?

In such initiatives, not only the scale is important, but also the content. The phrase “train a million people” requires clarification: does it mean completing an online course, obtaining a certificate, passing an exam? Unfortunately, clear criteria are lacking, and therefore behind the big numbers there is often an absence of a systematic approach. Even in the United States, according to Stanford University data, although 80% of teachers say they are ready to teach AI, half of them do not actually know the subject themselves. One should not forget that preparing instructors in AI also requires additional time. Thousands of teachers are needed across many areas: mathematics, networking, neural networks, infrastructure, and so on.

Uzbekistan demonstrates a high level of activity – indeed, a large number of events related to training specialists are being held there. In Kazakhstan, however, similar initiatives, including the Digital Almaty forums, training courses and projects, are often accompanied by many public statements but suffer in terms of quality and content. Training is frequently not aligned with the programs that are truly in demand in the market; there is a shortage of qualified instructors and a lack of practical orientation. All this gives these programs a formal character and limits their effectiveness.

In your view, how will artificial intelligence affect the labor market in Central Asia over the next 10 years? Is there reason to believe that AI will significantly change employment and professional requirements?

To be frank, I do not expect significant changes in the foreseeable future. In Kazakhstan, there is already an acute shortage of qualified specialists in key areas – cybersecurity, system administration, database development and management. There is a deficit not only of technical personnel but also of professional managers capable of carrying out large-scale IT projects. Many of those who hold leadership positions in the digital sphere have limited practical experience in managing major initiatives.

Meanwhile, the country’s population continues to grow, which creates additional pressure on the education system, healthcare, and the labor market. There are not enough kindergartens, schools, or university spots. The number of graduates is also increasing, and many of them simply have nowhere to apply their knowledge. AI, as a technological phenomenon, does not yet offer these people real entry points into employment. In difficult economic conditions, programs to create jobs are what’s needed.

Why is it so difficult to build a sustainable talent pool in the AI field?

The difficulty lies in a systemic lack of preparedness. Training AI specialists requires not only infrastructure but also qualified instructors capable of teaching at a level that meets international standards. Today, only a handful of such specialists exist in Kazakhstan. Educational institutions are simply not able to train tens of thousands of experts in areas like machine learning, big data analysis, or neural network development.

As a result, talented young people seek education abroad. Every year, thousands of Kazakhstani students leave to study in Russia, China, and Europe. China, in particular, is becoming a very popular destination – there they teach working with high-load systems that we simply do not have here.

Can we expect that AI, like digitalization in the 2000s, will eventually transform the labor market?

Comparing it to the digitalization of the late 2000s is quite appropriate. Back then, many predicted mass layoffs of officials due to the introduction of e-government – and indeed many lost their positions. However, this did not happen immediately, but rather as the result of a gradual, systemic effort: first the regulatory framework, then digital platforms, and only then structural changes.

The situation with AI is currently evolving in a similar way. In Kazakhstan, laws are being actively drafted and regulatory frameworks set up. But without human capital, academic programs, and a steady pipeline of specialists, this will not grow into a full-fledged knowledge economy. To change the situation, we need to send thousands of students right now to study in countries with advanced AI expertise – China, Russia, the US. Only after that can we talk about a real transformation of the labor market.

How do local startups and AI projects compete with global tech giants? Are there notable examples in the region, such as Higgsfield, Cerebra, AkylAI? What challenges and advantages do local players face?

Realistically speaking, it is not really possible to talk about competing with global tech corporations in our context – the scales are simply not comparable. Nonetheless, there are successful examples where startups from the region have managed to reach the international level. These initiatives are more one-off achievements demonstrating the existence of potential, rather than signs of sustainable competition with industry giants. The main problem remains the lack of developed infrastructure and a mature ecosystem capable of supporting the large-scale growth of such projects.

How does the current state of digital infrastructure in Central Asia affect the development and accessibility of AI?

The impact of infrastructure on the development of AI is critically important. Solutions in this field require significant computing power, the presence of specialized data centers, the use of hundreds of thousands of modern processors, and high-speed data transmission networks. For example, ChatGPT-3.0 operates using 150–180 billion parameters, while GPT-4 already uses 4.5 trillion. These parameters not only need to be stored but also processed, used to build new models, and so on. As a result, there is not a single data center in Kazakhstan capable of fully supporting AI tasks – we are talking only about basic data storage facilities. The construction of a high-speed data transmission network is also neither in planning nor in public discussion.

AI also requires specialized chips, access to which is limited or, in some cases, entirely absent. Moreover, the number of required chips is measured in the hundreds of thousands. In addition, working with networks that support AI-based projects requires the use of a new addressing system based on IPv6. Kazakhstan is still experiencing a severe shortage of the outdated IPv4 addresses, which have to be purchased on the black market.

The situation is even more acute in Kyrgyzstan: in many cases, basic digitalization of processes is still needed in government institutions. Under such conditions, it is premature to talk about AI. A similar situation can be observed in Tajikistan as well.

It is also worth mentioning the state of digital infrastructure when it comes to using AI products. Existing standards and data exchange protocols in infocommunication networks cannot ensure the required data transfer parameters. AI has ushered in a new era of the Internet, which must be built not only on supercomputers but also on infrastructure that is prepared to support it. The international technical community must begin working today on the creation of new protocols and standards that will be capable of supporting AI projects effectively across the world.

Can the Central Asian countries influence the development of global AI models in any way? For example, through digitizing cultural heritage, developing open data, or participating in international scientific initiatives?

Participation in global processes is possible, but the level of influence remains extremely limited. The region mainly acts as a consumer of data and technology, rather than a creator. The only way to change this balance is to form unique local datasets, provide open access to them, and actively participate in international initiatives. However, such efforts are being carried out at an insufficient scale. Without large-scale efforts to create and share open local data, the region’s influence on the development of AI models in the global context will remain minimal.

How significant are the risks associated with Central Asia’s absence in the training datasets of global AI models? Do the countries of the region have any way to influence the major tech players?

The risks are quite real. When a region is poorly represented in the datasets on which global AI models are trained, it essentially becomes “invisible” to these systems. Our languages, cultural context, and legal and administrative nuances simply are not recognized. This can lead to distorted decisions by such models, especially if they are used in government or legal processes.

In practice, this means we start using products developed based on a completely different reality – American, Chinese, or European. Although in Kazakhstan, for example, a large part of judicial practice has been digitized, having data available by itself does not guarantee the ability to influence global models. Access to these data is limited, and the models themselves are created outside the country.

Is it possible to engage with global companies and encourage the inclusion of local data in their products?

Theoretically – yes; practically – almost impossible. We live in an era of fragmentation of the digital space. Not long ago there was talk of globalization, technology sharing, open solutions. Today, however, countries are erecting barriers – digital, legal, political. Europe restricts cross-border data transfer, the US protects its digital assets, China is developing a closed infrastructure. This phenomenon has even been given a name – the “splinternet,” a fragmented internet.

In these conditions, countries like Kazakhstan find themselves in a vulnerable position: we are too small to be of interest to big investors, and we lack sufficient resources to create our own models. Even if we open up data, we have no way to ensure its large-scale processing.

If major players like China’s DeepSeek or the American company Presight agree to localize AI systems in Kazakhstan – will that bring benefits?

Localization is a compromise. On one hand, it allows compliance with data protection laws and partially takes the local context into account. On the other hand, the model still remains “foreign”: it is trained on someone else’s data, with different logic and priorities built in. That means the risk of misinterpreting local realities remains.

Besides, we face institutional constraints. Even if the government invests funds in localization, the real returns will be reaped first and foremost by the technology owner. Access to data, infrastructure, and scalability will remain with the external partner. This creates an asymmetry – we supply the content, and someone else extracts the value.

What steps can be taken to reduce dependence on global players and strengthen the region’s influence?

In the long run, the solution is to create local datasets and develop our own language models, as Kazakhstan has already begun doing with the Irbis-GPT project. This requires political will, sustained funding, and access to infrastructure. In addition, it is important to ensure these developments are open – otherwise they will remain internal projects without a real contribution to the global AI landscape.

But the key point is to recognize reality. We cannot hope that we will be “included” in the global AI space out of goodwill. We need to build bridges ourselves, create local initiatives, and actively participate in international scientific and technological processes to the extent our resources allow.

Can digitizing cultural heritage or developing open data programs improve the region’s representation in AI? Are there examples of such initiatives?

Yes, Kazakhstan already has some experience in this direction. Back in 2015, with support from the Ministry of Culture, the “Madeni Mura” (“Cultural Heritage”) project was implemented, under which cultural heritage objects were digitized. Digital routes were created and materials on traditions, history, and architecture were systematized. Unfortunately, the results of this project remained mostly in closed government systems and were not made publicly available. Nonetheless, the very fact that such a database exists represents significant potential.

In addition, Kazakhstan became one of the few countries to invest substantial funds in developing content in the Kazakh language. Starting in 2010, the government actively funded the expansion of the Kazakh-language Wikipedia – up to a million dollars per year. This enabled the creation of an impressive array of texts on a wide range of topics: culture, history, geography, industry. At one point, the Kazakh Wikipedia became one of the fastest-growing in the world, though later this growth sparked controversy, since the content was mostly generated using government resources.

It is precisely the content of the Kazakh-language Wikipedia that can be considered the foundation for building Kazakhstan’s own AI language model – Irbis-GPT. It uses the corpus of texts accumulated in recent decades, including Kazakh Wikipedia. This is a rare case where a country has not only a linguistic resource, but also a digitized cultural context suitable for machine learning.

In developing this model, a so-called “data lake” was assembled – a consolidated set of 90 government databases. These are not publicly accessible, but were provided to the developers as part of a pilot project. This allows Irbis-GPT to take into account the specifics of Kazakhstan’s reality and to generate more relevant answers.

Is it possible that such initiatives could strengthen Kazakhstan’s presence in the global AI space?

Only if two conditions are met: openness and sustainability. If the developments remain in internal systems and the models are used exclusively by government agencies, their impact will be limited. But if the government ensures access to these data for the scientific and technological community, they can become a full-fledged contribution to the development of global language models.

Furthermore, it’s important to understand that the digital representation of a language is not just a technological question. It is a matter of cultural independence, identity, and the ability to speak to the world on your own terms. Therefore, supporting such initiatives is a strategic task.

What forms of civil society participation do you observe in the processes of AI regulation and policy-making? How actively are non-profit organizations involved, and how is their interaction with government bodies structured?

There are many forms of interaction. There are working groups under government agencies, as well as structures like the National Chamber of Entrepreneurs “Atameken” in Kazakhstan. For example, we participate in the IT committee under Atameken, and also take part in sessions of Majilis working groups. Although formally we do not have voting rights – decisions are made by members of Parliament – we have the opportunity to convey our position.

In addition, roundtables, public hearings, forums, and expert discussions are regularly held – in Kazakhstan, as well as in Uzbekistan and Kyrgyzstan. Therefore, it would be incorrect to say that civil society is completely detached from the process.

How are these proposals actually taken into account? Are there instances where expert opinions and NGO comments influenced the content of legislation?

Unfortunately, in most cases the influence is extremely limited. After the events of January 2022, the situation changed. Whereas earlier there was a feeling that remarks from the civic sector could be integrated into final versions of documents, now that is more the exception.

For example, we prepared well-founded comments on the AI draft law – with participation from associations, lawyers, and technical experts. These proposals were sent to the head of the working group, but from then on the fate of these remarks depends solely on his decision. The discussion is not public, the criteria for selecting proposals are not explained, and a significant portion of recommendations are rejected without any explanation. Such practices reduce the effectiveness of public participation.

Which issues and topics in the draft AI law are causing the most concern for civil society?

Key topics of concern largely mirror the international agenda – in particular what is being discussed in Europe and the US. These include the legal aspects of copyright in the context of AI, questions of data transfer and protection, regulatory standards, risk-level classification, and restrictions on the use of high-risk technologies. For example, bans on manipulative algorithms, invasion of privacy, and defamation.

Also under discussion is the idea of creating a national AI platform – the government is considering centralizing development following the model of Singapore or South Korea. However, it remains unclear how applicable such a model is in our context.

Separate attention is being given to the certification and licensing of AI systems. Without clear standards, the market risks a proliferation of unreliable solutions and fake services. The issue of ethics is also being raised – this is probably the only area where there are at least some international framework documents, such as UNESCO’s recommendations. However, Kazakhstan so far lacks a local ethical code or clear guidelines for AI use.

What is your view on the idea of integrating Kazakhstan’s national AI platform into the global system? Could it really influence global processes?

Honestly, I find it difficult to imagine how that could be realized in practice. Such statements sound ambitious but are not yet backed by concrete mechanisms. Even at the national level, basic issues remain unresolved, not to mention global integration.

It is reminiscent of an eye-catching declaration like “Kazakhs have invented the wheel” – the phrase sounds confident, but there is no substantive content behind it yet. Before talking about influence on global processes, it is necessary to build a stable internal infrastructure.