Why does Kazakhstan require sovereign artificial intelligence capabilities when frontier models like ChatGPT are readily accessible? Why do state digital infrastructures require years to transform, and why does staff reskilling frequently fail to yield technology adoption? How can digital transformation genuinely serve citizens seeking public services?
In this episode of the RAEM podcast, Raimbek Iskendirov sits down with Anar Tuleubayeva, Vice Chair of the Association for the Development of Artificial Intelligence in Kazakhstan. Previously overseeing cybersecurity within state institutions, Anar directed engineering across national statistics, social security disbursement, and compulsory medical insurance platforms, participating as an expert contributor in drafting Kazakhstan’s National AI Law and Digital Code.
A discussion ostensibly centered on digital sovereignty quickly expanded into institutional governance: public decision-making frameworks, bureaucratic inertia against workflow reform, and the systemic disruption caused by leadership transitions.
The Strategic Imperative for a Sovereign Kazakh Language Model
Anar Tuleubayeva champions developing national foundational models as a prerequisite for long-term state resilience. Unrestricted access to foreign frontier technologies remains contingent on external boardrooms and geopolitical dynamics beyond sovereign domestic control.
Simultaneously, she insists sovereign models must solve concrete, everyday user requirements: high-accuracy Kazakh language processing, localized educational support, and verified access to domestic regulatory archives.
“Who understands the nuances of the Kazakh language better than we do?” she observes.
A second core rationale centers on capital retention. Foreign enterprise AI licenses consume extensive budgets across public agencies, domestic corporations, and private consumers. Tuleubayeva argues that strategic technology planning must forecast cumulative sovereign outflows over a ten-year horizon, evaluating what proportion could directly fund domestic R&D ecosystems.
She recommends public co-financing models to reduce access barriers for citizens, noting that accessibility must be matched by utility: users adopt tools solely when they efficiently solve specific workflows.
Why Cloning Consumer Platforms Misses the Strategic Mark
The dialogue addresses a persistent dilemma: should states finance domestic replicas of every dominant global consumer platform?
Iskendirov recalls prior initiatives attempting to launch localized social networks and video platforms, ventures that routinely collapsed against incumbents benefiting from global network effects, established habit loops, and immense capital reserves.
Tuleubayeva argues such expenditures must be judged against empirical return on investment: capital deployed versus verified societal utility relative to accessible global alternatives.
For national LLMs, genuine competitive advantage stems from proprietary linguistic depth and authentic national data corpora. The benchmark of success is delivering sufficient quality that global platforms seek integration with Kazakhstani models via API.
Hence, true technological sovereignty hinges on prioritization: identifying core competencies where domestic engineering can establish genuine regional leadership.
Civil Servants Seek AI Tools but Face Structural Barriers
Tuleubayeva notes strong latent demand for AI assistance among frontline civil servants. Former public-sector colleagues regularly consult her on accelerating administrative drafting and routine document workflows.
Yet adoption confronts the stringent nature of public administration: processing classified communications and personal citizen data prohibits sending sensitive materials to unregulated external cloud models.
Model accuracy presents another bottleneck. When available consumer tools generate hallucinations, staff must spend extra time verifying citations and correcting outputs, converting automation into an administrative burden.
As her former colleagues state: “We require reliable answers immediately.”
Under these operating conditions, academic training courses alone fail. Without tailored, compliant software interfaces and transparent operational guidelines, personnel revert to legacy manual workflows.
Why Digital Modernization Spans Decades
She highlights the digitization of national statistical reporting. Migrating legacy paper filings to centralized electronic databases required four years of software engineering and another three years to achieve institutional operational stability.
Implementation navigated uneven regional telecom connectivity alongside cultural skepticism from experienced staff accustomed to manual physical archives.
“Digital transformation invariably forces institutional skeletons out of the closet,” Tuleubayeva emphasizes.
Standardizing workflows surfaces latent inefficiencies, operational blind spots, and undocumented informal practices. Institutional resistance frequently stems from systemic anxiety regarding how automated audits reshape individual authority and operational accountability.
AI deployments encounter identical friction. If automated systems are perceived primarily as instruments of workforce reduction, organizational resistance stiffens. Prematurely replacing human decision-makers with uncalibrated algorithms risks destroying institutional trust.
Structural Reforms Proposed for National Health Insurance
A dedicated portion of the conversation explores her tenure within Kazakhstan’s Social Health Insurance Fund (FSMS).
During her appointment, the Fund lacked an autonomous transaction clearing platform providing real-time visibility into medical reimbursement claims, relying instead on intermediary systems.
Tuleubayeva proposed architecting a granular transaction ledger recording every clinical intervention as an authenticated digital ledger entry, linking service provision, patient identity, and financial settlement.
The initiative faced structural institutional resistance. Concluding that executive alignment could not be achieved under existing governance, she chose to resign.
The case study underscores that software architecture cannot succeed without political alignment: before building code, institutions must agree on the core operational problem and commit to organizational restructuring.
Preserving Human Access Points in Public Services
Discussing citizen interfaces, Tuleubayeva references personal frustrations with automated public call centers where rigid conversational bots prevented access to human caseworkers.
She emphasizes that deploying AI must preserve alternative service pathways: whenever automated systems fail to resolve an issue, immediate human escalation must remain accessible.
The same empathy applies to digital service design. Public architects must consider vulnerable cohorts who lack digital literacy, ensuring digital public infrastructure does not inadvertently disenfranchise citizens.
For the public, the benchmark of government digitization is measured by whether services are successfully accessed with minimal friction.
Retaining Cross-Functional Institutional Teams
In the concluding discussion, focus turns to leadership retention and public engineering talent.
Tuleubayeva critiques the common practice of dismantling experienced engineering units following executive turnover, which wipes out institutional memory developed over decades.
She reflects on personal career challenges, noting instances where male colleagues initially hesitated to report to female technical leadership before collaborative delivery established mutual respect.
Her response was to establish a three-month transitional commitment to align working methods before making organizational decisions.
Collaborative empathy is essential in enterprise technology: developers cannot optimize complex medical or administrative workflows in isolation from domain practitioners.
The discussion circles back to its central premise: establishing the structural conditions required for Kazakhstani engineering to build dependable, citizen-centric technologies.
The full discussion with Anar Tuleubayeva is available on the RAEM Podcast on YouTube.
Watch “Can Kazakhstan Achieve Technological Independence?” on YouTube
