AI literacy as a foundation for national innovation
Apolitical. An interview with Mkhitar Hayrapetyan, Minister of High-Tech Industry, Armenia
As part of the Government AI Campus, Apolitical is publishing a series of articles featuring leaders from the 2026 Government AI 100, a list recognising public servants around the world who are pioneering AI adoption, capacity building and regulation.
Apolitical spoke with Mkhitar Hayrapetyan, Minister of High-Tech Industry of the Republic of Armenia, who plays a central role in the country’s efforts to build its AI and innovation ecosystem. He has overseen initiatives including AI data centres and has been a vocal advocate for mandatory AI literacy across education systems.
Read also
You’ve been an advocate for mandatory AI literacy in schools and universities. How would you define AI literacy, and what core skills or understandings do you believe every student needs to prepare them to work responsibly and effectively with AI?
AI literacy is key in today’s world of adaptive skills and critical to someone’s future competitiveness. We’re relying on our human talents; that’s why we need to be careful with talent and skills development.
It’s not that everyone must become an AI engineer. It’s about ensuring that every graduate, whatever their profession, can safely and effectively use AI tools, understand core concepts such as data, models and limitations, critically evaluate AI outputs, and design and manage AI-enabled workflows.
I see AI literacy as three key abilities. First, understanding the basics: what AI is, how it works with data, and why it can sometimes be wrong or produce incorrect answers. Second, thinking critically and responsibly: being aware of risks like bias, privacy and misuse, and not blindly trusting AI or handing over responsibility to it. This also includes understanding that AI systems are shaped by the data they are trained on, which can reflect different cultural perspectives or biases. Third, practical use: using AI to solve problems and be more creative, while still checking information, making informed decisions and keeping humans in control, especially in important situations.
Our team was impressed last year at the ai/teens Worldwide Conference hosted by TUMO Centre for Creative Technologies in Yerevan. Teenagers, equipped with strong AI literacy, were not only using these tools but were confidently engaging in policy discussions and offering thoughtful recommendations. This is a powerful example of how literacy translates into agency, responsibility and leadership.
Thinking about skills development from a different angle, when designing AI education for very different groups like teachers, doctors or engineers, what principles have guided how you tailor skills and understanding to each context?
The principle is ‘common foundation, role-specific depth’. Everyone needs a baseline level of AI literacy: how AI works at a high level, what risks look like, how to verify outputs, and what responsible use means in a workplace. But after that, training must be tailored to tasks, decisions, and risk level. That is how you avoid two failures: overtraining (wasting scarce time) or undertraining (creating unsafe deployment and low trust).
Benchmarks are important to identify results. For teachers, the focus is not only on using AI tools but on teaching in a teacher-AI-student dynamic: pedagogy, assessment integrity, safeguarding, and a teacher’s power to make meaningful decisions and take action in their work. There is now strong international guidance that frames teacher competencies across ethics, AI foundations, pedagogy, and professional learning, precisely because teachers need more than tool ‘tips and tricks’.
For doctors and healthcare staff, the training centre of gravity shifts: patient safety, clinical accountability, data protection, model limitations in real clinical settings, and how to avoid automation bias. In medicine, ‘responsible use’ is inseparable from ethics and governance, including clarity on who is accountable when an AI-assisted recommendation is wrong.
For engineers and technical specialists, we go deeper into lifecycle skills: data quality, evaluation, robustness, security, documentation, monitoring, and the ability to meet governance expectations such as transparency and explainability where required. A practical way to frame this is to teach technical teams using trustworthy AI characteristics and risk management concepts that translate into engineering controls and operational practice.
What have been the most significant challenges in developing AI skills and adoption within the public sector, and what has actually helped address them?
The hardest challenges are organisational. The first challenge is around skills and confidence. Public servants are expected to use powerful tools responsibly, but many haven’t been trained, and some may already be using open AI tools. If leadership doesn’t create training, rules and safe internal pathways, you end up with shadow use, higher risk and uneven quality.
Second is data readiness and legacy systems. AI depends on reliable data and interoperable processes, but many government services were built in a pre-AI era. Data is often messy, incomplete, or hard to access, and using AI in real work is harder than it looks in demos. Data availability and legacy constraints are major blockers, alongside higher privacy and transparency expectations in government than in the private sector.
Third is change management. People worry about job displacement, risk of blame and ‘black box’ decisions. What has helped in practice are a few levers, like building capacity through partnerships, especially with academia, so that we don’t try to solve talent shortages only through hiring.
You’ve been involved in developing national AI infrastructure in Armenia, including initiatives like a new AI data centre and the AI Virtual Institute. What’s one piece of advice you’d give to governments embarking on similar large-scale AI infrastructure or ecosystem efforts?
Don’t treat “AI infrastructure” as a construction project; treat it as a national capability with a governance and access model from day one.
One of the most expensive failures is building compute that is under-used, captured by a narrow set of actors, or disconnected from ecosystem players, skills pipelines and public-value use cases.
In Armenia, the framing we’ve used is: widen access, reduce cost barriers, and connect compute to learning and product development (especially for students, startups, and researchers) through ecosystem instruments such as the AI Virtual Institute and subsidised access mechanisms. That logic is explicitly about democratising capability, not just ‘owning machines’. AI Virtual Institute is a strategic initiative of the Ministry of High-Tech Industry of the Republic of Armenia to empower AI innovators by accelerating ecosystem building.
Can you describe a project you’ve worked on where AI made a meaningful difference in government? What problem was it addressing, and how did AI help you reach your goals?
I think a strong example of ‘meaningful difference’ is where AI is used to reduce fraud and improve compliance outcomes in a way that is measurable and socially valuable. What matters isn’t just that a model can flag risk; it’s that the process becomes more consistent, auditable and targeted.
Another example is using AI to improve citizen interactions, like moving toward AI-supported call centres and creating a ‘citizen-AI dialogue platform’ that uses Armenian-language voice-to-text and text-to-voice capabilities to help citizens find information faster. That kind of project is about service quality and access: reducing waiting time, improving consistency of answers and freeing staff to handle complex cases while ensuring clear escalation to humans.
What is one strategic priority you believe every public sector leader should focus on today to lead responsible AI transformation and skills development?
I think the strategic priority is building a responsible AI operating model and treating skills as part of that operating model, not a side project. An operating model means: clear rules on what tools may be used for, training requirements by role, data-handling and privacy safeguards, procurement discipline, model monitoring and accountability for outcomes.
On skills specifically, ‘AI-ready government’ requires training at three levels: foundational literacy for the whole workforce, strategic literacy for leaders and advanced capability for digital/data professionals. Internal capability is important for compliance, accountability and effective use and that training needs differ across staff groups.
What excites you most about governments using AI, and what concerns you?
I’m excited by the chance to make government more responsive, effective and capable with the same or fewer resources. Automating repetitive administrative work, improving fraud and anomaly detection, supporting faster citizen answers and enabling better forecasting and policy design.
What concerns me is that AI can scale mistakes as easily as it can scale efficiency. The biggest risks are: bias and unfair outcomes, privacy violations, weak cybersecurity, over-reliance on generated content and low transparency that erodes trust. In high-stakes sectors like health, governance must prioritise ethics, human rights and accountability because the potential harm is tangible and personal. We need not forget that AI is a tool meant to help humans and professionals make better decisions, not the opposite.
My concern is also strategic: capability concentration. Compute, data and talent can concentrate power and opportunity in a narrow set of organisations or regions. That is why our ecosystem approach matters, pairing infrastructure with subsidised access models, education pipelines and inclusive participation, so the benefits are broadly shared and the public sector can adopt AI without losing democratic control.
Using AI will enhance human creativity. The human, whether a citizen or a public servant, will have more time to pursue their creative ideas. Digital solutions and automation will liberate people from everyday bureaucratic routines. People will engage in higher-value activities that require judgment, imagination, and critical thinking, leaving technical and repetitive work to AI.
