A multidisciplinary linguist at Estonian Business School and Tallinn University, Andres Karjus has worked tirelessly in raising public awareness on artificial intelligence (AI) since the recent explosion of this technology only two years ago. He also took part in the drafting of guidelines on the usage of AI in education at the Ministry of Education, as well as at EBS and Tallinn University. He is also involved in an expert network on AI convened by the Estonian Language Institute to chart the future of this language and culture in a world shared with extremely capable automatidation. At EBS, his work focuses on the impact of these new technologies on education. Countless seminars, talks, workshops and public lectures top this researcher’s activity as a now nationally recognised science communicator.
Interview by Pablo Veyrat
* * *
Pablo Veyrat: In these times of technological revolution, helping the public understand science, especially AI, is especially important. How did your interest in popularising AI begin?
Andres Karjus: I’ve been conducting various digital skills workshops on topics like data visualisation and analytics over the years. AI was just a natural addition to these. The more I saw people trying to understand it—especially with all the misconceptions around it—the more I felt the need to bring clarity, not just in academia but to a broader audience.
PV: And what misconceptions do you come across most often?
AK: Oh, many people think of AI as some kind of magic button that will solve everything or take over tasks completely. There’s this idea that AI will just “do your job for you,” but we’re not there yet. In my workshops and talks, I make a point to clarify that AI has limits, and it’s important to understand those. A lot of my work is about giving people a realistic, balanced understanding so they don’t overestimate or misuse the technology.
PV: Given where the technology is now, what do you think is a realistic expectation for AI?
AK: Well, that’s a big question. The current models we use—especially large language models underlying tools like ChatGPT—are nearing their capacity limits. The underlying architecture has a ceiling, so we’re seeing a plateau. However, there’s a lot of potential in combining these models with other tools and databases. For instance, integrating them with human expertise or specific datasets for certain tasks could be very powerful. The real potential lies in these hybrid solutions, but full automation? That’s still far away in many domains.
PV: You’ve worked with Estonia’s Ministry of Education on the use of AI in schools. What do you think would help Estonia make the most of AI in education?
AK: In Estonia, our education system is quite autonomous—schools and teachers have a lot of freedom. While that’s beneficial, I think some central support would be valuable, especially in terms of training and resources. I’ve noticed a lack of coordination among universities here. Each one is doing its own thing when it comes to guidelines, teacher training, and AI integration. Given Estonia’s small size, it would make more sense for us to collaborate and pool resources. We should be thinking about whether what we’re teaching is still relevant in a world where so much can be automated.
PV: And in terms of the impact on students—are there risks with using AI in education?
AK: Definitely. One big risk is that if students delegate their learning to AI, they won’t develop essential skills themselves. I also worry about something less obvious: the gradual disappearance of “apprentice” positions. AI can handle simple, repetitive tasks that used to be given to interns or junior staff to help them learn, thus taking away these entry points into the profession. Without these stepping-stone jobs, we could end up with a gap between what students know theoretically when they start and what they need in practical work, which will now involve only the more complex tasks. This could eventually affect hiring across industries.
These problems could be tackled through curricula focused on practice that would aim to close these gaps; or perhaps tailored internships at workplaces.
PV: So, how do you think we should motivate students to keep learning, especially when it’s so tempting just to use AI to do their assignments?
AK: Motivation is indeed a challenge. We’re actually running an experiment at EBS to measure how much motivation influences learning, even when AI tools are available. It’s tough, though—if a student can get an A with an AI-generated essay versus a B with their own work, it’s understandable why they’d be tempted to use the AI. Education will need to evolve to keep students engaged, and we’ll probably need more face-to-face methods to make learning feel meaningful.
PV: You’re part of a cross-institutional network aimed at creating AI solutions that work specifically for the Estonian language. How important is this in your opinion?
AK: Very. Currently, the popular AI tools are all based in English and are controlled by major American tech companies. These companies don’t specifically optimise their tools for smaller languages like Estonian. While these models can handle basic Estonian, their factual responses in Estonian cannot be fully relied upon, and relying on them for building products or services is risky, as the next model version might incidentally not have Estonian language capacity, just like current capacities are largely coincidental. This isn’t just a language issue; it affects our economy and our ability to integrate AI effectively across sectors in the future.
PV: Do you think the reliance on English-speaking AI is a threat to smaller languages?
AK: Yes, and it’s not just about language preservation. Without models that understand Estonian well, we’re forced to rely on external, cloud-based solutions, which raises data privacy concerns. In Europe, we’re very cautious about data protection. But it also means that European, or especially smaller local companies, face challenges in creating competitive AI models without good training data. Right now, the best option for some organisations here is to switch to English, but that has huge implications for language and culture.
PV: So what’s the solution?
AK: We need public, open AI models trained on Estonian language and cultural data. If we had those, they’d be a public good that any business, school, or government office could use. For example, local government departments could automate repetitive tasks while still following data protection laws. But to develop these models, we need more high-quality training data, and that’s a challenge under current European copyright laws, and especially for Estonian developers who might not have access to enough lawyers to litigate for data. If we don’t get this right, we risk falling behind other countries that are already automating their industries.
PV: What about AI’s impact beyond education and government—say, in businesses or other industries?
AK: AI is transforming how organisations operate, but the goal shouldn’t just be to replace people with machines. Instead, think about how you can free up employees from routine tasks so they can focus on more meaningful work. I tell people in my workshops to see automation as a way to expand their potential, not just cut costs. There’s a lot of interest in Estonia in this, and I believe that over time, as AI becomes easier to use, more industries will find ways to use it productively.
PV: Thinking of AI as a tool for freeing human potential rather than a replacement, do you have a vision of how society will look in 10 years or when AI is fully integrated?
AK: It’s hard to predict because there are so many factors at play, not least of which is climate change. But I’m cautiously optimistic. I think we can use AI to improve people’s lives, make education more personalised, and streamline government. However, we have to be proactive about preparing for different scenarios and updating our policies so we don’t get stuck with outdated methods or miss opportunities.
PV: One last question. Is there anything crucial about AI that you think the public is missing in discussions?
AK: Yes, I think people underestimate the importance of building AI systems that understand local languages and cultures. Most people don’t realise that the major AI tools they use are all built by American companies with no vested interest in small countries. If we don’t address this, Estonia—and similar nations—might miss out on AI’s full potential and increase dependency on American AI vendors. Language models trained on Estonian would let us preserve our language while benefiting economically. It’s an investment in our future.
(Answers have been edited for brevity and clarity)
