How would you describe your work in 90 seconds?

My work lies at the intersection of game theory and machine learning, often referred to as AI. It is driven by a simple observation: most AI models deployed today are trained to predict, not to participate. Yet intelligence does not develop in isolation. Humans learn through interaction, competition, and adaptation. AI systems can also be designed as collections of interacting agents. Interestingly, these systems often develop better models when placed in competitive environments. The challenge is that most current AI models are trained on static datasets collected in the past. When we want to design new policies or interventions that affect society, the relevant data often does not yet exist. This is where current models fall short.

As soon as a model is deployed in a strategic environment, the agents interacting with it begin adapting to it. Consider spam filtering: adversaries continually seek ways to bypass defenses, forcing models to evolve. In this sense, there is an ongoing game between AI systems and the people who use—or seek to exploit—them. Understanding the long-term outcomes of these interactions requires a game-theoretic perspective. This is why game theory plays an important role in AI and machine learning: It provides both the principles and algorithms needed to study multi-agent systems. For example, in one of our recent works, we show that agents in market settings converge to strategies that drive prices upward, even without communication, coordination, or explicit agreement. Simply by seeking to maximize their own wealth, they arrive at outcomes resembling price inflation in the model. In other words, these behaviors emerge naturally from the incentives built into the system and represent its long-run equilibrium.

How did you get in touch with informatics?

Growing up, I always wanted to become an architect and follow in my grandfather’s footsteps. I loved building things and was naturally drawn to subjects like mathematics, physics, and chemistry. I still remember my first encounter with the idea of algorithms, probably through listening to my father talk about them. Without really knowing what they were, I imagined them as abstract, almost mysterious mathematical objects. Then one day, I learned what an algorithm actually is, and I was fascinated by how it worked. In some ways, it matched my imagination, though not exactly. I went home that day thinking that if computers had existed in my grandfather’s time, he probably would have worked with them because they were so fascinating. That was when my interests began to shift. Around the same time, my older sister was studying computer science, and I occasionally attended her classes. I loved the discussions and sometimes even joined in, despite still being in high school.

As an undergraduate, I took part in a competition where I helped design an AI model. I found it incredibly rewarding to develop simple insights under practical constraints, such as efficiency and limited computation time. Some of those ideas proved surprisingly effective, and our team won first prize in an international competition. I was fascinated by how small, elegant insights could lead to strong performance. I have always enjoyed games, board games, and strategic problem-solving, and computer science gave me the language and framework to express that way of thinking and put it to use. It felt like a perfect fit.

What makes you happy in your work?

I think I am happiest when I have the opportunity to try to understand something that no one fully understands yet. I am drawn to problems where the answer is not known in advance. During my PhD, I gradually moved from more concrete problems to more abstract ones, partly because the range of applications is so broad. What I find most rewarding is discovering a simple way to describe a mechanism or property of a complex problem. The moment when a complex system reveals a simple underlying structure is especially fulfilling. Over time, I have also come to appreciate the people aspect of research. AI in particular attracts many highly talented researchers, and working with them is a unique experience. In discussions, there are moments of deep collective focus where you are fully absorbed in solving a problem together, which I find very rewarding.

More recently, I have also enjoyed working with students. They are very capable and motivated to find their own direction and contribute in meaningful ways. Seeing them develop over time is very satisfying, and supporting them is also inspiring and motivating for me. Another rewarding aspect is when an abstract idea becomes clearly expressible. Sometimes it starts with an intuition about a pattern in a complex system, but formalizing it so others can understand it is not always easy. When that succeeds, and experiments confirm that the idea also improves results, it is particularly satisfying.

Where do you see the connection between your work and everyday life?

My research focuses on how multiple agents interact in complex systems. Many real-world decisions are not isolated, but depend on others’ behaviors. In markets, for example, a company’s reward depends not only on its own pricing strategy, but also on competitors’ actions. Together, those factors determine outcomes such as revenue. Similar interaction effects appear everywhere, even in simple settings like traffic lights, where individual incentives lead to an overall equilibrium.

This perspective extends to my work on AI systems. The focus is less on what a single model or agent can do, and more on what a system of interacting agents can achieve together. In many applications, especially with large language models, different models specialize in different tasks, and their performance depends on how they coordinate and exchange information. The key idea is that system-level behavior is shaped by strategic interaction, not just individual intelligence. This is relevant in areas such as autonomous systems, smart cities, robotics, finance, and resource management, where outcomes emerge from multiple decision-makers interacting. Understanding these interactions helps design AI systems that are more reliable and better aligned with desired objectives.

A related direction is policy design. Policymakers can use simulations to anticipate how different agents—such as consumers and providers—might react to a rule or incentive structure. This is the idea behind mechanism design: creating rules of the game so that desirable system-wide outcomes emerge. A simple example is electricity pricing, where tariffs and penalties are designed to prevent grid overload while still balancing incentives for both users and providers. More generally, agents can be modeled as optimizing objectives like minimizing cost or maximizing utility, and their strategies evolve over time through repeated interactions. AI systems behave similarly, adjusting step by step rather than all at once. Studying these dynamics helps predict outcomes, test policies in simulation, and better understand complex adaptive systems where no single decision can be considered in isolation.

Why do you think there are still so few women in Computer Science?

This is a complex question, and historically quite interesting. Computer science initially had many women in the field—indeed, some of the earliest programmers were women—but over time the representation has shifted, and today increasing diversity in informatics has become an explicit goal.

I grew up in North Macedonia, where in my environment, there was no notion that women were less suited for computer science. In fact, there were many women in the field. I simply followed what I was good at—mathematics and physics—and what I enjoyed. In school, feedback is relatively immediate, so it is easier to recognize your strengths and interests. In that context, gender was never part of the equation. This changed abruptly when I moved abroad for my PhD. In my first course, I was the only woman in a group of about 30 students. At first, I did not think much of it and even felt motivated to prove myself. But over time, I realized that such environments do have an effect, especially when small interactions accumulate during already uncertain periods of personal and academic development. I was fortunate to have already built confidence in a different context before that experience, which helped me maintain perspective. Not everyone has that foundation.

In general, I believe the issue starts much earlier—shaped by societal expectations, lack of visible role models, and the perception of computer science as a non-female field. These factors influence choices long before university, and once an imbalance exists, it tends to reinforce itself. Still, small actions can make a difference: visibility, mentoring, and creating environments where people feel they belong. Ultimately, I think progress comes from both structural change and everyday interactions that help more people see themselves as part of the field.

Tatjana Chavdarova is an Assistant Professor at the Research Unit Machine Learning at TU Wien Informatics. Her current project, Multi-Player Artificial Intelligence, focuses on three core goals: Improving optimization for multiplayer games, building foundational frameworks for dynamic multi-agent AI, and advancing multi-agent reinforcement learning with socially conscious agents. Tatjana’s Vienna Research Group for Young Investigators is funded by the Vienna Science and Technology Fund (WWTF) and is set to run until 2033.

Discover the whole #5QW series.

Curious about our other news? Subscribe to our news feed, calendar, or newsletter, or follow us on social media.