���ҵ���

Building better AI for a more equitable world

Oana Ignat is an assistant professor of computer science and engineering at ���ҵ���. Her research focuses on building culturally grounded AI systems that address existing performance disparities and work equitably across diverse populations. She leads the , where she and her students work together to develop multimodal models, tasks, and datasets that emphasize inclusivity in AI across a range of demographic dimensions, including income levels, languages, cultures, age groups, and genders. 

Ignat is also a co-organizer of the workshop series at the Association for Computational Linguistics (ACL), one of the leading international conferences and research communities in natural language processing and AI. In addition, she co-hosts the panel sessions, helping students and early-career researchers navigate academia and industry, build confidence in their career paths, and connect with mentors and a broader global research community.

What questions or challenges are at the heart of your current work?

Current models work well only for a small fraction of the world’s languages and often perform better from higher-income and western populations. I aim to make AI systems more equitable across cultures, languages, and socioeconomic groups. I’m researching the populations these systems serve, and especially who they leave behind.. 

Beyond identifying these gaps, I build methods to reduce them through multilingual, multicultural, and human-centered AI design. It’s not enough for outsiders to come up with solutions. It’s important to involve the people affected by this technology, so we design solutions with these communities through a process called participatory design. We draw on their perspectives and lived experiences to meaningfully inform both the data and the design of AI systems, helping ensure they are more fairly represented and better served. You can find more about the challenges and solutions we propose in the paper, Why AI is WEIRD [].

Why is this issue important for the world to address at this time?

I identify performance disparities in AI models and look for ways to improve them so they don't further amplify biases in our society. If left unaddressed, these biases affect real-world decisions ranging from hiring and education to healthcare and access to resources.

This is a problem because AI systems are increasingly being integrated into our everyday life.. Current models are trained on data collected from the internet and tend to work better for communities that have access to opportunities and representation online. For example, if you ask an AI model to recognize a fridge from someone’s household, it’s much more likely to correctly recognize a fridge from a high-income household than one from a lower-income household, simply because those communities are represented differently on the internet, or not present at all. These kinds of gaps may seem small, but they reflect broader patterns of exclusion that can scale into larger societal impacts. We need to be more conscious about how we design AI systems and who they are designed for. In several of our published papers [], [], [], [], we measure these types of misalignments and propose methods for building more equitable AI systems.

Why have you chosen to dedicate your career to this research?

I come from a low-income region in Eastern Europe and did not have access to many opportunities growing up. I was fortunate to have mentors who encouraged me to apply for a Ph.D. in computer science in the United States, and that opportunity changed my life. It showed me how important mentorship and access to education can be.

I want to help others in a similar way as well. As a professor at ���ҵ���, I get to do that everyday through my teaching, research, and mentorship. By designing AI models that are more equitable for people who come from underrepresented communities like mine, I’m also working to make AI more accessible to everyone. 

This goal extends beyond my research. I also organize free, open, and recorded Q&A panels, like the sessions, where anyone can connect with senior mentors who want to give back to their community. I love it when students stop me at conferences to tell me how the mentorship panels helped them find an advisor or collaborator, or simply gave them confidence to start or finish their MS or PhD.

How have your students impacted your research?

My students are a big part of my research. They shape my ideas and the direction of my research. It’s a constant process of open discussion, brainstorming, and collaboration with my students. I try to empower them to think critically about science and come up with creative solutions. I appreciate being able to share my experiences and learn what motivates them as we collaborate on research papers, open-source models, and datasets. 

A fun recent project, , involved two of my MS students who came up with the idea to create an AI model that can translate between cross-cultural social media memes. At first, it might seem too playful for academia, but the project quickly became very popular. Our dataset had almost 20,000 downloads in just a few days because it really resonated with the broader community. It was a reminder to me that students have unique, creative ideas that can shape research in very novel and unexpected ways. I’m constantly learning from them.

What’s a book in your field that you think everyone should read?

I read Invisible Women by Caroline Criado-Perez a few years ago, and it stuck with me as a concrete example of how design decisions can shape how well systems work for different populations, especially in relation to gender differences. For example, in car safety testing, crash dummies were designed based on the body shape of men. Because of this decision, car safety features were built to protect the male body. It wasn’t inclusive, and as a result, women were more likely to be hurt in car accidents. The choices we make can lead to very drastic consequences, and this book really shows why fairness and equity need to be part of how we design systems from the start.