By Jay Seungjong Sun, PhD student, computer science, Sungkyunkwan University, Seoul; ICT Intern (2025)
I’m a second-year PhD student in Computer Science, based in South Korea. I completed my undergraduate degree at Kyung Hee University and my master’s degree at Sungkyunkwan University in South Korea, where I focused on Human-Computer Interaction and user experience design. I was always interested in how people engage with technology, but over time, I became more curious about how technology might engage with us. As artificial intelligence advanced, I found myself asking how machines might begin to resemble and communicate with humans—not just through surface-level interactions, but in ways that reflected deeper cognitive processes. That question is what led me to pursue a PhD.
I first learned about the ICT Summer Internship while attending EMNLP in Miami last November. I had the opportunity to meet Dr. Jonathan Gratch and Brian Deuksin Kwon, a PhD student at ICT. I was already familiar with their work in affective computing and negotiation agents, so I introduced myself and joined a workshop that Brian was leading. I was also invited to an ICT lunch during the conference, where I had meaningful conversations with other members of the team. After returning to Seoul, Brian told me about the summer internship program and encouraged me to apply. I was very glad to have the chance to reconnect with everyone this summer.
At ICT, I’ve been working with Dr. Gale Lucas, Director of ICT’s Technology Evaluation Lab. My project explores whether large language models might benefit from reasoning in a continuous space rather than using natural language. LLMs generate text by predicting a probability distribution over the next possible token. What we see—one word at a time—is just the discrete output of that process. But internally, the model is already operating in a continuous space. This raised an important question for me: what if we preserved that continuous representation, rather than forcing the model to commit to a single token?
My goal is to develop a negotiation agent that reasons via continuous vectors, not discrete tokens. Continuous thinking offers several benefits. It allows the model to hold multiple possibilities in mind at once, which helps with probabilistic reasoning. It also enables better performance on structured tasks, such as mathematical reasoning, graphs, and time series—areas where language-based reasoning often struggles. Negotiation is a particularly good use case. Human negotiation is full of ambiguity, multi-turn strategy, and incomplete information. But it’s difficult to teach those concepts to a model through natural language alone, especially given the limits of human-generated training data. Prior approaches have used reward modeling or linear programming to address this, but we are exploring a different method: self-play reinforcement learning using continuous representations. We hope this will help the model simulate more nuanced negotiation behavior, grounded in both logic and probability.
I have found that research becomes much stronger through discussion. One of the best parts of this internship has been our weekly Friday lab meetings. Each week, we share our progress, give each other feedback, and challenge ideas in constructive ways. I’ve learned that presenting my work—even when it’s still in progress—helps me clarify my own thinking. It’s also shown me that collaboration is more than just cooperation. It’s about building something together, and trusting that different perspectives will make the work better.
This summer has also given me time to think seriously about my future. After completing my PhD, I hope to return to the U.S. for a postdoctoral position. My experience at ICT has shown me what it means to work in an academic environment that values both deep technical skill and creative, interdisciplinary thinking. It is the kind of place I hope to find again.
For me, success is about bringing ideas to life and sharing them in a way that excites others. I enjoy thinking about unusual research directions and turning them into publishable work. One day, I hope someone reads one of my papers and says, “That’s a promising idea—I want to explore it further.” If that happens, I will feel I’ve contributed something meaningful.
This internship has also helped me grow in more personal ways. For the first time, I was invited to speak not only about my research but about myself—how I came to this field, why it matters to me, and what I hope to do next. That was a new experience for me, and a valuable one. I also learned from the mentorship of ICT researchers about what it really takes to build a career in academia—not just the technical work, but the persistence, collaboration, and communication skills required to navigate the journey.
If I were to offer one piece of advice to someone thinking about applying for the ICT internship in 2026, it would be this: if you bring a unique point of view to AI or computer science, ICT is a place where you will be taken seriously. You’ll have the chance to work with researchers who are both generous with their knowledge and rigorous in their standards. It is a rare and supportive environment for growth.
I’ll carry this experience with me as I return to Korea to continue my doctoral studies. I feel more confident in my research direction and more motivated to pursue questions that matter. I’m grateful to have spent this summer here—thinking, building, and learning alongside such talented people.
//