Festival of Learning and ICML in Seoul 2026

Although I watched Parasite, a Korean family drama that won the first top Oscar prize for a non-English film, and danced to Gangnam Style back in 2012, South Korea has not been at the top of my travel list. Still, my academic journey led me to experience this fascinating country firsthand in Summer 2026. My contribution to the International AI in Education conference (AIED) won me a lab-funded trip to the Festival of Learning, a co-location of three international conferences at the intersection of Computer Science and Education. Coincidentally, the International Conference on Machine Learning (ICML) took place the week after the Festival of Learning at the same conference center, which led me to extend my stay in Korea. This report will therefore blend my personal impressions of Korea with two conference reports.

Festival of Learning #

Attending AIED was the reason for this trip after getting my first PhD paper accepted there. It was my first time at AIED, but it felt familiar as there is a notable overlap with the LAK community, in which I found myself already two years earlier in Kyoto, Japan. A similar overlap exists with the EDM and L@S communities, making the co-location of these conferences a reasonable choice. This brought the venue to more than 1,200 registrations, prompting complaints that the two smaller conferences were being marginalized. Related to this theme, one of the panels discussed whether we are all AIED researchers now. Personally, I find it hard to draw a line between the kinds of work that each community produces, since both are concerned with methodological aspects as well as evaluation involving human learners.

The weekend before the official conference was reserved for workshops. A workshop by the National Tutoring Observatory informed me about current developments in the US in tutoring and related research. The project is set to collect large datasets of human tutoring to better understand what makes it so effective. I appreciate the close collaboration between researchers and tutoring providers whose representatives also gave talks at the workshop. This proximity to educational practice is what makes the field thrive. I learned that the National Tutoring Observatory faces challenges, including data privacy and the scalability of tutor dialogue annotation. Multimodal transcripts and annotation tools such as Sandpiper should help overcome these challenges. In another workshop on explainable AIED, I learned firsthand that it is not easy to agree on fundamental definitions and, more generally, find a common language given the field’s interdisciplinary nature. Although explainability has been around in more ML-focused venues for some time, it seems still not very popular in AIED.

During the main conference reception, I learned that my paper has been nominated for a Best Paper award, which was a great honor and showed me that the hard work had paid off. At the closing ceremony, the award in my nomination category was given to a paper that measured personalization to learner profiles in LLM tutors (Congratulations!). The overall best paper was awarded to colleagues from EPFL for their work on chatbot design and its effects in higher education (Congratulations to Jérome and colleagues!). I will hopefully get another chance to compete for the award as I continue my PhD journey!

A personal highlight of the conference was the teacher meetup, where conference attendees got the chance to talk to Korean teachers. At an evening event, organizers matched groups of researchers with a similar number of local teachers and suggested discussion questions. Volunteers were available for translation. However, I quickly figured that one of the teachers spoke decent English and could speak directly to them. After an initial question from the teacher group about how to detect students’ AI usage in assessments, which was hard to answer satisfactorily (since I believe this is not yet entirely possible), I got the chance to ask many questions to the teacher. South Korea had recently ramped up a program on AI-powered textbooks to replace traditional textbooks, which was ended after a short time. I was very curious about the digitization of education and how teachers work with it. The teachers at the meetup seemed generally tech-savvy and progressive but also told us that older teachers struggled with technological developments such as the AI textbook program.

Observations in Seoul and Busan #

Korean friends I had made during my studies previously had given me a preview of Korean food, making me put “not spicy” on top of my vocabulary study list, and hinted at the situation of Korean’s youth, which is marked by a competitive education system and a low birth rate ( 0.72 in 2023). Spending two weeks in this country, mostly in Seoul and a weekend in Busan, largely expanded this prior knowledge.

One of the first and recurring themes that I discovered was public safety. I quickly got used to the frequent sight of safety cameras. The metro walkways featured visible emergency equipment such as gas masks. Its tracks were separated from the platform by glass walls, allowing for driverless trains on some lines. Inside the metro, videos and announcements recalled safety instructions. Wall-mount flashlights were also a common security feature in almost every building, including all of my three accommodations. Above-ground, LED strips indicate red and green traffic lights on the pavement, in case anybody does not look up from their smartphone.

What struck me about the infrastructure and traffic in Seoul was how quiet it was despite being a huge metropolis. In Gangnam, the business district in which the conference venue was located, the numerous skyscrapers evoked memories of New York City. Still, the absence of honking cars and construction sites made it a relatively calm environment. I even discovered so-called dust-free zones that were underpasses with air filters and plants to provide pedestrians with healthy air. Interestingly, one rarely encounters trash bins in Seoul and Busan, a consequence of a “pay-as-you-throw” system against illegal disposal of household garbage. The two cities were still remarkably clean. Trash is diligently kept until a trash bin appears or one returns home. I even noticed workers in Busan scratching chewing gum from the floor.

As a guest, it is always hard to get to know the local population beyond service transactions. Luckily, Korean citizens I already knew or had just met were at the conference and happy to share their culture. This way, I spent an evening in a park on the Hangang riverside, eating instant noodles from a convenience store, a typical Korean experience I was told.

Speaking of Koreans that I interacted with outside of the conference scope, I was impressed by their unconditional helpfulness and openness. For example, a woman approached me while I was trying to figure out the hairdresser’s menu in Korean and volunteered to translate for me throughout my visit. On another occasion, I was lost while returning a plastic cup after a robot (!) at City Hall mixed me a drink, and another woman offered to handle the return and pay me the deposit she would receive digitally in cash. In general, Korean culture appeared considerate: Smokers enter cabins (separated between smoking and vaping) on the roadside, and I found cars parked reversed in lots to have a better overview and be quicker when leaving.

One of the things I wanted to better understand while traveling in Korea was the Miracle of the Han River, which encompasses the astronomical economic growth since the Korean War ended, placing Korea among the world’s ten largest economies in 2005. The country felt modern; not only are Seoul and Busan full of skyscrapers, but many of the cities I saw from the train were as well. Details such as tablets on tables for ordering and wireless chargers integrated into billboards at metro stations (and of course the robot that mixed my drink!) completed the picture. Bespoke tablets featured the US flag for English translations instead of the UK flag that I was more used to. It seems like the capitalist dream of the United States has become true in Korea with a society that could be more meritocratic but also stressful if your success entirely depends on exam grades instead of legacy and networking. On the other hand, the big conglomerates (called Chaebol), namely Hyundai, SK Group, Samsung, and LG, complicate entrepreneurship.

The special relationship to the United States became most apparent in my last two days in Seoul when I visited the War Memorial museum and the Demilitarized Zone between North and South Korea. In the War Memorial of Korea, the decisive role of the US in supporting South Korea in the Korean War (1950-53) became apparent. My friend and I stumbled into a temporary exhibition of poems in that museum and got to talk to older men who were exhibiting their poems. Despite the language barrier, they were willing to share their personal histories. Although, for the first time in 2025, a survey among South Koreans found no majority favoring reunification, the testimonials from the people we met clearly showed that the common history of the two states is not yet forgotten. As a German who is grateful for the reunification of my country, seeing two parts of the historically same state being so far apart from each other is saddening.

ICML #

ICML was the add-on of this trip, so I only gathered basic information before heading to the conference. I knew it would be larger than the Festival of Learning, but its scale still overwhelmed me. Whereas the Festival of Learning happened in smaller rooms, ICML booked all four halls of the conference center. One of the halls was entirely filled with a sponsor exhibition featuring booths from big tech, quantitative finance companies, and lesser-known organizations. I just followed attendees lining up in front of the hall at its opening and got washed into it with them and tried to understand what made them so impatient. It seemed to be the swag that made participants queue in multiple circles around stands such as OpenAI’s. The companies’ economic interest in top talent and brand value seemed to resonate among the participants. The many unofficial satellite events around the conference, sponsored by companies, were usually at capacity.

Research-wise, the conference was much broader than the Festival of Learning, spanning topics from deep learning theory to language models, computer vision, and their applications. In the poster sessions, I found myself mostly in the Social Aspects and Applications sections, hoping to catch something related to human learning or education. Another interesting genre of works was position papers that had their own submission track, did not expect empirical research, but contributed thought-provoking turns. Three of them stood out to me: one criticized the impact of big tech on AI research, whereas another criticized the dual-use nature of alignment research, since it also enables censorship. The third picked AI lock-in as a topic. A theme that is also of high interest in the AIED community and makes us reflect often on how heavy AI usage will impact agency and cognitive capabilities.

I also attended some oral paper presentations and keynotes. The keynote by Arvind Narayanan resonated the most with me. He was asking the question: What will be left for us to work on? I had come across his argument for treating AI as a normal technology earlier. Conference attendees can probably best imagine how AI progress will continue and change human work. Arvind’s response is to treat AI as human enhancement and focus on the development of taste, judgment, and agency. Project execution has become much faster; however, we should now focus on decision-making and delivery.

One of my favorite parts of the ICML was the AI x Education satellite event, where the organizer, Sungman Cho, brought together researchers, founders, and builders with a common interest. It allowed me to get to know more works and people related to education at ICML. For example, a work on educational video generation showed me that this kind of task is not yet possible in pixel space but can be mediated by generating code for programmatic video generation. EduMirror demonstrated the community’s interest in multi-agent simulation of social classroom dynamics, which could be used for teacher training if the simulation proves sufficiently realistic and challenging. Zach Pardos presented OATutor, an open-source adaptive tutoring system, and also promoted the Festival of Learning, which may lead to further synergies between the communities.