Seminar
Context-aware multimodal AI navigates hidden pathways in five centuries of art evolution
- Jinhyuk Yun
- Date
- Friday 2 October 2026
- Time
- Series
- CWTS Research Seminars
- Location
-
Willem Einthoven
Kolffpad 1
2333 BN Leiden - Room
- Online & common rooms at CWTS, with speaker in the room
About the seminar
We project 500 years of Western paintings into CLIP's high-dimensional semantic space and show that contextual features embedded in its pretrained representations form a continuous topological pathway aligned with artistic periods, styles, and individual artists, outperforming conventional formal-element analysis (e.g., color, composition). This lets us quantitatively trace long-run shifts in content, such as the gradual decline of religious imagery, as smooth trajectories through semantic space. Image-generation experiments that infuse future stylistic contexts into historical paintings further validate this alignment, producing artworks consistent with target period styles. Together, these results suggest that modern AI systems implicitly encode a navigable latent structure of cultural knowledge, opening a new quantitative lens for data-driven art history and, more broadly, for studying how creative and intellectual traditions evolve. In this talk, I situate this study within my broader research trajectory of treating human knowledge as a living, interactive network, from the structure of collective knowledge platforms and pathologies of the scholarly record such as predatory publishing, to node-split methods for bibliometric network analysis, showing how each strand reflects a common interest in the structure of knowledge itself, and how the tools we use to study it shape what we can see.
About the speaker
Jinhyuk Yun is an Associate Professor at the School of AI Convergence, Soongsil University, Seoul, South Korea, currently on sabbatical as a Visiting Researcher in the Data Science for Humanity group at the Max Planck Institute for Security and Privacy (MPI-SP), Bochum, Germany. He studies human knowledge as a living interactive network. AI is now fundamentally transforming this network: accelerating knowledge production while narrowing its scope, displacing human contributors, and restructuring the ecosystems through which ideas emerge and spread. His research spans collective knowledge platforms (Wikipedia, GitHub, Stack Overflow), scientific knowledge evolution, ecosystem pathologies such as predatory publishing and citation polarization, and creativity in cultural domains, most recently how AI systems encode and reveal the latent structure of creativity. His most recent work, published in PNAS, uses multimodal AI to trace five centuries of Western art evolution, which he presents today. He leads the Advanced Data Science Lab (ADSL) at Soongsil University.