Human analogical reasoning manifests in many pursuits, from scientific innovation to policy-making. Drawing inspiration from human analogical reasoning, robust analogy-making in AI could transform applications in domains such as law, education, and biology. However, realizing this potential requires bridging the interdisciplinary gap between AI and cognitive science through theoretically grounded, scalable, and integrated methods capable of robust, human-like abstraction.

Topics to be discussed in this Dagstuhl Seminar include the following:

1. Cognitive foundations of analogical abstraction and their implications for AI
2. Algorithmic and computational approaches to analogical abstraction at scale
3. Evaluation frameworks and benchmarks for analogical abstraction
4. Applications and impact of analogical abstraction in AI systems

Example questions within these topics are: Which components of human analogy are essential for AI analogical abstraction? What makes a good analogy? Can analogy be implemented as a general-purpose operation across modalities (e.g., text, vision, multimodal data)? How can we design process-sensitive evaluations that capture not only accuracy, but also reasoning steps? Why has analogical reasoning had limited impact in AI systems so far?

Discussion and Collaboration

No dedicated venue currently exists for advancing cross-disciplinary research on reliable AI analogical abstraction. This Dagstuhl Seminar is uniquely suited to bridge this gap. By uniting experts across cognitive sciences, AI (NLP, computer vision, commonsense reasoning), and applied domains like education and history, the seminar aims to establish a shared conceptual framework. Its ultimate goal is to catalyze collaboration and chart the future of analogy research, paving the way for next-generation intelligent systems capable of robust, human-like abstraction. The seminar prioritizes discussing open challenges over summarizing past publications. The format centers on perspective pitches, breakout sessions, and plenaries. To ensure effectiveness, the organizers will distribute a pre-seminar survey for participants to vote on and suggest topics, shaping the initial agenda.

Expected Outcomes

The seminar is expected to establish new links between researchers with complementary expertise. Primarily, we will design the sessions in a way that facilitates better understanding of the state of the art of analogical abstraction research across disciplines, identifies standing challenges, and inspires novel research directions. With the seminar, we aim to transform certain cross-disciplinary topics, such as using AI for education via analogy or commonality between biological and mathematical analogies. Second, the talks and breakout sessions during the seminar will outline a shared vision for a variety of challenging aspects of analogical abstraction, which we expect will lead to a perspective (consensus) paper initiated at the seminar and finalized afterward.

Creative Commons BY 4.0

Filip Ilievski, Marianna Marcella Bolognesi, Jay Pujara, and Ute Schmid

This seminar qualifies for Dagstuhl's LZI Junior Researchers program. Schloss Dagstuhl wishes to enable the participation of junior scientists with a specialisation fitting for this Dagstuhl Seminar, even if they are not on the radar of the organizers. Applications by outstanding junior scientists are possible until October 9, 2026.