Fellow-Klasse 2025/26 Prof. Dr. Kerstin Kunz

Arbeitsvorhaben am Marsilius-Kolleg

Efficient Knowledge Acquisition and Utilization in Human Cognition and Language Models

Humans face a new era where machine intelligence approaches parity with our cognitive abilities. Early science fiction scenarios have become reality - what lies ahead is hard to predict. In face of this, it is of utmost importance to understand the similarities and differences between human cognition and machine intelligence. Our project will compare information uptake in AI systems and humans that specialize in rapid language understanding and production: large language models and interpreters. We will analyze how both ingest new information and integrate it with pre-existing knowledge while

preventing forgetting of old information. For this we study and compare how the human brain and language models integrate new information for effective storage and retrieval via variable cues. This will allow us to understand and generalize strategies for efficient information uptake, consolidation and usability in both systems. In sum, our project aims to i) deepen our understanding of human cognition in comparison to knowledge processing in AI models, to ii) improve the learning strategies of LLMs and Simultaneous Interpreters, and to iii) transfer our insights to the domain of mental health. Our group is ideally suited for this project, as each field imports disciplinary insights that contribute to a holistic understanding of cognitive abilities that will be important across disciplines and society at large.