Report

On November 28 and 29, TU Wien Informatics and the Institute for Human Sciences (IWM) invited experts from a diverse range of disciplines to reflect on and discuss the future of Computer Science. The event, organized as a two-day workshop, focused on a possible Paradigm Shift in Computer science, the move from logic and algorithmic certainty to probability and large language models (LLMs). It was chaired by former Dean Hannes Werthner, Professor and Head of the Research Unit Databases and Artificial Intelligence Stefan Woltran, and Permanent Fellow at the IWM Ludger Hagedorn.

The workshop featured lively discussions across eight sessions; each keynote addressed distinct questions in computer science, offering insights from different perspectives and disciplines. Following each presentation, a panel of experts discussed the topic of the keynote and answered questions from the audience in a Q&A session. The central theme of the workshop, as reflected in the title, was whether current developments in Computer Science represent a paradigm shift or a gradual evolution. Key aspects of potential change included a shift from deterministic methods to probabilistic forecasting, the unanticipated dominance of data-driven (sub-symbolic) AI, the transition from systems providing guaranteed results to humans verifying the outputs of LLMs for factual accuracy, and the opacity of LLMs.

A Paradigm Shift?

Thomas Haigh opened the workshop with his keynote “History of AI”, which offered a historical overview of AI. He highlighted the non-linear nature of its development and the challenge of defining AI itself, suggesting that it often functions more as a “brand” than a coherent technology. Following this introduction, Thomas Eiter and Sepp Hochreiter compared symbolic (logic-based) and sub-symbolic (data-driven) AI, concluding that while symbolic approaches excel in causal reasoning and providing explanations, they face scalability challenges. Conversely, sub-symbolic AI demonstrates remarkable success in vision and language tasks but struggles with “hallucinations” (viewed by some as a feature rather than a bug). A recurring theme throughout the sessions of the first day was the potential and necessity of combining both approaches to create a more powerful computational and reasoning framework, a theme that was also explored in the last keynote of the day by Frank van Harmelen.