The GRCDI was glad to host a new AI Seminar Series in late 2025, led by Centre member Dr Kasim Terzić from the School of Computer Science, University of St Andrews. This in-person series for members of the University of St Andrews community kicked off in October 2025 with a group discussion session, followed by four lectures delivered by colleagues from the School of Computer Science (Dr Kasim Terzić, Dr Ruth Hoffmann and a joint lecture by Dr Nguyen Dang and Dr Phong Le). The series wrapped up in December with a further discussion session where participants reflected on and discussed a question central to research on artificial intelligence: whether intelligence can be computed.
The GRCDI team wish to thank Kasim for leading this fascinating series, as well as all the speakers (Ruth Hoffman, Kasim Terzic, Nguyen Dan and Phong Le) and everyone who came along to join the discussions.
Below you can find a summary of the series.
Wednesday 1 October 2025- Discussion session: What is intelligence (to you)?
Intelligence is notoriously difficult to define. This discussion session brought together practitioners from various fields to better
understand how each discipline understands and models intelligence. By understanding differences and similarities in how we understand intelligence, we can start to work towards an understanding of which aspects of intelligence can (and cannot) be computed.
Wednesday 15 October 2025-Lecture: Overview and history of AI (Dr Kasim Terzić)
Artificial intelligence is in all the news, with new advances and products being announced regularly. This talk gave a brief overview of AI as a field and introduce some basic problems, questions, and techniques associated with AI. This lecture also served as an introduction into more detailed explorations of various AI sub-fields in the weeks that followed.
Wednesday 29 October 2025-Lecture: Symbolic reasoning (Dr Ruth Hoffmann)
Symbolic Reasoning (or Symbolic AI) consists of the logical modelling and an exhaustive search for definite solutions to problems. Whether that is finding the solution of a sudoku, finding an optimal route for delivery vehicles or creating kidney matching chains, symbolic AI and logic are the building blocks of this type of reasoning. We explored the foundations of (Symbolic) AI, logic and search, and what type of intelligence it might represent.
Wednesday 12 November 2025-Lecture: Reinforcement learning (Dr Nguyen Dang and Dr Phong Le)
Reinforcement learning (RL) is a machine learning paradigm in which an agent learns to make decisions by interacting with an environment to maximise cumulative reward. In recent years, RL has emerged as one of the most exciting and impactful areas of Artificial Intelligence. Its significance was underscored by the 2024 Turing Award awarded to Richard Sutton and Andrew Barto, whose pioneering work established the foundations of the field. Today, RL underpins a wide range of applications, including robotics, games, recommender systems, healthcare, resource management, and even the training of large language models. In this talk, we provided an introduction to the basic concepts of reinforcement learning. We then illustrated how these concepts are realised in practice via a number of example applications.
Wednesday 26 November 2025-Lecture: Deep learning and transformers (Dr Kasim Terzić)
This lecture gave a brief overview of deep learning and how it relates to modern AI architectures such as transformers. We looked at
how deep neural networks can act as universal function approximators, and how this can enable them to perform tasks such as compression, semantic embedding, and next token prediction, techniques behind many of today’s advanced AI algorithms.
Wednesday 10 December 2025- Discussion session: Can we compute intelligence?
Following a series of lectures about various computational approaches commonly applied in sub-fields of AI, we gathered for an interdisciplinary discussion. Having looked at various aspects of intelligence including problem solving, knowledge, representation, learning, and reasoning, we looked at how the techniques discussed so far can (or cannot) be used to implement some of these aspects and behaviours.