HIDA Lecture Series on AI

The HIDA Lecture Series on AI offers valuable insights into the transformative role of AI. With a focus on Health, Ethics, and Public Administration, the lecture series shows how AI is revolutionising our lives in a variety of ways while also raising complex moral and ethical issues.

All speakers are highly qualified researchers from the Helmholtz Association and its partners.

Watch again: Past Lectures

• Clinical Large Language Models in Healthcare – Applications, Challenges, and Opportunities Learn more
• Large-Scale Brain Decoding - Taking Advantage of Physiological Diversity Learn more
• Machine Learning for Precision Medicine: Avenues and Roadblocks Learn more
• The Ethics of AI: Foundations of Applied Ethics and Emerging Risks from AI Development Learn more

Clinical Large Language Models in Healthcare – Applications, Challenges, and Opportunities

Abstract: 

This presentation focuses on the development and application of large language models (LLMs) in the Norwegian healthcare sector. Language models such as NorDeClin-BERT and NorBERT are already being used for tasks like medical data coding, the detection of adverse drug reactions, and the automated anonymization of patient texts.

Key challenges include the limited availability of annotated datasets, strict data protection regulations, and potential algorithmic biases. Current research efforts aim to address these issues through the use of multimodal AI systems, federated learning approaches, and synthetically generated data.

The overarching goal is to drive innovation in healthcare while maintaining high standards of data privacy and patient safety. 

Large-Scale Brain Decoding - Taking Advantage of Physiological Diversity

Abstract:

Decoding neural activity is crucial for neuroscience applications, from sensorimotor research to clinical diagnosis and brain-computer interfaces. Traditional statistical methods struggle with the complexity of brain data, making deep learning essential. However, neural data is highly heterogeneous, posing challenges for model training across diverse subjects, modalities, and conditions.

Blake Richards’ lecture presents novel tokenization techniques for training large-scale transformer-based neural decoding models. These models effectively integrate diverse data—across species, brain regions, and behaviors—improving performance as data diversity increases. This suggests that scaling neural decoding is feasible when models are designed to harness the richness of neural datasets.

Machine Learning for Precision Medicine: Avenues and Roadblocks

Abstract:

The lecture discusses the transformation in clinical neuroscience: Instead of the previous approach of comparing markers of brain structure, function or connectivity between groups or correlating them with behavioural phenotypes, advances in data availability and in multivariate statistical learning methods now enable the prediction of individual cognitive or clinical phenotypes. This development has initiated a revolution towards precision medicine.
However, the lecture highlights that the transition from proof-of-concept studies to practical applications might be more complex than often assumed. Technical and biological challenges, such as the limited dimension of biological variability, can compromise the power of predictions.

At the same time, ethical, legal and societal aspects represent significant hurdles for the practical implementation of such technologies and make it necessary to take these into account to a greater extent when developing new approaches in order to go beyond feasibility studies.

The Ethics of AI: Foundations of Applied Ethics and Emerging Risks from AI Development

Abstract:

Artificial intelligence shapes science, society, and everyday life, but it also introduces new ethical challenges. This lecture provides an introduction to the foundations of AI ethics, exploring sociotechnical systems and core principles such as autonomy, justice, and non-maleficence.

Using medical case studies, the lecture examines issues like bias, model opacity, and accountability. Theresa Willem concludes with strategies such as "embedded ethics," a proactive approach to integrating ethical considerations into the development of AI systems.