UiB AI #10: Theory, humans and natural sciences

Welcome to our next AI seminar! Topics cover theoretical research in the core of AI, studies interplaying between human learning and machine learning and application of AI methods.

Chair: Pål Grønås Drange, Associated Professor at the Department of Informatics

Program:

Fabio Massimo Zennaro (Department of Informatics): Causal reasoning and causal abstraction

This talk will touch on two aspects of human reasoning that are crucial for AI: causal reasoning and multi-level reasoning. We will discuss the role and the relevance of causal models in machine learning systems, as well as the ubiquity of multi-level reasoning across applications. We will then review some recent approaches based on causal abstraction, aimed at integrating causal reasoning and multi-level reasoning. We will conclude by suggesting possible developments and deployment of these methods. 

Asieh Abolpour Mofrad (Department of Informatics): Interfacing Psychology and AI: A Journey from Human Learning to Machine Adaptation

This talk explores the intersection of psychology and artificial intelligence, particularly focusing on the formation of stimulus equivalence classes--a key aspect of symbolic behavior, language,  and cognition. We will journey through the historical foundations of stimulus equivalence, its significance in understanding human behavior, and how computational models, specifically reinforcement learning-based models like Projective Simulation, can advance our understanding of this phenomenon. Furthermore, we'll mention recent enhancements in computational modeling, such as Enhanced Equivalence Projective Simulation (E-EPS), and their implications for studying learning mechanisms in both humans and artificial agents.

Asgeir Sorteberg (Geophysical Institute): AI in weather and climate research – The clash of two cultures

The presentation will highlight advancements in applying machine learning to weather and climate research. Beginning with an overview of the existing framework of weather and climate models, I will focus on recent advances using AI. Finally, I will explore the emerging opportunities and possible limitations in the use of AI within weather and climate research.

Iain Johnston (Department of Mathematics): Inferring accumulation pathways from pathogen evolution to disease progression

Many important processes throughout the sciences and medicine can be viewed as the serial accumulation of discrete features over time. These features may be, for example, different clinical symptoms as a patient's disease progresses, different genetic mutations as a tumour develops, different drug resistances as bacteria evolve, and so on. We are often interested in using data to learn about these processes. Does presenting with symptom X early mean a patient is high risk? Does mutation A make it more likely than mutation B will occur? If a bacterium is resistant to drugs 1, 2, and 5, what drug will it evolve resistance to next? What are the typical progression pathways for a given system, and how much do individuals take different, personal journeys through this space?

I'll talk about our work developing "hypercubic inference", a flexible class of methods for learning answers to these, and more, questions using data of different types and structures. I'll illustrate it with some examples from ongoing projects, which may include disease progression and risk stratification in severe malaria patients, multi-drug resistance in tuberculosis, and evolution of ovarian and blood cancers.

Jarl Giske (Department of Biological Sciences): A salmon makes up its mind

Behavioral choices in vertebrates (including fish and people) are based on the ability of episodic memory and that this can be used to predict the emotion a choice will lead to. Vertebrates choose the behavior that maximizes expected wellbeing. Behaviour and wellbeing are thus two sides of the coin for fish (and men). This can be used in a digital twin of a salmon, which in turn can become an honest broker between fish farmers, authorities and consumers about the condition of a farm or about the wellbeing of a salmon before it became a fish fillet. This requires a lot of process modelling and machine learning. This is a mixture of basic research, innovation and societal relevance.