A Theory for AI Computation?
Ana Ozaki, University of Oslo
Artificial Intelligence (AI) has become ubiquitous in our everyday lives. There has been intense research on tasks related to language generation and understanding and, as a result of the improvements in these tasks (see e.g. BERT, (chat)GPT, and Bard), AI has been applied to enhance user experience in household appliances, virtual personal assistants, search engines, recommender systems, among others. A big portion of articles on machine learning are based on experimental evaluations, where the goal is often to provide evidence of an improvement of accuracy results on benchmarks. While the advance of application-driven research in AI at such a pace is exciting, we also envision an urgent need of establishing and standardizing a formalization of certain notions used in the field, starting with basic conceptual questions. What is a learning algorithm? What does it mean (formally!) to say that it has learnt to perform a certain task? In this talk we cast light on abysmal literature gaps between AI and foundational research. We also highlight contributions made by multiple authors towards filling gaps.
Ana Ozaki is an associate professor at the University of Bergen and at the University of Oslo, Norway. She is an AI researcher in the field of knowledge representation and reasoning and in learning theory. Ozaki is interested in the formalization of the learning phenomenon so that questions involving learnability, complexity, and reducibility can be systematically investigated and understood. Her research focuses on learning logical theories formulated in description logic and related formalisms for knowledge representation. She is the principal investigator of the project Learning Description Logic Ontologies funded by RCN.
Endrocrine digital twin as an example of Medical AI
Helge Ræder, University of Bergen
The use of artificial intelligence is becoming increasingly prevalent in medical research and will continue to push the boundaries of what medical research is, and can do. Hormones are essential signaling chemicals that coordinate the function of the body’s various organs. They are secreted in pulses with distinct rhythms which may be ultradian (24h) or circadian (diurnal), and modulated by external factors such as sleep, food intake, physical activity and stress. Endocrine disorders arise when these rhythms are changed. A human digital twin is a virtual representation of a real person, created by collecting and analyzing data from various sources such as medical records, wearable devices, social media activity, and other digital interactions. An human endocrine virtual twin model would build on a spectrum of implemented, demonstrated, validated and early phase real-world data of the hormonal system, including clinical, device-based and other disease and well-being related data. The digital twin can be used to simulate and predict a person's behaviour, preferences, health, and other attributes, allowing for personalized and data-driven decisions.
Helge Ræder is a professor at the Clinical Institute 2 at the University of Bergen. He is a Vice Dean for Innovation at Faculty of Medicine and a Consultant Pediatrician. His research aims to characterize and better understand the major signaling pathways involved in endocrine disease.
AI and Games
Peter Wingaard, CEO Rain AS
This talk is about game AI. Largely it is a history of pretending to be smart.
In games, we do that with everything. "Will people think this random distortion is water?" "Does this look like real light to you?" "Does the enemy feel smart enough to be human, yet dumb enough to stick their head out for you to hit with a shot at regular and predictable intervals?"
Our AI's are no different.
Peter has been running the game studio Rain in Bergen since 2010, growing the company to 20 people.
With a lifelong passion for game design he has kept a hand on the technical aspects of game creation.