Dr. Osnat (Ossi) Mokryn
Senior Lecturer, Department of Information Systems
Herta & Paul Amir Faculty of Social Sciences
Field of Research: Statistical Inference in AI, Human-AI Collaboration, Cognitive Processes
Teaching AI the Power of Surprise
Can a machine be surprised?
“Teaching machines to learn from surprises, as humans do, could transform the way we train artificial intelligence – making AI learning more efficient and more explainable.”
– Dr. Osnat Mokryn
The Project
AI today requires vast amounts of data to learn. Dr. Mokryn’s Learning via Surprisability (LvS) is a novel method that enables AI systems to form expectations and learn from the unexpected, much like humans do.
- We have hypothesized that LvS mirrors human cognitive processes, identifying discrepancies between new samples and expectations.
- This method parallels the brain’s predictive mechanisms, where mismatches between expectation and reality trigger attention and adjustment.
- Applications span from social media analysis and historical data mining to biomedical research, with the potential for more efficient and explainable AI systems.
Fundraising Goals
Your support will advance the development and application of the novel method of Learning via Surprisability (LvS) in AI:
- Enabling more efficient, lower-resource AI learning across multiple domains.
- Addressing a critical question in medical learning: Can an algorithm trained on one population generalize effectively to another?
- Applying LvS to biomedical data, where training on multiple sources could extend to unseen subpopulations.
Meet Osnat Mokryn
A first-generation academic with a BSc and MSc in computer engineering from the Technion-Israel Institute of Technology, and a PhD in computer science from the Hebrew University of Jerusalem. Throughout my studies, I worked in parallel, gaining extensive experience in the high-tech industry and developing expertise in managing and designing highly complex systems.
Today, I head the Social, AI and Networks (SCAN) Laboratory at the University of Haifa. We research creativity and cognitive aspects of human-AI decision-making and develop statistical inference methods for modeling and mining high-dimensional data. Applications include decision-making under stress, recommender systems, epidemiology, time series, and temporal archival data.
At the heart of my research is the development of a novel method for Learning via Surprisability (LvS). Inspired by human cognitive processing of surprisability, LvS is a characterization method that captures meaningful information and provides interpretable results aligned with human reasoning. It has already been applied to authorship and impersonation detection on social media, topical historical archival mining, time series analysis, and the identification of novel immunological and biological information.
