Conference on Informatics and Information Technologies (CIIT).

Seventy years after the Dartmouth workshop, AI paradigms have evolved dramatically. Yet, despite the superhuman performance of foundation models on specialized tasks, robust commonsense reasoning remains an elusive challenge. This talk provides a framework for commonsense reasoning, bridging AI, cognitive psychology, and linguistics. I will examine the evolution of AI models through the lens of abstraction tasks, such as drawing analogies in narratives, understanding causality in videos, and decoding brain teasers. Specifically, I will discuss methods that combine neural learning with structural reasoning—such as entailment trees and structural mapping engines—which show promise in addressing these complexities. Finally, I will critically reflect on the term 'commonsense AI' itself, challenging the assumption that such knowledge is universally shared.

Filip Ilievski is an Assistant Professor at Vrije Universiteit (VU) Amsterdam and an affiliated scientist at USC’s Information Sciences Institute. He is also a member of the ELLIS Unit Amsterdam Management Team and a scientific coordinator at the Digital Sustainability Center (DiSC). Dr. Ilievski’s research investigates the development of human-centric AI, specializing in commonsense reasoning, neurosymbolic methods, and analogy. He currently leads the NWO-funded "Human-Centric AI with Common Sense" project, where he collaborates with cognitive and communication scientists to bridge the gap between neural learning and structural reasoning, and between theories and computational methods.

With over 100 peer-reviewed publications and two books, Dr. Ilievski has developed foundational resources for the field, including the CSKG (a consolidated graph of commonsense knowledge), multimodal benchmarks, and novel methods for visual abstraction. He is an active organizer of workshops and tutorials at major AI conferences like AAAI, ACL, and IJCAI. Recently, his work highlighting the challenges of visual abstraction in AI was featured in a BBC series.

In spite of being one of the most advanced industry in the planet, novel approaches from Artificial Intelligence and Deep Learning are seldom applied in air transport. The natural question is, why? In this talk we will review some solutions that have been proposed to integrate AI and DL approaches in air transport and air traffic management, from delays' prediction to separation assurance; and, more importantly, where and why they fail. Using some research works done in my group as starting point, we will discuss how both fields can be merged: in which topics can air transport get a real benefit, and which properties the solutions need to have to be operationally feasible.

Massimiliano received his Ph.D. degree in Electrical and Computer Engineering from the Universidade Nova de Lisboa, Portugal, in 2014. He is currently a researcher at CSIC-IFISC, where he coordinates the ERC Starting Grant ARCTIC. With more than 140 peer-reviewed publications in international conferences and journals, he has extensive experience in data mining and machine learning research, covering both theory and applications, and has collaborated with scientists from all over the world. His main research interests include Data Science and its application to real-world problems, including the modelling and understanding of air transport systems and the analysis of complex biomedical datasets. He is a member of the editorial team of Nature Scientific Reports, the European Journal of Social Behavior, PeerJ, PeerJ Computer Science, and Complexity.

I will address the recent trends and open problems at the intersection of modern Cryptography and the disruptive field of Artificial Intelligence. Then I will describe recent work conducted with two of my PhD students on defining a framework for performing a full set of cryptographic operations over the message space provided by modern LLMs. We proposed a novel cryptographic embedding framework that enables covert Public Key or Symmetric Key encrypted communication over public chat channels with human-like produced texts. Some unique properties of our framework are: 1. It is LLM agnostic, i.e., it allows participants to use different local LLM models independently; 2. It is pre- or post-quantum agnostic; 3. It ensures indistinguishability from human-like chat-produced texts. Thus, it offers a viable alternative where traditional encryption is detectable and restricted.

Danilo Gligoroski got the Ph.D. degree in computer science from Ss Cyril and Methodius University in Skopje in 1997. From 1997 to 2008, he was an Assistant Professor in the Faculty of Natural Sciences at Skopje University. Since 2008, he is a Professor of information security and cryptography with the Norwegian University of Science and Technology (NTNU). He is an author of more than 180 scientific publications and more than 10 inventions.

His main research interests are in the application of various algebraic structures in cryptography, information security, and coding theory.