The tool, called LogosXAI, is the work of Professor Krzysztof Michalik from WSB Merito University who has spent two years developing the LogosXAI system, a Polish tool designed to explain AI decisions.

XAI, or Explainable Artificial Intelligence, is a growing area of AI research focused on making algorithmic decisions transparent to humans.

Professor Michalik said: “With XAI, a doctor can see why an AI system proposes a given diagnosis and what data it relies on. This offers greater confidence in the technology and aids in clinical decision-making.

“In banking, XAI allows to understand why the system approves or rejects a loan. The customer and analyst see the criteria that have the greatest impact. In education, LogosXAI allows students to experimentally test how neural networks work: they can remove individual neurons and observe how the model's accuracy changes. This is learning by doing, which makes artificial intelligence no longer an abstraction.”

According to the Merito Center for Higher Education Development, most AI models currently function as a "black box," showing only inputs and outputs without revealing how conclusions are reached.

“LogosXAI allows us to partially explain the operation of deep neural networks — technologies that are extremely effective, but still opaque to users,” Michalik says in a press release.

The first scientific paper on the subject, “XAI and Hybrid Systems: Toward a Transparent AI for Education and Beyond,” was published in Intelligent Systems and Applications (Springer, Lecture Notes in Networks and Systems series) and presented at the 11th International Intelligent Systems Conference (IntelliSys) in Amsterdam in 2025. The paper shows that a hybrid approach to AI can combine the power of neural networks with improved interpretability.

A second paper, “AI for Education, Education of AI: Exploring the Role of Explainable AI (XAI) through LogosXAI,” appeared in the Lecture Notes in Computer Science series (Springer) and was presented at the 26th International Conference on Artificial Intelligence in Education (AIED 2025) in Palermo, Sicily.

"With LogosXAI, students can literally sneak a peak into a neural network and see how removing individual neurons affects the model's accuracy. This is experiential learning that facilitates understanding the principles of AI," Michalik says.

In both studies, Michalik describes his own method, HiNeS-P (High-Impact Neuron Selection – Pruning), which identifies and removes neurons that have the greatest influence on a network’s performance.

"In short, the idea is to identify the network elements that truly determine the outcome, and then simplify the model by removing less important neurons. This makes the network smaller, faster, and easier to understand, without significantly compromising quality," Michalik explains.

With the HiNeS-P method, the system not only reduces computational costs and enables AI to run on smaller devices, known as edge hardware, but also helps reveal which parts of the network are responsible for key decisions. "It is a bit like sifting through noise in a conversation — only what truly matters for the meaning of the message remains," Michalik says.

"Thanks to research on XAI, we have the opportunity to not only use its results but also understand how they are created. This is a step toward AI that we can trust because we see how it works,” he adds.

PAP - Science in Poland

zan/ agt/

tr. RL