Embedded Files
Unveiling the Inner Workings of GPAIS: Prediction and Explanation of Emergent Abilities in Neural Networks
International Joint Conference on Neural Networks - IJCNN 2026
June 21 - 26, 2026
Maastricht, The Netherlands
Aim and Scope
This special session is organized in alignment with the goals of the recently established IEEE CIS Task Force on General Purpose AI Systems (https://cis.taskforce.ieee.org/gpais), which recognizes that AI is undergoing a significant paradigm shift from narrow, task-specific models to general-purpose systems. This task force asserts that the Computational Intelligence (CI) community is uniquely positioned to address the critical challenges this transition presents, including robustness, alignment, and explainability. General Purpose AI Systems (GPAIS), particularly large-scale neural networks, have demonstrated a remarkable and often surprising characteristic: as they are scaled, they acquire qualitatively new and unpredictable capabilities. These "emergent abilities," such as in-context learning or multi-step reasoning, were not explicitly programmed but arose spontaneously from the interplay of architecture, data, and immense computational scale.
While these abilities are powerful, their unpredictability represents a fundamental scientific challenge and a significant barrier to safe and reliable deployment. We currently lack a rigorous framework for forecasting which new abilities will emerge, at what scale, and through which mechanisms. The central goal of this special session is to bring together researchers dedicated to transforming the study of emergence from a descriptive, post-hoc observation into a predictive and explanatory science.
To achieve this, we must look inside the "black box." Understanding why an ability emerges requires a deep dive into the internal computations of the network. This involves leveraging techniques from mechanistic interpretability to reverse-engineer the neural circuits that implement these new skills. Furthermore, to make these systems trustworthy, we need methods to translate these complex findings into human-understandable terms, a challenge addressed by Explainable AI (XAI). This session will provide a focused forum at IJCNN to present and discuss cutting-edge research aimed at predicting, analyzing, and explaining the emergent phenomena that define the frontier of modern neural networks.
Topics of Interest
This special session invites high-quality submissions that address the core challenge of understanding and predicting emergent abilities in GPAIS. We are particularly interested in work that moves beyond simply benchmarking performance and instead provides fundamental insights into the mechanisms driving these powerful, unexpected behaviors. We encourage contributions that leverage techniques from mechanistic interpretability to identify the specific circuits underlying emergent skills, as well as research that develops novel theoretical or empirical frameworks for forecasting the onset of new capabilities before they appear. Furthermore, we welcome work on XAI that is specifically tailored to making the reasoning processes of GPAIS transparent to developers and end-users. The goal is to foster a collection of research that collectively builds a more rigorous, predictable, and safe science of large-scale AI.
The list of potential themes includes, but is not limited to:
Prediction and Detection of Emergent Abilities:
Forecasting new capabilities using theoretical frameworks and scaling laws.
Systematic detection and causal analysis of emergent behaviors.
Analyzing the role of data and architecture in skill emergence.
Mechanistic Interpretability for Understanding Emergence:
Reverse-engineering the neural circuits that implement emergent abilities.
Analyzing information flow and internal representations to explain new skills.
Probing and identifying model components responsible for emergent behaviors.
Explainable AI for GPAIS:
Novel XAI methods for explaining the complex reasoning of GPAIS.
Assessment and mitigation of the risks posed by GPAIS to fundamental rights:
Methods to assess and mitigate risks to fundamental rights.
Design of a regulatory model to address ex ante the risks posed by GPAIS.
Participatory governance to forecast the capabilities of GPAIS and their integration in risk management cycles.
Submission and Publication Information
Paper Submission Deadline: January 31, 2026
Please submit your paper directly through the IEEE-WCCI 2026 submission website, selecting this special session as the main research topic.
For paper guidelines, please visit Information for Authors.
For submissions, please select the single topic "IJCNN SS44: Unveiling the Inner Workings of GPAIS: Prediction and Explanation of Emergent Abilities in Neural Networks" from the "Special Sessions" on the IEEE-WCCI 2026 submission website.
Submitted papers will be peer-reviewed with the same criteria as other IJCCN 2026 tracks.
The papers accepted for the special session will be included in the IJCNN 2026 proceedings and will be published by the IEEE Xplore Digital Library.
Committee
Dr. Leonardo Concepción
Dept. of Computer Science and Artificial Intelligence.
DaSCI, Andalusian Research Institute in Data Science and Computational Intelligence
University of Granada, Spain
Dr. Zacharoula Papamitsiou
SINTEF AS, 7034 Trondheim, Norway
zacharoula.papamitsiou@sintef.no
Prof. Dr. Athena Vakali
Department of Informatics
Aristotle University of Thessaloniki, Greece
avakali@csd.auth.gr
Andrea Palumbo
Centre for IT and IP Law and IMEC
Faculty of Law
KU Leuven, Belgium
Prof. Dr. Isaac Triguero
Dept. of Computer Science and Artificial Intelligence.
DaSCI, Andalusian Research Institute in Data Science and Computational Intelligence
University of Granada, Spain
See you in Maastricht!
Acknowledgements
This special session is part of the Project “Ethical, Responsible and General Purpose Artificial Intelligence: Applications In Risk Scenarios” (IAFER) Exp.:TSI-100927-2023-1 funded through the Creation of university-industry research programs (Enia Programs), aimed at the research and development of artificial intelligence, for its dissemination and education within the framework of the Recovery, Transformation and Resilience Plan from the European Union Next Generation EU through the Ministry for Digital Transformation and the Civil Service.
Page updated
Google Sites
Report abuse