Introduction

Human-interpretable AI models [1] are playing an increasingly important role in Artificial Intelligence (AI). Today, a large part of the technologies employed by AI and SIGKDD researchers is based on Deep Neural Networks (DNNs). Yet, the lack of transparency of DNNs prevents a safe deployment of these models in critical contexts that significantly affect users. Consequently, decision-making systems based on deep learning are facing constraints and limitations from regulatory institutions [2], which increasingly demand transparency in AI models [3]. Even though standard eXplainable AI (XAI) emerged to address the need to interpret DNNs, several works are arguing that it may not have achieved its goal [4,5].

To really explain DNN decision-making process, there is a growing consensus that human-interpretable explanations are required. Human-Interpretable AI (HI-AI) methods either provide post-hoc explanations by extracting the symbols that have been automatically learnt by the models (e.g., T-CAV [6]), or directly design intrinsically interpretable architectures (e.g., CBM [7]). Among other qualities, these explanations resemble better the way humans reason and explain [8], help to detect model biases [9], are more stable to perturbations [10], and can create more robust models [11].

Workshop Topics

Topics of interest include, but are not limited to, the following:

• Explainable-by-design models, novel approaches to creating machine learning and deep learning models that are intrinsically explainable or interpretable. • Post-hoc methods for Interpretable AI, novel approaches on post-hoc interpretable AI. These include but are not limited to approaches working on higher-level features such as concepts.

• Theoretical analyses of existing methods, showing what

existing interpretable methods can achieve both from an explanation and a generalization point of view.

Program

This workshop aims to advance the research on HI-AI by offering a diverse program designed to enhance participants' knowledge, and foster collaboration and innovation. The following list contains the invited speakers who will give keynote talks at the HI-AI workshop, and the expected topics that their talks will cover. All invited speakers have already confirmed their presence. Program Outline. Table 1 reports the workshop program. Firstly, we will give an overview of the key aspects of HI-AI to ensure all attendees have a solid understanding of the background concepts and terminology. Secondly, the workshop features three invited talks from experts in the field, who will share their insights and latest research findings. These talks will provide valuable perspectives and inspire new ideas. Thirdly, we will offer participants the chance to engage in dialogue with prominent scientists during a long coffee break with poster presentations, encouraging collaborations and knowledge-sharing. Also, the workshop program includes three contributed talks from selected contributions. We will recognize the most interesting contribution with a Best Workshop Paper Award. We have allocated 40 minutes for each invited talk, allowing for a 30-minute presentation followed by a 10-minute Q&A session. We allotted the same time for the poster sessions.

Paper Management

Paper management. We published the Call For Papers (CFP) on the workshop website 1 . The CFP focuses on short papers, which can be research papers, theoretical analysis papers, or vision papers. Table 1: Draft of the program outline.

In the case of research contributions, we asked paper authors to make their code and data openly available to ensure reproducibility.

The review process has been double-blind. We have used OpenReview to ensure the final decisions for each paper are made by the organisers with no conflict of interest. All accepted papers will be published on the workshop website, which will remain active and accessible after the conference concludes. Additionally, we took contact with an external editor (CEUR-WS) to create an archival version of these papers for authors who wish to participate in a subsequent publication.