Tutorial 1 - ICEDEG 2025

Title: Fundamentals of (and Tools for) Trustworthy Artificial Intelligence in Smart Health

Outline of the Tutorial

Artificial Intelligence (AI) is pervading many aspects of our society. This poses challenges to avoid people being put aside when their own data are processed by AI systems, which provide decisions that may result in harmful discrimination. Our focus is on knowledge representation and how to enhance human-centered information processing in the context of Trustworthy AI. Endowing AI with trustworthiness encompasses technical and non-technical challenges. In this tutorial, in addition to technical aspects (i.e., disruptive human-centered technologies as well as human-friendly computer tools aimed at covering all phases of the design, analysis, and evaluation of trustworthy intelligent systems), we will also discuss Ethical, Legal, Socio-Economic and Cultural (ELSEC) implications of AI. Special emphasis will be placed on certifying if intelligent systems comply with European values. Assuming explainability as a prerequisite for trustworthiness, Explainable AI (XAI in short) is an endeavor to evolve AI methodologies and technology by developing intelligent systems capable of generating decisions that a human can understand, but also capable of explicitly explaining their decisions. This way, it is possible to scrutinize the underlying intelligent models and verify if automated decisions are made based on accepted rules and principles so that decisions can be trusted and their impact justified. Accordingly, intelligent systems are expected to naturally interact with humans, thus providing comprehensible explanations of decisions automatically made. Even if this tutorial will introduce the main concepts and methods in the context of XAI in general, a major focus will be on how to properly deal (and compute) with words and perceptions in generating and evaluating textual explanations for smart health. More precisely, we will consider the explainable design of Fuzzy Sets and System in combination with pre-trained Large Language Models for paving the way from interpretable machine learning to Trustworthy AI. Such systems deal naturally with uncertainty and approximate reasoning (as humans do) through computing with words and perceptions. This way, they facilitate humans to scrutinize the underlying intelligent models. Moreover, human-AI interaction is natural and faithful.

The tutorial is organized as follows:

• Unit 1. Explainable Artificial Intelligence (XAI)
  • Introduction: Concepts and Nomenclature
  • Data Explainability and Model Explainability
  • Generation and Evaluation of Explanations
  • Resources: datasets and software tools (XAI in the Lab: Hands-on Session)
• Unit 2. Trustworthy Artificial Intelligence (TAI)
  • Introduction: Concepts and Nomenclature
  • Ethical Guidelines and Assessment List for TAI
  • Detection and prevention of biased data and models
  • Legal, Socio-economic, and Cultural implications of AI
  • Resources: datasets and software tools (TAI in the Lab: Hands-on Session)
• Unit 3. Trustworthy Smart Health Applications
  • Active and Healthy Aging

Targeted audience, prerequisite knowledge:

 This tutorial is of interest for researchers, practitioners, and students (PhD, MSc, BSc, or undergraduate students) working in the field of AI and healthcare. Attendants are expected to participate actively in interactive hands-on sessions with their laptops. Most software will run online, but instructions for installing software, if needed, will be provided in advance.

Importance of this topic for the ICEDEG community:

This tutorial is appropriate for the ICEDEG community because the use of AI in Healthcare, especially in clinical decision-making but also in Health Management and Policy recommendations, is becoming more prevalent and impacting citizen’s quality of life. Moreover, the use of AI in Healthcare can have significant moral and ethical implications, potentially exacerbating existing health disparities. Despite this, ethics is often an afterthought in the model development process, a practice that is at odds with the fundamental principle of medicine: to do no harm. The main aim of this tutorial is to equip AI practitioners in positioning foundational ethics as an integral part of the model development process, rather than a quality that models are evaluated on post-hoc. We will address technical challenges in embedding ethical principles into the very genesis of human-centered model development, representing a paradigm shift from measuring and mitigating impacts to preventing them; thus contributing to leveraging Trustworthy AI.

History of prior tutorials from the instructor(s):

• “Responsible and Trustworthy AI for Healthcare”, IEEE BHI 2024 (Houston, USA)
• “Using Fuzzy Sets and Systems for Explainable Artificial Intelligence - How and Why”, IEEE WCCI 2024 (Yokohama, Japan)
• “Fundamentals of (and tools for) Trustworthy Fuzzy Systems”, SFLA 2024 (Toledo, Spain)
• “Fuzzy Sets and Systems for Detecting and Mitigating Hallucinations in Large Language Models”, SFLA 2024 (Toledo, Spain)
• “Explainable Fuzzy Systems”, EurAI ACAI 2022 (Barcelona, Spain)
• “Explainable Fuzzy Systems: Paving the way from Interpretable Fuzzy Systems to Explainable Artificial Intelligence” FUZZ-IEEE 2021 (Luxembourg)
• “Paving the way from Interpretable Fuzzy Systems to Explainable Artificial Intelligence Systems”, IEEE WCCI 2020 (Glasgow, Scotland)