The 13th IEEE International Conference on Intelligent Systems

We are concerned with the smart environments – exemplified by smart homes, cities, etc. – which are in general “small worlds” composed – on the one hand – of sensors, computers, analytic and algorithmic tools and techniques, and – on the other hand – of human beings, individuals, groups and societies who are the “addressees” and possibly the beneficiaries.

First, we provide a survey of main developments in the field starting with the traditional techno-centric (technology driven) approach to smart environments in which just technical capabilities are accounted for. Then, we proceed to a more advanced human-centric approach in which the human needs, preferences, judgments, etc. are to be accounted for, to increase the comfort or well-being of the humans. Finally, the value-centric (driven by ethics, societal values, etc.) approach, with emphasis on societal impact and moral responsibility.

We show the power of AI (artificial intelligence) which can provide effective and efficient new analytic, algorithmic, decision making and reasoning models, using mostly data-driven machine learning based approaches. We start with a simple AI-assisted approach, and then, to deal with more complex problems, we proceed to the AI-enabled approach, notably in the decision support systems (DSSs) perspective. We advocate the use of the new, so-called AI-native ecosystem perspective in which AI solutions, notably related to sensors, actuators, data analytics, interoperability, privacy, security, trust, etc., are used right from the beginning of the smart environment conceptualization and development.

Finally, we present the idea of a new direction, the LLM-enabled approach, in which tools and techniques based on the use of the Large Language Models (LLMs) to implement natural language driven decision support and knowledge integration are employed.

An example from the critical infrastructure planning is shown for illustration.

As data increasingly drives research, innovation, artificial intelligence, and decision-making, the question is no longer simply whether data are available, but whether we can trust them. Data veracity—the extent to which data are accurate, reliable, consistent, traceable, and fit for purpose—is becoming a critical challenge in increasingly complex and dynamic data ecosystems.

In this talk, I will explore how data management needs to evolve to make veracity a first-class principle. Traditional approaches to data quality often focus on detecting errors and ensuring consistency at a particular point in time. Yet in modern data environments, data are continuously generated, transformed, integrated, enriched, and reused. Their meaning and reliability can change along the way, while emerging AI systems introduce new forms of uncertainty and make it increasingly difficult to distinguish reliable information from plausible but incorrect outputs.

The talk will connect these challenges to practical examples from real-world flexible query answering, decision support systems and AI-supported data analyses. These examples will illustrate the need for adequate data veracity assessment and handling techniques. As improving data veracity is not always possible, stakeholders should at least be clearly informed about the confidence they can have in the results of their querying, decision-making and data analysis processes as informed use and reuse definitely became an undeniable prerequisite in current and future AI-based information systems.

I will discuss and illustrate how computational intelligence can contribute to making data—and the processes that produce and transform them—better explainable and more transparent and trustworthy. Rather than treating veracity as a property that can simply be assigned to a dataset, I will argue that it should be understood as a dynamic, contextual, and traceable characteristic of the entire data lifecycle.

Building future-proof data management therefore means designing systems that do not merely preserve data, but preserve our ability to understand, evaluate, and trust those data as contexts, technologies, and uses evolve.

Machine Learning has recently established itself as the main approach to designing and implementing AI. However, in spite of the significant improvement in the accuracy of predictions, most ML models suffer from lack of interpretability and the recommendations made by most AI systems suffer from lack of explainability. This is a significant problem especially for safety critical applications where wrong recommendations from an AI system may have serious consequences.

The presentation will highlight recent developments in Explainable AI. It will also focus on Surrogate Modelling as a promising research trend that is aimed at finding a good balance between accuracy and explainability of machine learning models used in AI systems. This balance guarantees reliable and understandable solutions that improve some AI features such as safety and trustworthiness. These features can facilitate and speed up the successful adoption of AI on a wider scale.