Keynote Lectures

Abstract
Cyber-human systems are formed by the coordinated interaction of human and computational components. In this talk, I will argue that these systems can only be designed as trustworthy systems if the interoperation between their components is meaning preserving. For that, we need to take the challenge of semantic interoperability between these components very seriously. I will discuss a notion of trustworthy semantic models and defend its essential role in addressing this challenge. Finally, I will advocate that engineering and evolving these semantic models as well as the languages in which they are produced require a hybrid intelligence augmentation program resting on a combination of techniques including formal ontology, logical representation and reasoning, crowd-sourced validation, and automated approaches to mining and learning.

Abstract
Artificial Intelligence (AI) can bring competitive advantage to industry by improving system autonomy, decision support and the ability to offer higher value-added products and services. Delivering the expected service safely (conformance to requirements), meeting stakeholder expectations (trustworthiness, usability...) and maintaining service continuity will determine its adoption and use in industry. Moreover, concerns such as ethics, accountability, liability, security, privacy, and trust are receiving increasing attention in many industries. In addition, we see frenetic activity in standardization and regulatory bodies. For example, quality is the focus of the SQuaRE (Systems and software Quality Requirements and Evaluation) series of standards ISO/IEC 25000:2014, and AI quality is addressed in ISO/IEC DIS 25059. The principles of risk management are explained in ISO 31000:2018 and AI risk is specifically addressed in ISO/IEC FDIS 23894, and the High-Level Expert Group set up by the EU to advise on the European AI Strategy has published the European Commission's AI Act. A successful strategy to overcome these challenges requires collective actions around the objectives of a common industrial and reliable AI strategy to strengthen synergies and develop engineering best practices. The keynote will emphasize the importance of trustworthy AI engineering with a sound end-to-end methodology and tools to support the overall lifecycle of an AI system. This includes analyzing and meeting stakeholder expectations and specifications (such as regulation and standardization bodies, customers, and end-users) and assessing and managing AI-related risks to maintain trustworthiness in the system of interest, such as safety and security. The 'confiance.ai program' approach revisits conventional engineering, including data and knowledge engineering, algorithm engineering, system and software engineering, safety and cyber-security engineering, and cognitive engineering. The goal is to ensure the system's compliance with requirements and constraints, assess and master AI-technologies related risks, and maintain trustworthiness between stakeholders and the system of interest (e.g. RAMS - Reliability, Availability, Maintainability, and Safety - properties).

Abstract
What does software innovation look like? Using a series of well known examples, I’ll show you that it’s not about technology breakthroughs, or even radically new ideas. Most often, a software innovation succeeds because it offers a new piece of functionality (aka, a concept) that eliminates some constraint that made previous solutions much less attractive. I’ll explain how applications can be described as compositions of concepts and I’ll show how this idea can be put into practice in several ways: designing more usable software; aligning products across a company’s offerings; and generating the code of entire applications automatically with an LLM.