The ALFIE (Assessment of Learning technologies and Frameworks for Intelligent and Ethical AI) project represents a groundbreaking initiative under the EU Horizon Europe programme aimed at democratizing access to ethical AI development through Automated Machine Learning (AutoML) technologies. This deliverable presents the initial definition of three trial use cases, comprehensive user requirements analysis, and the foundational validation framework for the ALFIE platform. The first pilot use case, led by Bosch, focuses on compliance alerts classification to automate the prioritization of Business Partner Screening alerts for regulatory compliance monitoring. The second pilot use case, coordinated by UAB, addresses website accessibility evaluation through the development of AI tools that automatically assess and improve web content accessibility for users with disabilities. The third pilot use case, managed by CTL, concentrates on ethical driver monitoring by creating unbiased emotion and drowsiness detection systems for autonomous vehicles. This use case addresses the critical challenge of mitigating demographic biases in facial recognition models to ensure equitable performance across diverse populations. Through a comprehensive User-Centred Design approach, the project systematically identified and prioritized requirements across two distinct user categories. The analysis revealed that a significant portion of end-user requirements focused on accessibility features, bias detection capabilities, and multilingual support functionality. These requirements consistently focused on transparency, explainability, and ethical AI behaviour as fundamental expectations for platform acceptance. The examination of the designer user requirements showed many requirements emphasized on data management capabilities, comprehensive bias detection and mitigation tools, and flexible model deployment options. Designer user requirements encompassed technical specifications for code generation, local development environments, and seamless integration with existing organizational systems, reflecting the more technically sophisticated needs of users responsible for creating AI solutions rather than consuming them. The ALFIE project implements a comprehensive three-phase validation framework designed to ensure systematic evaluation of platform capabilities, effectiveness, and trustworthiness across diverse application domains. This approach allows for continuous improvement while maintaining rigorous standards for technical performance, user acceptance, and ethical compliance. The initial validation phase focuses on establishing robust foundations through thorough data preparation, quality assessment, and baseline performance measurements using existing state-of-the-art solutions, demonstrating core AutoML capabilities. The second phase emphasizes real-world integration and user-centered evaluation through comprehensive prototype testing with diverse participant groups representing different backgrounds, technical expertise levels, and accessibility needs. Iterative refinement processes ensure that user feedback directly informs platform improvements, with particular attention to usability, accessibility, and trustworthiness factors. The final phase provides comprehensive evaluation under realistic operational conditions through full-scale demonstrations across all three trial scenarios. Performance validation employs both quantitative metrics aligned with established KPIs and qualitative assessments capturing user satisfaction, trust, and perceived value, while cross-scenario analysis identifies transferable insights for broader platform optimization.
Abad et al. (Thu,) studied this question.