• AI democratization
• Machine Learning
• Participation
• Visual Programming Languages

Artificial Intelligence (AI) is propelling society towards an increasingly algorithmic future, bringing profound transformations across industry, academia, and social dynamics. By leveraging data-driven predictive analytics, AI boosts operational efficiencies and sparks innovation in product development and decision-making processes. Central to AI's evolution are Machine Learning (ML) systems, whose integration into existing processes marks a significant paradigm shift. However, the future impact of these systems on society remains both promising and uncertain due to varying performance across different domains. Moreover, the technical knowledge required to design and manage ML-based systems is extensive, leading to growing complexity that often confines these systems to the exclusive domain of computing experts. As a result, domain specialists or decision-makers are frequently excluded from the design process, relying heavily on the expertise of technical professionals. Acting merely as end-users, their understanding and trust in ML decision-making mechanisms are limited, raising ethical concerns.

On the one hand, computing experts lack in-depth knowledge of the domains their applications impact, while on the other hand, domain specialists are aware of their specific needs but lack the technical expertise to contribute effectively to system design. Research in AI democratization aims to bridge this divide by promoting broader user participation and collaboration to integrate technical expertise with domain-specific insights. Achieving this balance is crucial for ensuring ethical decision-making processes. Tools like Visual Programming Languages (VPLs) and no-code platforms can transform end-users from passive consumers into active contributors in the design process by simplifying ML complexities. When supported by Natural User Interfaces (NUIs), these tools can reduce user cognitive load and further enhance accessibility for novices through natural and intuitive interactions.

This thesis introduces PyFlowML, and its version for touch tabletops. By incorporating Explainable AI (XAI) techniques, it aims at exploring how VPLs can foster participation and collaboration of both domain specialists and computing experts in the design of ML-based systems. The research includes a systematic literature review, expert and user-based testing, comparative analysis, and hypothesis testing. This work yielded promising results, demonstrating that VPLs can enhance novices' understanding of ML decision-making, foster broader user participation and collaboration, and ultimately contribute to AI democratization.