Designing Positive AI

Abstract

The rise of Artificial Intelligence (AI), from curatorial AI in YouTube to generative AI in ChatGPT, demonstrates both potential for progress and risks of harm. Adopting a Positive Design approach aimed at directly enhancing human wellbeing, this dissertation develops the concept of Positive AI. It explores the role of designers in steering AI innovation towards holistically supporting flourishing, not just optimizing profit or user engagement. Through human-centered methods, this research seeks to advance understanding and techniques for assessing and iteratively improving AI's impact on wellbeing.Key questions address how wellbeing manifests in AI systems, how it can be measured, how to design interventions, and how to evaluate them. Outcomes include conceptual frameworks, case studies demonstrating approaches, proposed methods, and evaluations, aimed at laying a robust foundation for AI that fosters human flourishing. A cybernetics perspective organizes the challenges for designing AI for wellbeing, emphasizing the importance of feedback loops connecting assessments and actions.A longitudinal case study at Delft University of Technology presents "My Wellness Check," a cybernetic system for community wellbeing during COVID-19. The project, spanning two years and engaging 20,311 participants, demonstrates the application of cybernetic principles in a complex sociotechnical context. Building on these insights, a novel method is developed to systematically integrate wellbeing into AI design through distinct phases, from contextualizing wellbeing needs to continuously aligning AI behavior with wellbeing goals. The method's effectiveness is exemplified through diverse student projects, with expert evaluations providing evidence of its practicality and efficacy.Finally, the research synthesizes insights into a set of recommendations, charting concrete next steps for researchers and practitioners across fields to further mature these nascent perspectives and capabilities towards Positive AI. Key recommendations include integrating human-centered design methods, balancing immediate desires with long-term wellbeing, contextualizing wellbeing through participatory processes, establishing multidimensional feedback loops, shifting from harm mitigation to actively cultivating flourishing, and embracing Positive AI as an ongoing process. Through multifaceted efforts spanning advocacy, policy, and community building, the Positive AI agenda can progressively guide innovation trajectories towards enhancing societal wellbeing. While further work is needed to fully realize its potential, this dissertation makes important strides in laying the groundwork for AI that actively prioritizes human flourishing through integrative, collaborative design.