Welzijn.AI: A Conversational AI System for Monitoring Mental Well-being and a Use Case for Responsible AI Development

We present Welzijn.AI as new digital solution for monitoring mental well-being in the elderly, as a use case illustrating how recent guidelines on responsible Artificial Intelligence can inform Welzijn.AI's Technology and Value dimensions. Here Technology concerns the description of an open, well-documented and interpretable envisioned architecture in light of the system's goals; Value concerns stakeholder evaluations of Welzijn.AI. Stakeholders included, among others, informal/professional caregivers, a developer, patient and physician federations, and the elderly. Brief empirical evaluations comprised a SWOT-analysis, co-creation session, and user evaluation of a proof-of-concept implementation of Welzijn.AI. The SWOT analysis summarises stakeholder evaluations of Welzijn.AI in terms of its Strengths, Weaknesses, Opportunities and Threats. The co-creation session ranks technical, environmental and user-related requirements of Welzijn.AI with the Hundred Dollar Method. User evaluation comprises (dis)agreement on statements targeting Welzijn.AI's main characteristics, and a ranking of desired social characteristics. We found that stakeholders stress different aspects of Welzijn.AI. For example, medical professionals highlight in the SWOT analysis Welzijn.AI as the key unlocking an individual's social network, whereas in the co-creation session, more user-related aspects such as demo and practice sessions were emphasised. Stakeholders aligned on the importance of safe data storage and access. The elderly evaluated Welzijn.AI's accessibility and perceived trust positively, but user comprehensibility and satisfaction negatively. All in all, Welzijn.AI's architecture draws mostly on open models, as precondition for explainable language analysis. Also, we identified various stakeholder perspectives useful for researchers developing AI in health and beyond.