Welzijn.AI: Developing responsible conversational AI for the care of older people through stakeholder involvement

Bram M A van Dijk; Armel E J L Lefebvre; Marco R Spruit

doi:10.1016/j.maturitas.2025.108616

Welzijn.AI: Developing responsible conversational AI for the care of older people through stakeholder involvement

Maturitas. 2025 May 23:199:108616. doi: 10.1016/j.maturitas.2025.108616. Online ahead of print.

Authors

Bram M A van Dijk¹, Armel E J L Lefebvre¹, Marco R Spruit²

Affiliations

¹ Health Campus The Hague, Leiden University Medical Center, Turfmarkt 99, 2511 DP The Hague, The Netherlands.
² Health Campus The Hague, Leiden University Medical Center, Turfmarkt 99, 2511 DP The Hague, The Netherlands; Leiden Institute of Advanced Computer Science, Leiden University, Einsteinweg 55, 2333 CC Leiden, The Netherlands. Electronic address: m.r.spruit@lumc.nl.

PMID: 40446629
DOI: 10.1016/j.maturitas.2025.108616

Abstract

Objectives: We present Welzijn.AI, a new digital solution for monitoring (mental) well-being in older populations, and illustrate how the development of systems like Welzijn.AI can align with guidelines on responsible AI development.

Study design: Three different evaluations with various groups of stakeholders were designed to disclose new perspectives on the strengths, weaknesses, design characteristics, and value requirements of Welzijn.AI. Evaluations consisted of expert panels with patient representatives, general practitioners, researchers, and older people themselves. Panels were involved in interviews, a co-creation session, and an evaluation of a proof-of-concept implementation of Welzijn.AI.

Main outcome measures: Interview results were summarized in terms of Welzijn.AI's strengths, weaknesses, opportunities and threats. The co-creation session ranked a variety of value requirements of Welzijn.AI with the Hundred Dollar Method. The proof-of-concept evaluation entailed analysing proportions of (dis)agreement on statements targeting Welzijn.AI's design characteristics, and ranking desired social characteristics.

Results: Stakeholders in the interviews acknowledged Welzijn.AI's potential to combat loneliness and extract patterns from the linguistic behaviour of older people. The proof-of-concept evaluation complemented the design characteristics most appealing to older people to achieve empathetic and varying interactions. Stakeholders linked the technology to the implementation context: Welzijn.AI can unlock an individual's social network, but practice sessions and continuous support should also be available to empower users. Yet, stakeholders also disclosed challenges to a proper understanding of the application, and issues concerning privacy were also highlighted.

Conclusions: Incorporating all stakeholder perspectives in system development remains challenging. Still, our results benefit researchers, policymakers, and health professionals who aim to improve the care for older people through the use of technology.

Keywords: Artificial intelligence; Conversational AI; Geriatric care; Language biomarkers; Large language models; Remote monitoring; Responsible AI; Well-being.