Welzijn.AI: Developing Responsible Conversational AI for Elderly Care through Stakeholder Involvement

TL;DR

Welzijn.AI presents a conversational AI for monitoring elderly well-being by querying EQ-5D-5L domains and analyzing language biomarkers, with responses summarized into clinical-style scores. The work follows the CEHRES roadmap, conducting three stakeholder evaluations (expert panels, co-creation, and an elderly proof-of-concept) to elicit requirements and gauge acceptance. Findings reveal potential to reduce loneliness and reveal behavioral patterns, but also highlight challenges in goal clarity, privacy, and user comprehension, prompting mitigations such as demos and help desks. The paper provides practical guidance for researchers, policymakers, and health professionals seeking to develop responsible AI for elderly care that meaningfully integrates stakeholder perspectives and context.

Abstract

We present Welzijn.AI as new digital solution for monitoring (mental) well-being in elderly populations, and illustrate how development of systems like Welzijn.AI can align with guidelines on responsible AI development. Three evaluations with different stakeholders were designed to disclose new perspectives on the strengths, weaknesses, design characteristics, and value requirements of Welzijn.AI. Evaluations concerned expert panels and involved patient federations, general practitioners, researchers, and the elderly themselves. Panels concerned interviews, a co-creation session, and feedback on a proof-of-concept implementation. 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. User evaluation comprised analysing proportions of (dis)agreement on statements targeting Welzijn.AI's design characteristics, and ranking desired social characteristics. Experts in the panel interviews acknowledged Welzijn.AI's potential to combat loneliness and extract patterns from elderly behaviour. The proof-of-concept evaluation complemented the design characteristics most appealing to the elderly to potentially achieve this: empathetic and varying interactions. Stakeholders also link the technology to the implementation context: it could help activate an individual's social network, but support should also be available to empower users. Yet, non-elderly and elderly experts also disclose challenges in properly understanding the application; non-elderly experts also highlight issues concerning privacy. In sum, incorporating all stakeholder perspectives in system development remains challenging. Still, our results benefit researchers, policy makers, and health professionals that aim to improve elderly care with technology.

Figures (4)

• Figure 1: Welzijn.AI comprises modules for interaction (blue) and analysis (orange). Regarding interaction, the transcription module concerns an open weights Whisper model radford2023robust that transcribes user speech to text as input to the LLM module. The Large Language Model module supports the conversation with the open weights Llama3.3-70B model grattafiori2024llama, prompted to structure the conversation around the EQ-5D-5L. The speech module is an open weights CoquiTTS VITS model that converts LLM responses to natural speech kim2021conditional. Regarding analysis, user input in text and audio formats are analysed in the speech/text analyser module. Text is also input to the summarizer module, a Llama3.3-70B model that summarizes user responses into a score.
• Figure 2: The CEHRES roadmap (reproduced from van2011holistic) involves five phases. In short, Contextual Inquiry discloses the contexts of prospective stakeholders of the system; Value Specification entails translating stakeholder values into user requirements; Design involves the (co-)creation of prototypes with stakeholders; Operationalisation involves bringing the technology to the market, and the Summative Evaluation concerns an evaluation of the technology's actual use. In this work we focus mainly on the first two phases.
• Figure 3: Proof-of-concept static application interface of Welzijn.AI, translated to English. Left and middle panels illustrate how Welzijn.AI was presented to the elderly (Section \ref{['elderly_eval']}), with real conversations about various EQ-5D-5L topics (mobility, self-care and mood). Right panel shows a work-in-progress dashboard illustrating how text-to-score module score outputs can be tracked over multiple conversations and topics, allowing signalling decline in well-being and quality of life.
• Figure 4: Value requirements of Welzijn.AI regarding its technical, environmental, and user-related aspects, ranked with the HDM. Requirements are listed as follows. Technical: (T1) periodically asking consent, (T2) available on-off button, (T3) robust, portable design, (T4) automatic shutdown in case of multiple persons present, (T5) long battery life, (T6) limited human-likeness, (T7) gradual flow of the conversation, (T8) safe data storage, (T9) speech recognition. Environment: (E1) available WiFi, (E2) help desk, (E3) power outlet, (E4) social media functionality, (E5) promotional campaign, (E6) agreements on access to data. User: (U1) education for caregivers and user, (U2) demo, test and practice session, (U3) creating awareness to reduce stigma on using assistive technology.