Throughout our lives, we interact daily in conversations with our friends and family, covering a wide range of topics, known as open-domain dialogue. As we age, these interactions may diminish due to changes in social and personal relationships, leading to loneliness in older adults. Conversational companion robots can alleviate this issue by providing daily social support. Large language models (LLMs) offer flexibility for enabling open-domain dialogue in these robots. However, LLMs are typically trained and evaluated on textual data, while robots introduce additional complexity through multi-modal interactions, which has not been explored in prior studies. Moreover, it is crucial to involve older adults in the development of robots to ensure alignment with their needs and expectations. Correspondingly, using iterative participatory design approaches, this paper exposes the challenges of integrating LLMs into conversational robots, deriving from 34 Swedish-speaking older adults’ (one-to-one) interactions with a personalized companion robot, built on Furhat robot with GPT $$-$$ - 3.5. These challenges encompass disruptions in conversations, including frequent interruptions, slow, repetitive, superficial, incoherent, and disengaging responses, language barriers, hallucinations, and outdated information, leading to frustration, confusion, and worry among older adults. Drawing on insights from these challenges, we offer recommendations to enhance the integration of LLMs into conversational robots, encompassing both general suggestions and those tailored to companion robots for older adults.

Voice Agents (VAs) are touted as being able to help users in complex tasks such as cooking and interacting as a conversational partner to provide information and advice while the task is ongoing. Through conversation analysis of 7 cooking sessions with a commercial VA, we identify challenges caused by a lack of contextual awareness leading to irrelevant responses, misinterpretation of requests, and information overload. Informed by this, we evaluated 16 cooking sessions with a wizard-led context-aware VA. We observed more fluent interaction between humans and agents, including more complex requests, explicit grounding within utterances, and complex social responses. We discuss reasons for this, the potential for personalisation, and the division of labour in VA communication and proactivity. Then, we discuss the recent advances in generative models and the VAs interaction challenges. We propose limited context awareness in VAs as a step toward explainable, explorable conversational interfaces.

Background: Welfare technology has been launched as a concept to accelerate digital transformation in care services, but the deployment of these technologies is still hindered by organisational resistance, lack of infrastructure, and juridical and ethical issues. This paper investigates decision-making among municipal actors in the application and deployment of welfare technology from a procurement process perspective. The study explores the perceptions and negotiations involved in purchasing welfare technology at each stage of the procurement model, revealing the impact of technical, economic, juridical and ethical competence on the mapping, planning, procurement, implementation and management of welfare technology. Methods: The study presents empirical findings from qualitative interviews conducted among municipal actors in Sweden. Semi-structured interviews were gathered in 2020 among procurement managers, IT managers, and managers in social administration in three different municipalities (n = 8). Content analysis and systematic categorisation were applied resulting in the division of procurement practices into sub-categories, generic categories and main categories. Results: Challenges in the application and deployment of welfare technology occur at all stages of the procurement model. In mapping and planning, barriers are identified in the need analysis, requirement specification and market analysis. In the procurement stage, economic resources, standardisation and interoperability hinder the procurement process. Implementation and management are complicated by supplier assessment, legislation, cross-organisational collaboration and political strategy. Building on these findings, this study defines ‘procurement competence’ as consisting of technical, economic, juridical and ethical expertise in order to assess and evaluate welfare technology. Technical and ethical competence is needed in early stages of procurement, whereas juridical and economic competence relates to later stages of the model. Conclusions: Procurement competence is associated with the application and deployment of welfare technology in (1) assessment of the end-user’s needs, (2) estimation of the costs and benefits of welfare technology and (3) management of juridical and legislative issues in data management. Economic and juridical decisions to purchase welfare technology are not value-neutral, but rather associated with socially shared understandings of technological possibilities in care provision. Optimisation of procurement processes requires a combination of capabilities to introduce, apply and deploy welfare technology that meets the demands and needs of end-users.

Conversational agents (CA) are increasingly used to manage and coordinate household chores and everyday activities at home. However, these technologies should be adaptive to age-specific characteristics in order to be considered beneficial for the ageing population. This study presents a participatory design of a conversational agent to provide cognitive support in recipe following and nutrition advice for adults aged 65 and over. Through a qualitative thematic analysis, the study explores older adults’ expectations, interactions and experiences with the agent in order to identify age-specific challenges of interacting with CAs. Data consists of a participatory design workshop with eight older adults (aged 65 and over), followed by a Wizard of Oz study with ten older adults interacting with the agent in the kitchen environment in a laboratory setting. Results demonstrate that older adults consider conversational agents as beneficial for providing personalised recipe recommendations, advising the user to choose appropriate ingredients and reminding them of their dietary intake. When interacting with the agent older adults displayed challenges with confirmation and repetition, questioning and correcting, the lack of conversational responses, and difficulties in hearing and understanding the multi-modal interaction. Older adults experience agents as collaborators, but not as conversational partners. The study concludes that the accessibility and inclusiveness of conversational agents regarding voice interaction could be improved by further developing participatory methods with older adults.

This work aims to provide initial guidelines towards developing companion robots with large language models (LLMs) to be part of everyday lives of older adults. Using iterative participatory design (co-design) approaches, we analyze the challenges of applying LLMs for multi-modal open-domain dialogue, deriving from older adults' (one-to-one) interactions with a personalized companion robot, built on Furhat robot with GPT-3.5. An initial study with 6 Swedish-speaking older adults (65 and older) showed that the robot frequently interrupted the users, responded slowly and repetitively, engaged in superficial conversations, and caused a barrier in the interaction due to foreign language (English). Upon incremental technical developments to address these issues, participatory design workshops were conducted with 28 Swedish-speaking older adults. While the interactions (in Swedish) were smoother, less disrupted, and more varied in topics and responses, further challenges were observed due to hallucinations and obsolete information, and disengagement cues, causing frustration, confusion, and worry.