Dementia is a growing concern worldwide as the population ages. By 2050, the aging demographic is projected to double, with the number of people living with dementia anticipated to reach 139 million [1]. This surge is particularly pronounced in low- and middle-income countries, where approximately 63% of people living with dementia live [2]. The challenge of dementia in India stands out, given its vast population and rising incidence of dementia, coupled with significant gaps in resources and clinician accessibility. Today in India, 8.8 million people aged 60 years and older live with dementia, a figure that is projected to exceed 14 million by 2050 [3]. Yet, the diagnosis, care, and treatment landscape in India reveals a staggering 90% gap [4]. Addressing the challenge of dementia in India requires an understanding of the sociocultural factors that limit access to care and a widespread lack of dementia awareness among all stakeholders. Studies in India indicate that people living with dementia can be stigmatized and neglected by society and their families, creating critical barriers to accessing care [5]. Furthermore, the financial and mental strain placed on caregivers and family members is enormous [6]. Research in South India underscores the demand for stronger technology adoption to support people living with dementia and caregivers [7].

Socially assistive robots (SARs) and conversational artificial intelligence (AI) technologies hold potential to enhance independence and well-being at home [8-10]. Examples of home assistance include medication management and reminders [11], cognitive stimulation and engagement [12], companionship [13], and activities of daily living [14], while relieving the burden placed on caregivers [15]. Furthermore, SARs have shown potential in improving neuropsychiatric symptoms [16] and reducing agitation [17].

Voice interaction provides an accessible and natural user interface to interact with technology without the need for complex controls, promoting usability among older adults [18] and providing safety in scenarios where voice may be the only form of communication. For instance, conversational AI integrated with home monitoring technology can track behavior and alert stakeholders to risks, such as falls or cognitive decline [19]. Conversational SARs offer multimodal interactive features beyond speech, including facial expressions, embodiment, and gestures. Previous work has argued that physical embodiment and social presence can enhance engagement and enjoyment in human-robot interaction (HRI) [20-22]. Tapus et al [23] defined engagement as a sustained collaborative connection between humans and robots. Heerink et al [24] reported social presence as a key robot design feature for older people, while Nishio et al [22] showed that older adults were more verbally engaged with an embodied robot than a virtual one. Despite the promise of conversational robots, the effectiveness in supporting well-being and daily living activities among people living with dementia remains underexplored.

User-centered design with stakeholders is critical for developing useful and acceptable robots [25]. This approach can help identify perceived benefits and barriers for specific support contexts during the early stages of HRI design [26]. Previous research suggests that cultural factors influence robot acceptance [27,28], making it crucial to consider these in robot design. However, few studies in robotics for dementia care have involved stakeholders [29-33], and none of these have addressed the necessary cultural adaptations and feasibility challenges inhibiting real-world deployments in low- and middle-income countries, particularly in India.

While there is a lack of studies exploring SARs for dementia care in India, their acceptance in supporting daily care and well-being is growing among Indian older adults [34,35]. To the best of our knowledge, this is the first study to investigate the cultural feasibility of conversational SARs for dementia care in India through stakeholder involvement. In the context of this study, cultural feasibility refers to the suitability and acceptance of technology to support dementia care within India’s cultural context, including stakeholders’ perceptions and the barriers to its integration into daily life. This is particularly pertinent in India, given sociocultural factors that shape user perceptions of robots for well-being support in residential settings, such as deep-rooted family caregiving responsibilities and varying levels of digital literacy across rural and urban populations [36].

In this study, we assessed the cultural feasibility of conversational robots in India. We engaged a total of 29 stakeholders, including people living with dementia, caregivers, family members, and dementia care professionals. People living with dementia interacted with 3 conversational robots with varying interactive modalities. We evaluated engagement through behavioral observations and video analysis, assessed robot acceptance using standardized questionnaires, and gathered stakeholders’ attitudes and perceptions through interviews. Our research was guided by the following research questions (RQs):

We engaged a total of 29 stakeholders in India, including people living with dementia (n=11), formal and informal caregivers (n=11), and dementia care professionals (n=7). People living with dementia, often accompanied by a caregiver or both a caregiver and a family member, participated in a single session that involved verbal dialogue with 3 conversational robots. Each session lasted approximately 45 minutes and included a semistructured interview. We encouraged caregivers and dementia care professionals who observed the sessions to share their perceived benefits and barriers to using technology for dementia care in India.

After providing informed consent, learning about the purpose of the study, and having the opportunity to ask questions, participants verbally engaged with the robots in English. Robots were placed on a table in front of participants (Figure 1), one at a time. We aimed to understand how people perceived robots with different interactive modalities: (1) voice only, (2) facial expressions with voice, and (3) embodiment combined with voice and facial display. Accordingly, robots were presented to participants in the same order to minimize distractions. A psychologist and a researcher jointly assessed engagement and moderated the session, respectively, while a video camera recorded the session. Data were collected via direct observation of engagement, video recordings for offline assessment of engagement, a robot acceptance questionnaire, and a semistructured interview to explore the perceived benefits and barriers to technology adoption in India.

We involved 11 people living with mild to moderate dementia, as determined by established clinical assessments (mean age 78.6, SD 8.95; 2 female participants). Two weeks before the study, participants’ cognitive and daily function were evaluated using the Alzheimer’s Disease Assessment Scale-Cognitive Subscale (ADAS-Cog) [37] and the Clinical Dementia Rating (CDR) [38] (mean ADAS-Cog 21.4, SD 7.1; mean CDR 1.8, SD 0.5). Participants were recruited based on the severity of cognitive impairment and communication ability. Specifically, we included participants with a CDR score of ≤2 (mild to moderate cognitive impairment) and sufficient ability to engage verbally in English with the robots. The study excluded individuals who were unable to provide informed consent or had impairments that precluded them from hearing or seeing the robot.

The study was conducted at the Schizophrenia Research Foundation (SCARF) Centre for Active Ageing and the in-patient unit of the Sri T S Santhanam Centre for Elderly Care in Chennai, India. All participants lived in urban settings and had an educational background (mean years of education 13.2, SD 4.0). Most participants (n=7) lived at home with their partner, 2 had formal caregivers, and 2 were inpatients. Participant information is detailed in Table S1 in Multimedia Appendix 1. Generally, participants were not familiar with voice technologies. Of the 3 participants who had heard of commercial devices, including Amazon Alexa, none had used them.

Participants engaged with 3 conversational social robots (Rs): a smart speaker (R1), a virtual affective robot (R2), and an embodied social robot (R3). R1, an Amazon Echo device, prompted users with a well-being questionnaire querying mood, sleep, fatigue, anxiety, activity plans for the day, and a breakfast recall to trigger memory of a recent event. The well-being questionnaire was developed following our previous work that analyzed daily Alexa interactions in households with people living with dementia [19]. R1 interacted through speech and screen displays. Participants initiated the questionnaire using a specific invocation phrase, as instructed by the researcher.

R2, a research platform built on our previous work [39], is a virtual affective robot displayed on a tablet that simulates facial expressions. The robot’s face was integrated with a bespoke conversational agent designed to assist in daily activities, offering support based on the user’s blood pressure readings. R2 interacted through speech combined with a virtual affective face.

R3, the social robot Miko, displayed facial expressions through animated eyes and mouth [40] and included physical embodiment. While R3 could move autonomously, its motion was restricted for this study to avoid distracting behaviors. Participants initiated dialogue with R3 using an invocation name, as instructed by the researcher.

Interactions with R1 and R2 followed a structured, health-related questionnaire. R1 interacted through voice and screen display, while R2 combined voice with a virtual affective face, simulating basic expressions. In both cases, the robots adapted their verbal responses based on user input (eg, the robot asked for details if the user expressed concern). In contrast, interactions with R3 were more open-ended and social: participants asked general questions about the robot (eg, “Hello Miko, where are you from?”) and culturally specific questions (eg, “Hello Miko, what is famous in Chennai?”) from a provided list and were encouraged to ask additional questions of personal interest. Interactions with R3 had a higher degree of multimodality, combining voice with the robot’s physical embodiment and animated facial expressions. Average dialogues with R1, R2, and R3 lasted 4 minutes 53 seconds, 4 minutes 3 seconds, and 4 minutes 52 seconds, respectively. Examples of interaction transcripts are provided in Multimedia Appendix 1.

The study was approved by the Ethics Committee of SCARF, India (Protocol SRF-DC/17-A1/JUN/2021). Written informed consent was obtained from all participants before participation. Participants were informed that sessions would be video recorded for research purposes, that only deidentified information would be analyzed, and that data were securely stored on local institutional servers and were accessible only to members of the research team. Participants were compensated with INR 1000 (US $11) for their time in the session. They retained the right to withdraw from the study at any time.

We first present findings from engagement with the robots, assessed both from on-site behavior observation and video analysis. We further report results from robot acceptance questionnaires and insights from semistructured interviews.

People living with dementia (n=11) and caregivers (n=11) completed a posttrial questionnaire to assess their robot acceptance, overall experience, and attitudes toward adopting conversational robots in India. Caregivers responded based on their observations and how they envisioned the technology in their care setting.

Table 2 presents descriptive statistics for each robot condition and the reliability coefficient (Cronbach α) of the underlying constructs. Although no statistically significant differences were reported between the means of each construct, some trends emerged. In general, participants expressed positive attitudes toward the conversational robots. Perceived sociability and usefulness had the highest mean values, indicating that participants viewed the robots as useful tools to engage in verbal dialogue and support well-being and independence at home. Particularly, 80% of respondents agreed that Alexa (R1) could improve social interaction skills, 78% agreed with the virtual affective robot (R2), and 65% with the embodied robot (R3; Figure 3A).

Lower mean values were generally found in the perceived ease of use (Table 2), as participants noted they were not ready to use such systems independently due to unfamiliarity with voice technologies. Approximately 50% of respondents felt they could not interact independently with R1 or R2, and 40% reported the same for R3 (Figure 3B). The anxiety subdimension (reverse-scored in Table 2) showed low mean values across all robots, indicating that people living with dementia and caregivers did not find the robots intimidating. Furthermore, the majority expressed trust in following the robots’ suggestions or action prompts, with 65% of respondents agreeing for R3, 61% for R2, and 50% for R1 (Figure 3C).

We explored stakeholders’ perceived benefits and barriers of using SARs for dementia care in India. Views were collected from the full stakeholder group, including people living with dementia, formal and informal caregivers, and dementia care professionals, providing expert-informed insights into how this technology could work in the person’s daily routine. Three main themes emerged from the interviews: (1) perceived valuable support scenarios to promote dementia care in home and clinical settings in India, (2) preferences for robot interactive modalities, and (3) factors affecting feasibility in India.

Health inequality significantly increases the impact of dementia in underprivileged and lower socioeconomic communities [47,48]. The challenge of dementia in India stands out due to its vast population, high incidence rates, limited resources, poor clinician accessibility, and a striking 90% gap in diagnosis and care [4]. To address this gap, we involved 29 stakeholders, including people with lived experience of dementia, to investigate the cultural feasibility of conversational SARs to support dementia care in India. People living with dementia interacted with 3 robot platforms with different interactive modalities: a smart speaker (R1), a virtual affective robot (R2), and an embodied social robot (R3). Our study provides insight into (1) the engagement of people living with dementia with the conversational robots, (2) robot acceptance, (3) the perceived utility in dementia care, and (4) the challenges that need to be addressed for integrating this technology into real-world care settings in India.

Our findings indicate that people living with dementia positively engaged with the 3 robots in verbal dialogue, with variations in some engagement metrics (refer to RQ1). Participants exhibited shorter visual engagement durations and lower attention scores with R1, the smart speaker. Conversely, the perceived trust in this robot as an assistive companion was generally lower compared to the other platforms. However, it registered longer positive emotional engagement, which could be attributed to the novelty effect [49], that is, initial engagement that wears off over time, resulting in decreased interest and changing attitudes toward robots. Sessions were designed with an increasing level of multimodality in mind, allowing participants to experience the incremental interactive features each robot offered. For this reason, R1 was the first robot that participants interacted with, and given their overall unfamiliarity with voice technology, this suggests R1 sparked initial curiosity. However, the smart speaker R1 may lack interactive features to sustain attention over time compared to the other 2 robot platforms.

In addition, we found that overall R3 was more challenging to interact with, a finding supported by the robot’s lower ratings for its perceived role as a conversational partner. This difficulty may be attributed to R3’s requirement to be called by its invocation name for each query, which many participants found challenging to recall. Consequently, participants would often ask R3 a question without activating its listening mode. Despite these challenges in triggering the robot, R3 elicited high verbal engagement overall, with some participants independently asking questions beyond those on the list provided. This could be due to its multimodal interactive features, including embodiment. Research has shown that the physical embodiment and social presence of SARs can promote engagement compared with virtual agents [20,22,24,50]. Despite the overall engagement with R3, some participants perceived R3 as a “toy” and voiced concerns about feeling infantilized by technology.

Stakeholders were open to the adoption of conversational technology to support dementia care in India and identified key support functions (refer to RQ2). Supporting routines at home was viewed as a promising way for conversational AI to enhance the independence of people living with dementia while reducing the burden placed on caregivers. Emphasis was placed on reminders for daily tasks and alerts for behavioral risks, such as agitation or wandering. In addition, the potential to have regular conversations about topics of interest and prompt users to act in their daily routines was deemed beneficial to promote cognitive function and provide companionship.

Overall, our findings underscore the need to personalize verbal interactions to users’ cognitive needs over time, culturally adapt the robot’s communication style to sustain user comfort in task-based or more conversational tasks, and prioritize effectiveness in natural language communication. Building on these insights, the following section outlines design implications and recommendations for integrating conversational AI into dementia care settings in India.

Our study identified improvements needed to culturally adapt conversational robots for dementia care in India (refer to RQ3). Our findings suggest improvements in speech processing in noisy environments and the ability to handle user pauses and longer speech. Response delays and turn-taking during verbal dialogue with robots are well-recognized challenges that remain unresolved [51]. Local accent interpretation should also be addressed. In our experiments, speech recognition errors often led to participant disengagement. The limited performance of automated speech recognition systems on Indian accents remains a recognized gap, largely due to the lack of representative training data [52]. However, recent research has shown promising improvements in both speech interpretation and generation of Indian-accented English. For instance, fine-tuning Whisper on Indian-accented speech has improved recognition accuracy, reducing the word error rate to 15% [53]. In addition, some commercial text-to-speech services provide Indian English voices, such as Amazon Polly, Microsoft Azure Neural TTS, and ElevenLabs. Future HRI feasibility studies in India could integrate these adapted speech recognition and synthesis models in conversational robotic systems to enhance user engagement.

Adapting the robot’s interaction style and appearance to align with cultural norms is another important aspect. Particularly, participants valued a communication style characterized by kindness and appreciation, which should be reflected in the robot’s tone, choice of words, and facial expressions if present. Participant feedback indicated that cultural customization of the robot’s appearance is a desired feature to make it more relatable, including the option to change eye colors, add a bindi, and modify clothing. Furthermore, introducing robot prototypes in local clinics as a preliminary step could promote familiarity and acceptance.

To address privacy concerns related to voice data collection, it is important to educate users, especially caregivers, about the robot’s support functions and limitations, what data is collected, and who has access. This requires transparent informed consent alongside compliance with established privacy and security standards, for example, the General Data Protection Regulation and Health Insurance Portability and Accountability Act. Technical safeguards, such as local storage or edge processing of sensitive data, and feature anonymization in voice analysis, can further reduce privacy risks inherent to speech processing.

Participants were informed of their right to withdraw from the study at any point. Future recruitment should further strengthen data governance by incorporating a consent clause that enables data withdrawal from protected institutional servers. In addition, future robot deployments should provide accessible training resources that help users understand the system’s capabilities and purpose, ultimately fostering long-term trust and acceptance. Such predeployment training should be tailored to differences in infrastructure and digital literacy between urban and rural Indian settings, which directly affect the feasibility of conversational AI implementation. Future deployments in India should consider the tradeoff between using local (on-device) models for speech recognition and synthesis and cloud-based architectures, such as those tested in this study. While the former reduces reliance on stable Wi-Fi connectivity, it operates with a more limited vocabulary and reduced contextual flexibility for verbal engagement. Adopting a hybrid design that combines local processing for essential interactions with cloud-based retrieval to sustain richer dialogue may provide a more robust and equitable pathway for conversational AI deployment across India’s diverse settings.

Engaging people affected by dementia in research presents inherent challenges, and the requirement for fluency in English further limited the size and diversity of the participant cohort, as in India, more men from the current older generation acquired English as a second language. However, this study offered practical design considerations obtained from stakeholder feedback to enhance robot acceptance and the feasibility of implementation in India. Our study was designed to capture initial impressions and cultural perspectives from HRIs. Future longitudinal studies could address continuous interactions in home settings and the associated long-term implications.

Furthermore, future within-subject studies that systematically counterbalance the order of robot presentation could provide deeper insights into how specific robot features influence user acceptance. A future iteration of the work should implement the culturally tailored design features emerging from stakeholder feedback on robot prototypes for validation in real-world residential and clinical settings. In addition, as large language models continually change the landscape of open dialogue, integration with conversational robots represents a promising direction for engaging support among older adults [54]. Yet, potential risks, including the spread of misinformation or biases, should be addressed in the cultural feasibility exploration with stakeholder involvement.

Given the distinct impact of dementia on individuals, our findings cannot be generalized. However, we propose design considerations obtained from stakeholder perspectives on necessary improvements for sustained engagement when introducing robots into new cultural contexts. This includes incorporating features for local accent interpretation, customizing facial expressions, and implementing strategies to promote familiarity and accessibility.

We pioneered a cultural feasibility study of conversational robots in India, an underrepresented context in AI and robotics for dementia care. We involved 29 stakeholders to evaluate human-robot engagement, robot acceptance, and perceptions regarding adopting the technology in India. Participants accepted and were willing to engage with conversational robots in verbal dialogue and recognized their potential to support daily routines, alleviate loneliness through personalized conversations, and promote cognitive engagement. Cultural adaptations to robot design and implementation are essential for feasibility in India, including enhanced speech processing for local accent interpretation and effective communication in noisy settings; a kind and appreciative communication style; customizable appearance through tone, choice of words, and facial expressions; and a phased introduction to promote familiarity with technology, starting in local clinics. This work paves the way for a more inclusive and culturally aware design of conversational robots for dementia care, particularly for feasible, real-world deployments and sustained user engagement in residential settings in India.