Effectiveness of artificial intelligence robot interventions on psychological health in community-dwelling older adults: A systematic review
Article information
Abstract
Purpose
The global older adult population is rapidly growing, intensifying the burden of elderly care. To alleviate this challenge of an aging society, interventions utilizing artificial intelligence (AI) technology are becoming widespread. This review aimed to examine the effects of AI robot interventions on the psychological outcomes of community-dwelling older adults through a systematic literature review.
Methods
The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) method was employed to identify and select relevant studies. Seven electronic databases were thoroughly searched for eligible studies from June 1st to 30th, 2023. Methodological quality was assessed using RoB 2.0 or RoBANS 2.
Results
Thirteen studies (five randomized controlled trials and eight quasi-experimental studies) were selected in the systematic review. Among the selected studies, eight provided AI robot interventions individually, whereas five used a group format, primarily addressing older adults with cognitive impairment or dementia. Depression was the most frequently addressed psychological outcome, with six of ten studies reporting significant effects. Additionally, five studies each highlighted significant effects on emotions, such as positive expressiveness and enjoyment. However, quality of life, anxiety, and loneliness revealed divergent results.
Conclusion
AI robots show potential in alleviating psychological challenges for older adults. However, due to the scarcity of high-quality studies, the review recommends conducting more randomized controlled trials with rigorous designs. This review is expected to provide valuable insights for planning and executing AI robot interventions to improve psychological outcomes for community-dwelling older adults in future research.
INTRODUCTION
With the advancement of healthcare technology worldwide, life expectancy has increased. Additionally, the values surrounding marriage and childbirth among young adults have shifted, leading to low birth rates and rapid aging [1]. Currently, older adults constitute 18.2% of the total population in South Korea, classifying it as an aging society. By 2025, more than 20% of its population is expected to be older adults, transforming it into a super-aged society [2]. This rapid increase in the older adult population’s proportion presents a serious social issue, significantly increasing older adult care related national and individual burdens [1]. To mitigate these burdens and improve the quality of life of older adults, it is essential to support their independent daily living by utilizing their remaining abilities to the fullest extent; this can enable them to continue living in their communities (aging in place) rather than residing in care facilities [3].
The coronavirus disease (COVID-19) pandemic has highlighted the need for various non-face-to-face services, leading to the expansion of artificial intelligence (AI)-based non-face-to-face care services in the healthcare sector. Specifically, AI robots, which combine AI technology and robotics, have been developed to interact emotionally with humans and are utilized in various ways in older adult care services [4,5]. Such robots, often referred to as socially assistive robots, include Pepper (distributed to Japanese nursing homes), ELLIQ (distributed to older adults living alone in the United States), and Hyodol (distributed to older adult households in South Korea) [6,7]. These AI robots not only help manage medication and daily tasks for older adults but also recognize emotions—by analyzing facial expressions and voices—engage in simple conversations, and offer recreational activities such as games [6,7]. AI robots used in older adult care improve cognitive function, enhance their quality of life, and provide emotional support by reducing depression, anxiety, isolation, and loneliness [8-10].
Emotional health issues such as depression, anxiety, isolation, and loneliness are major causes of suicide in older adults and are closely related to their ability to perform daily activities and the presence of disabilities [11,12]. As people enter old age, social networks decline owing to the death of peers and retirement, increasing the risk of emotional problems, such as depression and loneliness [13,14]. However, to date, most AI robot intervention studies have focused on older adults living in care facilities, with few studies analyzing the effects of emotional support in community-dwelling older adults [15,16]. Recently, the Ministry of Health and Welfare of South Korea proposed expanding care services through AI robot development as part of a second comprehensive support plan for older adults living alone, emphasizing the need to verify the effectiveness of such interventions in community-dwelling older adults [13]. Therefore, this study aims to systematically review the literature on AI robot interventions conducted for community-dwelling older adults domestically and internationally, identify the elements and delivery methods of these interventions, and verify their effectiveness on emotional health (depression, positive emotions, quality of life, anxiety, and loneliness).
METHODS
Ethic statement: This study was exempted from approval by the Institutional Review Board as it is a review of the literature using previously published studies.
1. Study Design
This systematic review aimed to identify and comprehensively analyze AI robot interventions targeting community-dwelling older adults. The description of this study was written in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) reporting guidelines [17] (PROSPERO registration number: CRD42023437842).
2. Literature Selection and Exclusion Criteria
The core questions were selected according to the Participants, Intervention, Comparison, Outcomes, and Study Design (PICO-SD) framework to comprehensively examine AI robot interventions: the participants were community-dwelling older adults; interventions were AI robot-based interventions; outcomes focused on psychological factors; and study designs were limited to experimental research designs (randomized controlled trials or quasi-experimental studies). However, no specific criteria were set for comparison interventions. Based on previous research findings indicating an increase in AI-related literature in the healthcare field after 2010 [18], we targeted literature published after 2010. The exclusion criteria were as follows: 1) unpublished theses or literature presented only as abstracts, 2) case studies and protocol literature, 3) literature with inaccessible full texts, and 4) literature not in English or Korean.
3. Literature Search
Literature searches were conducted between June 1 and June 30, 2023. The international databases included PubMed, EMBASE, PsychINFO, and CINAHL, whereas the domestic databases included the Research Information Sharing Service (RISS), Korea Academic Information Service (KISS), and DBpia. Preliminary searches identified the key search terms as “older adults,” “artificial intelligence,” and “robot.” The search terms were adjusted for each database using a combination of indexed terms and the natural language used in Ovid-MEDLINE. For older adults, terms such as “elderly,” “older adults,” “senior,” “aged,” and “geriatric” were used. For AI robot interventions, terms such as “artificial intelligence,” “AI robot,” “socially assistive robots,” and “companion robots” were used. Emotional outcomes were searched for using terms such as “emotion*,” “psycho*,” “depression,” “anxiety,” and “mental health.” For domestic databases, “older adults” and “robot” were combined for searches. Additional manual searches were conducted based on references in the literature.
4. Literature Selection
EndNote 20 bibliographic management software was used for literature selection. Duplicate literature from each database was removed, and the remaining titles and abstracts were independently screened by three researchers according to the selection criteria. The full texts of the selected articles were reviewed in the second screening to finalize the literature selection. Reasons for exclusion were documented at each stage. Three researchers independently conducted the literature selection process, and any disagreements were resolved through discussion.
5. Quality Assessment of Literature
The quality assessment of selected literature was conducted using the version 2 of the Cochrane risk-of-bias assessment tool (RoB 2.0) for randomized controlled trials and the revised risk-of-bias assessment tool for non-randomized studies of Interventions (RoBANS 2) for quasi-experimental studies [19,20]. The RoB 2.0 tool includes five items: “bias arising from the randomization process,” “bias due to deviations from intended interventions,” “bias due to missing outcome data,” “bias in the measurement of the outcome,” and “bias in the selection of the reported result.” Each item was assessed as “low risk,” “some concerns,” or “high risk,” and the overall bias was evaluated. The RoBANS 2 tool comprises eight domains: “target group selection,” “confounders,” “measurement of intervention,” “blinding of outcome assessors,” “incomplete outcome data,” “selective outcome reporting,” “comparability of the target groups,” and “outcome assessment.” Each domain was rated as “low risk,” “high risk,” or “uncertain risk.” The quality assessment was independently conducted by three researchers, and any discrepancies were resolved through discussion.
6. Data Extraction and Analysis
Microsoft Excel was used to create a data extraction form to extract data from the final selected literature. General characteristics of the literature, such as authors, publication year, country, study design, participant age and number, and participant characteristics were extracted. To analyze the characteristics of the intervention programs used in the literature, data on the AI robot type, intervention duration, intervention content and methods, location of implementation, individual/group setting, provider, outcome variables, and main results were extracted. The data extraction process was independently conducted by three researchers using the same format, and any disagreements were resolved by discussion.
RESULTS
1. Literature Selection Results
According to the selection criteria, 13 articles were selected (Figure 1). In total, 19,222 articles were retrieved from the seven databases used in this study: 778 from PubMed, 3,278 from EMBASE, 4,297 from PsycINFO, 10,017 from CINHAL, 574 from RISS, 73 from KISS, and 205 from DBPia. An additional 11 articles were manually searched from the references of previous studies. After excluding 3,122 duplicates using EndNote bibliographic management software, 16,111 documents were reviewed according to the inclusion and exclusion criteria. Three researchers independently reviewed the titles and abstracts of the articles, eliminating 15,895 that did not meet the inclusion criteria, resulting in a total of 216 articles being selected. Subsequently, the full texts of the remaining 216 articles were reviewed, and ultimately, 13 articles were selected.
Flow diagram of study selection process.
AI=Artificial intelligence; PICO=Participants, Intervention, Comparison, Outcomes.
2. Quality Assessment Results
Figure 2 shows the quality assessment results of the 13 selected articles. For the five randomized-controlled trials, a quality assessment was conducted using RoB 2.0, and only one article was found to have a low risk of bias in the randomization process [21]. A low risk of bias due to deviations from intended interventions and missing outcome data were found in three [21-23] and four [21-24] articles, respectively. Bias in outcome measurement [23-25] and bias in selection of the reported results [21,22,24] were rated as low risk in three studies each. Overall, three studies had some concerns [21-23], while two studies had a high risk of bias [24,25]. Among the quasi-experimental studies, six out of eight were rated as having a high risk of bias due to the lack of control over confounders [26-31]. In the domains of outcome measurement and selective reporting of outcomes, six studies [27,29-33] and seven studies [26-30,32,33] respectively showed a low risk of bias. However, four studies each were rated as having a high risk of bias in terms of comparability of groups [26,29-31] and measurement of intervention [27,28,30,33].
Risk of bias of selected studies.
Kim (2020a)=Cited from reference 28; Kim (2020b)=Cited from reference 27.
3. General Characteristics of the Selected Studies
The general characteristics of the 13 selected articles are summarized in Table 1. Three studies (23.1%) were conducted from 2010 to 2015 [22,26,31], five (38.5%) from 2016 to 2020 [21,25,27-29], and five (38.5%) from 2021 [23,24,30,32,33], indicating a recent increase in published articles. The most common country was South Korea with four studies (30.8%) [24,27,28,32], followed by the United States with three studies (23.1%) [25,26,30], Japan with two studies (15.4%) [22,23], and Italy, France [29], Israel [33], and Spain [31], and New Zealand [21] each with one study (7.7%). Regarding study design, five studies (38.5%) were randomized controlled trials [21-25], and eight studies (61.5%) were quasi-experimental designs [26-33]. Six studies (46.2%) focused on older adults with cognitive impairment or dementia [21,24,28,29,31,32], four studies (30.8%) focused on older adults living alone or in isolation [22,25,27,30], one study (7.7%) focused on older adults living with caregivers [21], and two studies (15.4%) did not specify any particular condition [26,33]. The most common average age of participants was in the 70s (nine studies, 69.2%) [22-24,26-29,31,33], with one study in the 60s [25] and one in the 80s [32], while two studies (15.4%) did not provide this information [21,30].
General Characteristics and Methodology of 13 Studies (N=13)
4. Characteristics of AI Robot Interventions
The characteristics of the AI robot interventions performed in the 13 selected articles are summarized in Table 2. The most common type of AI robot was humanoid, which was used in six studies (46.2%) [22,24,26-29], followed by animal-shaped robots in two studies (15.4%) [21,30]. All studies using humanoid and animal-shaped robots reported significant emotional effects, whereas only two out of four studies using other types of robots reported significant effects [32,33].
Characteristics of Artificial Intelligence (AI) Robot Interventions of Included Studies (N=13)
The intervention methods applied were individual interventions in eight studies [22,25-28,30,32,33] and group interventions in five studies [21,23,24,29,31]. Notably, in all studies providing individual interventions, there were no intervention facilitators, and only two out of the eight studies included older adults with cognitive impairment or dementia [28,32]. In contrast, group interventions were facilitated by an intervention facilitator in all but one study [23]. Among the group interventions, four studies targeting older adults with cognitive impairment or dementia showed significant effects on depression [21,24,29] or anxiety [31], while the study targeting healthy older adults [23] did not show significant effects.
When looking specifically at the intervention facilitators, the most common approach, seen in nine studies (69.2%), did not involve human facilitators [22,23,25-28,30,32,33]. In these cases, the intervention programs were conducted either by AI robots [23,25-28,32,33] or with AI robots treated like pets, requiring no facilitator [22,30]. Other approaches included interventions provided by trained professionals or medical staff [24,29,31], and in some cases, the researchers themselves provided the intervention [21,22,27,28,30,32,33].
The settings for the interventions were primarily home [21,22,25,27,28,30,32,33] or senior care facilities [21,24,29,31,33], with some studies not specifying the location [23,26]. Home-based interventions most frequently involved integrated management interventions in four studies [25,27,28,32], living with and communicating with AI robots in three studies [21,22,30], and exercise and cognitive therapy in one study [33]. Senior care facilities were divided into dementia centers [24,29], day care centers [21,31], and senior welfare centers [24,33]. Cognitive therapy was the most common intervention provided in these facilities [24,29,33], followed by combined exercise and cognitive therapy in one study [31] and living with robots in one study [21].
Regarding the total duration of interventions, the most common period was 1~2 months, found in six studies [21,22,24,25,29,30], with one study applying a single, one-time intervention [33].
The intervention content most frequently involved integrated management interventions in four studies [25,27,28,32], all conducted at home. In these integrated management interventions, AI robots performed various roles, including communication, medication guidance, scheduling, contacting family and friends, playing music and games, movement detection, and alerting to abnormalities. In contrast, three studies focused solely on simple communication with AI robots [21,22,30], three studies on memory or cognitive interventions [23,24,29], one study on exercise intervention [26], and two studies on both cognitive and exercise interventions [31,33].
5. Psychological Effects of AI Robot Interventions
The psychological effects of the AI robot interventions are summarized in Table 3. The most common outcomes measured were depression and positive emotions, followed by quality of life, anxiety, and loneliness.
Results of Research-Related to AI Robot Program of 13 Studies (N=13)
1) Depression
Of the 10 studies that reported depression as an outcome [21-25,27-29,31,32], six [21,24,27-29,32] reported significant reductions in depression in the intervention group. Of these, five studies [21,24,28,29,32] targeted older adults with cognitive impairment or dementia, and three studies [21,24,29] involved group interventions. Four of the six studies reported significant effects using humanoid robots (NAO, Sil-Bot, and Hyodol) [24,27-29], whereas the remaining two used animal-shaped [21] or mechanical [32] robots.
2) Positive Emotions
Positive emotions were measured in six studies [21,22,25,26,29,33] and were variously reported as pleasure [22], expressiveness [21,29], enjoyment [26,33], and happiness [25]. The measurement methods also varied, with studies using custom-developed items for yes/no [22], 5-point [33], and 10-point [26] scales for pleasure and enjoyment, whereas happiness was measured using the Subjective Happiness Scale, a validated and reliable instrument. Two studies measured expressiveness by observing participants’ smiling faces [21,29]. Four studies [21,26,29,33] reported the significant positive effects of interventions conducted once or twice weekly, or as a single session. By contrast, two studies [22,25] found no significant effects of daily interaction with AI robots at home.
3) Quality of Life
Quality of life was reported as an outcome in four studies [23,27,28,30]. Two studies [27,30] reported significant effects of interactions with AI robots at home, targeting lonely or isolated older adults. In contrast, two other studies [27,28] found no significant effects in older adults with cognitive impairment or diabetes. Another study by Otake-Matsuura et al. [23] found no significant effects in healthy older adults participating in group conversation programs led by AI robots.
4) Anxiety
Anxiety was reported as an outcome in three studies [28,29,31], all of which targeted older adults with cognitive impairment or dementia. Two studies [29,31] involving interventions conducted by professional healthcare providers or therapists found no significant effects on anxiety. By contrast, one study [28] on interventions led by AI robots without human facilitators reported a significant reduction in anxiety.
5) Loneliness
Two studies [25,30] reported loneliness as an outcome. Tkatch et al. [30] found significant effects in reducing loneliness among isolated older adults using pet-shaped AI robots. However, Sidner et al. [25] found no significant effects on loneliness in older adults living alone when using Reeti, an AI robot in neither animal nor human.
DISCUSSION
This study conducted a systematic review to understand the characteristics and psychological health effects of AI robot interventions on community-dwelling older adults. Most of the intervention participants were older adults with cognitive impairments or those living alone, and humanoid and animal-shaped robots were most commonly used. AI robot interventions effectively reduced depression and improved positive emotions in community-dwelling older adults; however, their effects on quality of life, anxiety, and loneliness were inconsistent.
Regarding the form of the robots, 45.4% of the interventions used AI humanoid robots. This contrasts with previous studies on older adults living in care facilities, in which animal-shaped AI robots were more commonly used [15]. Previous studies have suggested that AI humanoid robots are perceived as companions by community-dwelling older adults and positively contribute to psychological health [34]. Some concerns have been raised about the “Uncanny Valley Effect,” in which robots that closely resemble humans can evoke discomfort [35]. However, this phenomenon is reported only among younger and middle-aged adults, not among older adults [36,37]. AI humanoid robots, with their human-like appearance, physical movements, and vocal responses, can provide stronger emotional support to older adults than other robot designs [29,37,38]. In countries such as Europe and Japan, AI humanoid robots are gaining attention not only for psychological health but also for meeting future older adult care needs [29,39]. However, in South Korea, studies related to AI humanoid robots, such as Sil-Bot and Hyodol, are limited [24,27]. Considering the increasing older adult population and low birth rates, further research on AI–humanoid robot interventions in South Korea is required.
This study found that AI robot interventions were effective in reducing depression among community-dwelling older adults with mild-to-moderate cognitive impairment. This aligns with previous systematic reviews where most participants were residents of care facilities [15,40]. In South Korea, 31.2% of community-dwelling older adults have cognitive impairment and its severity is associated with higher levels of depression [41]. Managing depression in community-dwelling older adults with cognitive impairment is crucial for their continued community residence [42]. AI robot interventions, which include communication with robots and cognitive therapies, such as games and music, provide positive experiences, thereby reducing depression. A previous study has also suggested that structured programs are more effective in improving psychological health than merely living with an AI robot [15]. However, owing to the small sample size and heterogeneity in intervention content, duration, and frequency among the included studies, further large-scale studies with systematic designs and rigorous controls are needed to provide stronger evidence.
AI robot interventions have also helped improve positive emotions in older adults. The included studies measured positive emotions using various concepts and tools such as pleasure, expressiveness, enjoyment, and happiness. This variation likely reflects the different cognitive abilities of the older adult participants. Given the extensive use of AI technology in older adult care [37,43], exploring unified concepts and appropriate measurement methods for the positive emotions experienced through AI robot interventions is necessary. Studies that reported no significant effects on positive emotions all had in common that the participants were isolated older adults, and the interventions were provided daily. Socially isolated older adults living alone may have a lower technology acceptance [44], which could explain the lack of a significant effect. The novelty effect, in which the initial interest in an AI robot diminishes over time, may also play a role [45]. On the other hand, interacting with an AI robot once or twice a week may have been able to generate older adults’ continued curiosity and interest through novel activities, which may have led to more positive emotions. Therefore, determining the optimal frequency and duration of AI–robot interactions is important to enhance technology acceptance and active engagement.
This study focused on community-dwelling older adults who underwent AI robot interventions. Home-based interventions included comprehensive management, such as medication guidance and communication with friends and family, and games to support cognitive function, whereas facility-based interventions only focused on improving cognitive or physical function. AI robots used by community-dwelling older adults should include functions that enhance daily convenience and should be easy to use [46], differentiating them from those used in care facilities. Since 2017, the conversational AI robot Hyodol has been distributed to older adult households in South Korea, allowing caregivers to monitor medication, meals, and movements through an application [27]. AI robot interventions are becoming an effective means of supporting independent living and continuous residence in the community for older adults. Group interventions were found to be more effective than individual interventions in promoting social interaction, regardless of living environment [15]. Given the increasing proportion of older adults living alone, AI robot group interventions could be a promising approach to enhance social interactions and reduce negative emotions, such as depression and loneliness [47].
This study had several limitations. First, many selected studies reported low-quality assessments. Future studies should adhere to the Consolidated Standards of Reporting Trials-Artificial Intelligence (CONSORT-AI) [48] for more systematic and rigorous AI robot-intervention applications. Second, owing to the heterogeneity in participants, AI robot types, intervention durations and methods, and outcome measures, a meta-analysis could not be performed, limiting the provision of quantitative evidence for AI robot interventions. Finally, studies published in languages other than English and Korean were excluded. Despite these limitations, this study provides valuable evidence for effective AI robot interventions targeting community-dwelling older adults based on participant characteristics.
CONCLUSION AND RECOMMENDATIONS
This study confirmed the positive effects of AI robot interventions on the psychological health of community-dwelling older adults. Some have argued that AI technology in older adult care undermines the essence of caregiving by focusing solely on functional aspects [49]. However, this systematic review revealed that AI robot interventions can effectively reduce depression and enhance positive emotions in community-dwelling older adults, with significant effects on anxiety, quality of life, and loneliness. Thus, AI robot interventions are emerging as effective tools for older adult care in the era of the Fourth Industrial Revolution, forming an integral part of gerontechnology. Based on the findings of this systematic review, the following recommendations are proposed for nursing research and practice in community-dwelling older adults: (a) Prioritize humanoid robots for AI interventions considering their human-like appearance and ability to promote social interaction and technology acceptance among older adults; (b) Implement group rather than individual interventions to enhance social interactions and positive emotions; (c) Develop and use unified concepts and appropriate measurement tools for positive emotions in future research to ensure consistency and comparability; and (d) Conduct large-scale studies with systematic designs and rigorous controls to provide robust evidence for AI robot interventions.
Notes
Authors' contribution
Study conception and design acquisition - YP; Data collection - YP, SJC, HJK, and HNJ; Interpretation of the data - YP, SJC, HJK, and HNJ; Drafting and critical revision of the manuscript; YP, SJC, HJK, and HNJ; Final revision: YP and HNJ
Conflict of interest
No existing or potential conflict of interest relevant to this article was reported.
Funding
This work was supported by the 2023 Graduate Student Research Grant from the Research Institute of Nursing Science, Seoul National University.
Data availability
Please contact the corresponding author for data availability.
Acknowledgements
None.