Integrating a Large Language Model Into a Socially Assistive Robot in a Hospital Geriatric Unit: Two-Wave Comparative Study on Performance, Engagement, and User Perceptions
Bibliographic record
Abstract
BACKGROUND: Addressing the complex medical and psychosocial needs of older adults is increasingly difficult in resource-limited care settings. In this context, socially assistive robots (SARs) provide support and practical functions such as orientation and information delivery. Integrating large language models (LLMs) into SAR dialogue systems offers opportunities to improve interaction fluency and adaptability. Yet, in real-world use, acceptability also depends on minimizing both technical and conversational errors, ensuring successful user interactions, and adapting to individual user characteristics. OBJECTIVE: This study aimed to evaluate the impact of integrating an LLM into a SAR dialogue system in a hospital geriatric unit by (1) comparing system performance and interaction success across 2 experimental waves, (2) examining the links between robot errors, interaction success, and multidimensional user engagement, and (3) exploring how user characteristics influence performance and perceptions of acceptability and usability. METHODS: Over an 8-month period, 28 older adults (≥60 years of age) attending a geriatric day care hospital (Paris, France) participated in a single-session evaluation of a SAR. Interactions took place in the day care hospital and were video-recorded across 2 waves: wave 1 (basic dialogue system) and wave 2 (LLM-based system). From the recordings, system performance (error types and interaction success) and user engagement (verbal, physical, and emotional dimensions) were coded. Acceptability and usability were measured using the Acceptability E-Scale and the System Usability Scale. Sociodemographic data were collected, and quantitative results were supplemented with a thematic analysis of qualitative observations. RESULTS: Following LLM integration, error-free interactions increased from 27.8% (10/36) to 70.2% (66/94; P<.001), comprehension failures decreased from 47.2% (17/36) to 17% (16/94; P<.001), and interaction success rose from 25% (9/36) to 74.5% (70/94; P<.001). Acceptability (Acceptability E-Scale: 12.8 vs 20.8; P=.003) and usability (System Usability Scale: 40.0 vs 60.4; P=.04) were significantly higher in wave 2. Engagement scores did not differ significantly between waves, though emotional engagement correlated positively with interaction success (r=0.28; P=.008), and age was negatively associated with both physical engagement (r=-0.30; P<.001) and acceptability (r=-0.20; P=.03). CONCLUSIONS: Behavioral engagement with a SAR in geriatric care is shaped by both system performance and individual user characteristics. Improvements in dialogue quality observed in wave 2, coinciding with the integration of the LLM, were associated with higher interaction success and enhanced user experience. Nevertheless, other contextual or group-related factors may also have contributed to this outcome. These findings highlight the importance of combining multimodal behavioral analysis with self-reported measures to inform the iterative, user-centered design of socially responsive robots in clinical contexts.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.009 | 0.016 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".