Promoting Cognitive Health in Elder Care with Large Language Model-Powered Socially Assistive Robots

Research Background and Problem Statement

• Identified Problems or Challenges: With the global aging population, there is an increasing demand for technologies that support cognitive health and enable early detection of cognitive decline. In elder care, traditional cognitive assessment tools (e.g., paper-based neuropsychological tests) are labor-intensive, prone to errors, and face challenges in user engagement. Additionally, due to concerns about technological privacy and cognitive impairments, the large-scale real-time deployment of socially assistive robots (SARs) remains at the research stage for older adults.
• Significance: As the population with cognitive impairments grows and caregiver shortages become more acute, the development of home-based technologies that support cognitive health and alleviate caregiver burden is urgently needed. These technologies can not only enhance the ability of older adults to live independently but also provide healthcare professionals with real-time cognitive assessment data.
• Research Motivation and Related Work: Previous studies have demonstrated the potential of socially assistive robots in elder care, mental health support, cognitive training, and companionship roles. However, there is a lack of long-term investigations into how such robots can effectively support cognitive health in home environments, especially studies focusing on robots with language interaction technologies and adaptive learning capabilities tailored to specific scenarios.

Proposed Solution

• Proposed Solution:

  • Develop a socially assistive robot powered by large language models (LLMs) to promote and assess cognitive health in residential environments for older adults.
  • Design tasks based on clinically validated tools, including picture description tasks and semantic fluency tasks, integrated with natural language interactions between the robot and users.
  • Conduct a five-week study to evaluate whether the robot can effectively enhance cognitive health while reducing reliance on human administrators.

• Innovations:

  • Technical Aspect: Integrating large language models (e.g., GPT-4) with socially assistive robots to improve the naturalness and flexibility of interactive dialogues.
  • Application Aspect: Conducting the first evaluation of the potential for robots to replace human administrators in cognitive health tasks and examining their long-term interaction behaviors with users.
  • Design and Privacy Protection: Employing AI-driven real-time monitoring mechanisms to ensure the safety and appropriateness of LLM-generated language, with an emphasis on user privacy protection.

• Implementation Steps and Key Technologies:

  1. Build a robot platform named Blossom, which is simple, affordable, and scalable through 3D printing.
  2. Integrate a cloud-based speech recognition and LLM response generation system to optimize voice interaction capabilities.
  3. Design adaptive interactions for the two tasks (e.g., providing prompts for semantic tasks) and dynamically adjust the robot's behavior to enhance engagement.
  4. Conduct multi-round experiments over five weeks, including three weeks of robot-managed tasks and comparative tests managed by human administrators at the beginning and end of the study.

Research Outcomes

• Specific Results:

  1. Task Performance Improvement: After three robot-managed sessions, participants showed significant improvement in the description task, including providing more details, using fewer repetitive words, and requiring fewer prompts. In the semantic task, participants reduced the use of repetitive words and reliance on prompts.
  2. Increased Social Engagement: During interactions with the robot, participants made more social comments compared to sessions managed by human administrators, indicating that the robot could foster stronger social engagement.
  3. User Acceptance: With repeated interactions, participants' acceptance and trust in the robot significantly increased, with the system usability score (SUS) rising from 80.7 to 85.6, reaching an "excellent" level.

• Advantages Compared to Existing Solutions:

  • The robot can sustain task management more consistently, with long-term interactions being more easily accepted by users and reducing dependence on human administrators.
  • LLM technology enhances the flexibility of language interactions, making task prompts and sentence generation more natural and adaptable to multilingual scenarios.

• Experimental or Evaluation Results:

  • The study demonstrated that LLM-powered SARs could improve participants' cognitive task performance, showing potential for maintaining user cognitive engagement and encouraging sustained use.
  • Autonomous robots, compared to human administrators, enhanced social interaction during tasks and created a more positive user experience.

• Limitations and Future Directions:

  1. Small Sample Size: The study included only 22 participants; future research should expand to more diverse populations across different regions and cultural backgrounds.
  2. Long-term Deployment: This study lasted only five weeks; longer-term interventions are recommended to evaluate the actual impact on cognitive health.
  3. Technical Optimization: Further improvements are needed in the real-time responsiveness of the voice interaction system and in reducing latency issues, along with clearer solutions for privacy protection.
  4. Broader Task Scope: More diverse tasks targeting other cognitive domains (e.g., memory, attention) need to be designed to sustain long-term user engagement.

Overall, this study provides the first empirical evidence supporting the potential of socially assistive robots integrated with LLM technology to promote cognitive health in elder care. It offers an important direction for developing low-cost, highly adaptable, and long-term deployable cognitive support technologies in the future.