As societies face rapid demographic aging, ensuring both the cognitive health and emotional well-being of older adults has become a critical concern. In response, this study proposes a dual-purpose conversational robot that functions not only as a platform for lifelong learning, but also as a personalized care companion for elderly users.

Built upon a multi-task large language model (LLM) framework, the robot seamlessly integrates four key functions:
1. Casual conversation to stimulate memory, social engagement, and cognitive activity;
2. Personal assistant tasks, such as reminders, schedule management, and daily information support;
3. Vital sign monitoring and anomaly detection, providing a watchful presence to ensure physical safety and timely response in emergencies;
4. Cognitive function assessment through natural daily dialogue, designed as an extended form of conventional screening tools such as HDS-R (Hasegawa's Dementia Scale–Revised) and MMSE (Mini-Mental State Examination). Unlike traditional structured interviews, this assessment passively and continuously infers cognitive status through everyday conversation, analyzing features such as fluency, recall, language structure, and response latency.

A distinctive feature of the system is its ability to adapt both conversation topics and interaction styles—such as response speed, linguistic complexity, and emotional tone—based on the cognitive assessment results. This personalization allows the robot to maintain appropriate cognitive stimulation while minimizing stress or disengagement for the user.

By combining these functions, the system fosters a friendly, human-compatible interaction style that encourages daily engagement while supporting autonomy and health monitoring. The educational aspect is embedded in everyday dialogue, enabling users to maintain verbal fluency, recall past experiences, and engage in knowledge-rich exchanges that subtly promote cognitive resilience.

At the same time, this platform serves as a learning environment for caregivers and healthcare students. In a pilot study conducted in a long-term care facility, interaction logs were analyzed to explore elderly users’ engagement patterns. Interestingly, it was found that residents’ satisfaction was more closely related to the number of dialogue turns—that is, the sustained back-and-forth flow of conversation—than to the specific content of what was said. This insight not only informs robot design but also serves as a valuable learning point for human caregivers: consistency and presence often matter more than information in communication with older adults.

This dual approach—empowering the elderly as learners while providing educators and caregivers with real-time contextual learning—illustrates a novel and practical fusion of AI-based education and elderly care. This work demonstrates how generative AI and multi-modal sensing can be synergistically applied to address both the educational and caregiving needs of aging societies, aligning with the goals of inclusive, lifelong, and technology-enabled learning.

Keywords: Learning, Wellbeing, Elderly Care, LLM, Conversation Robot.