"Self-Learning and Evolving": ETRI Develops Autonomous Growth-Type Sleep Counseling AI

A technology for an "autonomous growth-type sleep counseling artificial intelligence (AI) agent" that evolves by independently acquiring knowledge without external intervention has been developed.

The Electronics and Telecommunications Research Institute (ETRI) announced on June 12 that it has developed an AI counseling system in which the AI autonomously acquires sleep information required by users and evolves its counseling approach based on each individual's sleep status and responses.

ETRI researchers are explaining the technology of an autonomous growth-type sleep counseling AI agent. Provided by Electronics and Telecommunications Research Institute (ETRI) Korea

This system operates by recognizing the user's sleep deprivation status, proposing customized solutions, and learning individual characteristics and preferences through ongoing conversations.

In particular, this system features a dual structure in which an "autonomous growth-type AI," responsible for inferring and making judgments about specific situations, and a "human understanding AI," which comprehends the user's state and intentions to respond appropriately, share roles and collaborate.

According to ETRI, the two AIs divide responsibilities and cooperate, enabling the system to continuously advance the level of counseling by adding new experiences (information) to its existing knowledge base.

Most notably, the system incorporates a "zero-shot learning" approach, which allows it to flexibly respond to unexpected questions, even when there is no pre-existing training data, by generating answers to new queries.

To validate the technology in real-world settings, ETRI researchers collected sleep state data from a large number of ordinary individuals and analyzed the counseling content of professional counselors to build high-quality AI training data.

The information generated during this process is accumulated in the form of a knowledge graph, and as the number of users increases and time passes, the precision and personalization of the counseling improve. This is described as a sleep AI that remembers knowledge and evolves autonomously, much like a human.

Additionally, the process of integrating various types of information?such as sleep patterns, activity levels, conversation history, and survey results?into new knowledge and continuously evolving the counseling content based on this is implemented using a Retrieval-Augmented Generation (RAG)-based large language model (LLM), which has recently attracted significant attention.

Song Hwajeon, head of the Complex Intelligence Research Laboratory at ETRI, stated, "This technology goes beyond simple sleep counseling and is expected to be widely utilized in various fields such as digital healthcare, intelligent home services, and elderly care. We hope it will be used as a true 'evolutionary artificial intelligence' technology, where AI interacts and accumulates and expands knowledge like a human."

This research was conducted as part of the "Research on Fundamental Technologies for Autonomous Growth-Type Complex Artificial Intelligence," a project promoted by the Ministry of Science and ICT.

This content was produced with the assistance of AI translation services.