ETRI develops AI system that configures and monitors fiber-optic networks using plain language

South Korean researchers have developed technology that lets AI configure telecommunications networks and check for errors without requiring operators to manually enter complex commands.

The Electronics and Telecommunications Research Institute (ETRI) announced Wednesday that it has developed "LLM-based optical access network operation technology," which allows a large language model (LLM) to interpret plain-language instructions from users, automatically configure optical communications networks and verify potential errors before applying any changes.

Optical access networks are critical infrastructure that delivers high-speed communications services to homes, businesses and mobile base stations — the last stretch of connectivity closest to end users. Often described as the capillaries of a telecommunications network, they involve a complex web of interconnected equipment and functions that has traditionally required trained technicians to memorize device-specific commands and operating procedures before making any configuration changes.

The technology ETRI developed automates that process using AI. When a user enters a network operation goal in natural language, the AI analyzes the intent of the instruction, plans the necessary steps on its own and then controls the actual optical access equipment.

Researchers applied the technology to a range of tasks required for real-world network operation, including device registration and configuration. Interoperability tests with optical access equipment confirmed that networks can be managed using natural-language commands alone.

ETRI also developed a mechanism to filter out erroneous commands before they can cause problems in a live network, even when the AI generates an incorrect instruction.

The system validates AI-generated control procedures in a virtual environment that reflects the current network state and operational data before applying them to the actual network, reducing the risk of outages caused by misconfiguration or faulty commands.

The institute also secured technology that allows AI to be used in network operations without exposing sensitive network information to outside parties.

Using a commercial external LLM for network management carries the risk of transmitting critical network configuration and operational data to outside servers. In response, ETRI is developing an "optical access-specialized local LLM" that runs entirely within the network itself.

ETRI plans to advance the technology beyond systems that act only on human commands, toward an "AI-native optical access network" in which AI autonomously monitors network conditions and makes its own judgments and control decisions.

To that end, the institute will also develop an "optical access foundation LLM" trained comprehensively on optical access network operational knowledge, network status data and accumulated operating experience.

"We have laid the technical groundwork for AI to understand natural-language commands from operators, carry out the necessary tasks, and verify results before actual deployment," said Jo Seung-hyeon, head of ETRI's Wired/Wireless Optical Access Research Laboratory. "Going forward, we will advance AI-native optical access technology in which AI autonomously recognizes, judges and controls network conditions."