Ericsson is enhancing the autonomy of telecommunication networks by applying artificial intelligence (AI) across overall network operations. The plan is to automate major operational tasks such as fault response and quality optimization, and in the long term, realize an Autonomous Network (AN) that minimizes human intervention.
Thomas Kinnman, Head of AI and Automation Development at Ericsson, said in a recent interview with Electronic Times, “Ericsson's business is to support the software and products necessary for telecom operators to move toward L4,” adding, “Most of the solutions we currently provide are already in a state capable of supporting L4.”
L4 is a high-level automation stage among the AN autonomy levels presented by TM Forum. It is evaluated based on telecom operators' networks and operational processes, rather than individual equipment.
Ericsson points to possessing end-to-end technologies covering OSS/BSS fields in addition to network fields from radio access network (RAN) to core, IMS, and transport network, along with network operation know-how through managed services, as its competitiveness in the AN field.
The core foundation is the Ericsson Intelligent Automation Platform (EIAP). It expanded the scope of existing RAN-centric automation to the core and applied an open structure so that other equipment vendors and third-party developers can develop automation applications. Kinnman, head of AI and automation development, said, “We are opening the ecosystem so that various multi-vendors as well as our own company can develop their own apps,” adding, “Recently, we have expanded it to implement core automation applications beyond RAN.”
Agentic AI is also utilized for AN advancement. Rather than applying the same AI to all network tasks, automation technology is selected according to user requirements. Kinnman explained, “Depending on what use case you have in mind when utilizing agentic AI, the required technology varies,” adding, “If it is an automation domain that can be solved with scripts alone, it can be handled without incurring high costs.” For functions requiring microsecond-level processing like beam selection, and fault root cause analysis or service operations, since the required latency and AI model size differ, computational cost, security, and reliability are considered together.
Automation effects are also showing up in numbers. According to Ericsson, operating expenditure (OPEX) efficiency for cell optimization tasks improved by 75%, and downlink throughput of problematic cells rose by 43%. In a enterprise network slicing case of a European Tier-1 telecom operator, total work time input into design, creation, operation, and termination was reduced from 2,520 hours to 67.29 hours. Costs decreased from $202,000 to $4,810.
As the scope in which AI directly engages in network operations broadens, security and control have also become important. Ericsson restricts tasks and permissions that can be executed by each AI agent and monitors and audits activities separately. Kinnman said, “Rather than granting the same permissions to all agents, we define in advance so that specific agents can perform only specific tasks.” Based on a zero-trust architecture, external attacks and internal network anomalies are also checked.
