Mobile Giants Launch 'Open Telco AI'
The GSMA, the global mobile operators' association, has launched a new 'Open Telco AI' initiative. The goal is to create interoperable, industry-standard AI models to boost network operations, customer service, and security across the telecom sector.
The initiative addresses a key industry gap: while general-purpose AI models are advancing, they often fall short on telecom-specific tasks. A recent GSMA Intelligence report revealed that only 16% of generative AI deployments in the telecom sector have been applied to network operations. AT&T and AMD have been announced as founding partners. AT&T is contributing a family of open telco-grade AI models, while AMD is providing the necessary computing power for training and fine-tuning these models through its GPU platforms and cloud partner TensorWave. This effort is not starting from scratch; it builds on the collaborative spirit of the GSMA's earlier 'Open Gateway' initiative, which focused on creating a standardized framework of APIs for universal access to mobile network capabilities. The Open Telco AI project will have a portal to provide access to open models, tools, and datasets. A broad coalition of industry players are supporting this initiative as contributing partners, including Huawei, Nvidia, Orange, SK Telecom, and Vodafone. This wide-ranging collaboration also includes academic institutions like King Abdullah University of Science and Technology and Purdue University, which are contributing data, models, and use cases. The performance of the AI models will be measured and tracked through a "Telco Capability Index." This index will evaluate model performance on a growing set of tasks that are specific to the telecommunications industry, ensuring the developed AI is fit for purpose. The ultimate goal is to move beyond generic AI and develop systems that understand the specific complexities of network data and documentation. This will enable more accurate automation of network operations and help improve customer experiences.