Telecom AI: Increasing AI Integration in Networks of the Future

This article is part of the essay series – Nations, Networks, Narratives: World Telecommunication and Information Society Day 2025.  

The past two decades have witnessed a revolution in telecommunication technologies. The 2000s saw the introduction of 3G technology, which offered a higher bandwidth (up to 21 Mbps) that allowed support for multimedia applications and services on mobile phones, to 5G in 2020, which pushed data rates up to the gigabytes per second scale and facilitated optimisation techniques like network slicing. Current trends in telecom technologies suggest that the following years will see the introduction of 5G-A (Advanced) and 6G technologies. One crucial element of the evolution of telecom has been the integration of machine learning (ML) and artificial intelligence (AI) algorithms for complexity management in telecom networks. The role of AI in the 3G era was limited to fault detection and predictive maintenance. From 4G onwards, AI began to play a more expansive role in telecom networks such as optimisation of network resources, data traffic management, spectrum allocation and eventually for developing AI-driven customer support systems.

AI and telecommunications form a recursive feedback loop where AI is used to manage complexity and facilitate more devices and services on the networks, and the injection of more devices and services in turn increases the demand for more powerful and capable AI systems. 

AI and telecommunications form a recursive feedback loop where AI is used to manage complexity and facilitate more devices and services on the networks, and the injection of more devices and services in turn increases the demand for more powerful and capable AI systems. Till 5G networks, the integration of AI was an added feature to optimise network performance. On the other hand, the upcoming 5G-A and 6G technologies are being designed to be “AI-native” by having AI built into their architectures from inception, such as by co-locating AI capabilities with Radio Access Network functions (RAN). A notable benefit of AI-native architectures will be the capability to extend AI services to edge computing through “multi-tenancy” or hosting edge AI applications, voice, data and video on the same telecom networks, thereby reducing latency. While AI-native telecom networks will be crucial for technological projects like smart cities, Internet of Things (IoT), autonomous driving and so forth, they also intensify risks such as unethical surveillance capabilities, data harvesting that may infringe on consumer privacy, and increased energy consumption that may put increasing pressure on local energy infrastructures.

Promises of ubiquitous AI in networks

The most significant benefit that AI offers for advanced networks is managing network complexity arising from the exponentially larger number of digitally connected devices that have entered the market over the previous decades. The ubiquitous spread of digitalisation is creating a scenario where human capabilities may not be enough to handle ever-increasing network traffic created by billions of devices that are active at any given moment. AI algorithms in this context can be used for efficiently analysing vast streams of network data by employing modalities like Convolutional Neural Networks (CNNs) to predict areas of congestion and anomaly detection, Long Short-Term Memory (LSTM) Networks for predicting traffic demand trends and Generative Adversarial Networks to generate synthetic data for training models with limited data to protect privacy.

AI-driven solutions are already being integrated into the telecom industry. For instance, telecom industry leader Vodafone has partnered with Google Cloud to embed generative AI into its network infrastructure to streamline network engineering, analyse contractual data, and provide real-time information to field technicians to reduce the number of field dispatches. A foundational promise of future networks with embedded AI is automating operational decisions and creating a “zero-touch paradigm” where minimal human intervention will be required for network management. Telecoms are already making headway towards this paradigm shift through initiatives like China Mobile’s Level 4 autonomous networks that have demonstrated the ability to increase data analysis efficiency by over 70 percent and reduce annual operational expenditure by US$ 7 million by using generative AI solutions to increase network optimisation.

The ubiquitous spread of digitalisation is creating a scenario where human capabilities may not be enough to handle ever-increasing network traffic created by billions of devices that are active at any given moment.

Another crucial benefit of AI-native telecom networks is increased security against cyber threats and fraud that can target consumers as well as network operators. In 2024 alone, the global telecom industry experienced a record-breaking rise in ransomware attacks, indicating that increasing digital connectivity across the globe to bridge the digital divide must be accompanied by scalable cybersecurity measures. To address this issue, AI-native networks may offer the benefit of leveraging Threat Intelligence Platforms (TIPs) that, instead of relying on historical data, can perform real-time data analysis to promptly identify emerging threat patterns through predictive capabilities. Moreover, AI-enabled TIPs can also automate threat detection and redressal mechanisms and bring the zero-touch paradigm to cybersecurity as well.

Dual-use Risks

While telecom AI offers a range of benefits, the increasing integration of automotive modalities in networks presents risks and policy challenges. Due to its nature as a dual-use technology, AI can also be used by bad actors to implement increasingly sophisticated threats such as AI-generated malware attacks, enhanced phishing campaigns and automated vulnerability exploitation. Furthermore, the expansive capabilities of AI-native networks also raise ethical questions about government overreach and surveillance operations. In the US, the Department of Defense has already established a FutureG team to study the impact and development of 6G technology. One emerging modality of import is Integrated Sensing and Communication (ISAC) that will allow the integration of sensing capabilities directly in the 6G network infrastructure. This sensor integration will enable methods such as high-precision positioning, environment mapping, object detection and gesture recognition by allowing AI systems to analyse radio waves, “effectively transforming the network into a distributed sensor array.” While the initial use of such methods likely will be in the defence sector, the potential for sophisticated surveillance techniques in civilian environments needs extended deliberation by policymakers.

One emerging modality of import is Integrated Sensing and Communication (ISAC) that will allow the integration of sensing capabilities directly in the 6G network infrastructure.

Another risk of expanding AI adoption is the rise in AI biases. Various examples have been brought forth in recent years of biased representation of certain demographics and genders in the training data and output of generative AI. In the telecom AI context, such issues can manifest through biased pricing and service offerings, biased allocation of resources and bandwidth to specific areas, problematic customer service instances, fraud detection systems unfairly flagging demographic groups and so forth. Similar to generative AI, policies for regulating telecom AI should focus on ensuring that training datasets for AI models do not reflect historical biases through algorithmic transparency requirements. Furthermore, international collaboration through inter-governmental and public-private partnerships is required to arrive at collectively accepted technical benchmarks that can provide best practices and guidance for telecom operators.

Concluding Remarks

A major policy challenge arising from increasingly sophisticated AI systems, generally, and in the telecom sector specifically, is the flexibility and adaptiveness of regulations in the face of a fast-moving technology. Furthermore, with the emergence of AI agents that can operate across systems and make decisions, there arises a complex set of questions regarding unpredictable emergent behaviour resulting from interactions between autonomous agents. Attempts at regulating such a dynamic set of technologies may require policy approaches to be flexible with an emphasis on co-regulation of networks by the public as well as the private sector. Given the complexities emerging from growing network traffic, regulations may be more efficacious if they are outcome and principles-based, with a focus on desired results and collectively agreed upon governance standards rather than the means of achieving such results. The distributed nature of telecommunication networks and the threat of cross-border security risks will also require increased collaboration through international platforms such as the International Telecommunications Union and the Global Coalition on Telecommunications. Such platforms can be used to promote multi-stakeholder engagements, prevent regulatory fragmentation, and also prevent onerous regulations that may hinder innovation.

Siddharth Yadav is a Fellow with the Technology vertical at the ORF Middle East.