The Role of Gen AI in Telco: Handle with Care

Generative Artificial Intelligence (Gen AI) is having a substantial impact on the telecoms industry, with a number of highly interesting use cases emerging in the last year. Chatbots and copilots are particularly promising, offering assistance that is more finely tuned to the user than was previously possible with traditional Artificial Intelligence (AI). The primary use case is leveraging Gen AI chatbots for customer care services; Vodafone, Verizon, and MTN, among others, are already doing that with great success. A close second is to use Gen AI copilots for software and Application Programming Interface (API) development (with many using Microsoft’s GitHub Copilot), for Information Technology (IT) & security support (Nokia uses Gen AI to enhance its Extended Detection and Response (XDR) offering for telco Security Operations Centers (SOCs)), and for data analytics (Netcracker GenAI Telco Platform for Business Support System/Operations Support System ((BSS/OSS) integration). Beyond that, Gen AI is being tested for dynamic network optimization and predictive maintenance (AT&T is leveraging it to predict and address network congestion). There are many other potential use cases on the horizon, not just in customer service and network operations, but also in marketing and sales, for example.  

The excitement around Gen AI is driven, in part, by the rapid and significant gains that can be delivered by Large Language Models (LLMs). Traditional AI has not been able to significantly improve processes for the telecoms industry; in large part due to the difficulty in training effective models for networks as large as telcos that often span various cellular generations and different countries. But Gen AI’s ability to contextually understand, and importantly, discover complex patterns and make intuitive inferences in very large datasets, lends itself particularly well to the large complex telecoms networks of today.

In the security space, Gen AI is proving to be an exceptional tool to support security analysts not only in threat detection, but also in response. Nokia’s XDR solution, part of its NetGuard Cybersecurity Dome offering, includes a Gen AI assistant that provides highly detailed contextual information about adverse events, correlating and pulling data from various sources, with the ability to offer remedial next steps and automated playbooks for the analyst to run through. This type of assistant is highly attractive in a domain where there is a massive shortage of available security analysts and suitable threat detection tools that are specifically targeted at telco environments.

Equally, Gen AI is being used for fraud management as well, such as detecting Subscriber Identity Module (SIM) card cloning or billing fraud. Panamax and Neural Technologies offer revenue protection and fraud management solutions for telcos, and both are incorporating Gen AI into their platforms. The LLMs are much better at identifying fraudulent behavior than traditional AI. Their ability to continuously ingest data and refine their detection capabilities is key, closing the gap with adversaries that often change tactics and have plagued cellular networks ever since they emerged.

While Gen AI can offer significant advantages to telcos in many applications, it is also important to understand its limitations.  The first is to train the models on the right context. While they may be able to contextually infer from ingested data, that context needs to be defined accurately. A Gen AI model won’t work properly unless it’s been trained for the right context. This requires time and effort initially.

Beyond that, the decision to use any AI should be made with careful consideration. Telecommunication networks are part of a nation’s critical infrastructure. The industry tends to be more risk-averse for that reason. Many perceive the unbridled use of AI, especially to act directly on live networks, to be counter to national security prerogatives. There is too much scope for misconfiguration, as well as misuse, especially because AI can be subject to adversarial manipulation. Interruption or degradation of a telecommunication network could have significant economic repercussions and even pose a danger to people, especially as many other critical infrastructures rely on connectivity for their day-to-day functioning (hospitals, utilities, etc.).

But there are also other issues to consider. Telcos possess substantial amounts of confidential data about their subscribers; sharing these with a Gen AI model that hasn’t been fully vetted from a security perspective could be problematic in terms of privacy and data protection. The onus is on the telco operator to ensure that subscriber data are secured.

Ultimately, there is a balance to be reached between the efficiency and financial gains to be had with Gen AI and the potential risks it may pose to telcos as a critical infrastructure. There is no doubt that Gen AI is an appealing and highly promising technology, especially as the last few years have proven particularly difficult for telcos faced with high 5G Capital Expenditure (CAPEX), low Return on Investment (ROI), a global economic downturn, and aggressive competition from the tech sector.

But Gen AI, as with any AI, is not a silver bullet. And its capacity to alleviate those burdens will only be realized if it is used and implemented appropriately; this means both technology innovators and implementers should undertake comprehensive risk and cost assessments, ensure contextual training that is focused on telcos, and iterative and incremental deployment to ensure it does not pose a risk to national or subscriber security. As with any new technology adoption, the key is understanding the use case and Gen AI is no exception. It may well deliver for the telco industry, but it must be developed and handled with the care it deserves if any of its promises are to be fulfilled. For telcos, this means discussing openly with technology innovators like Nokia, Netcracker, Microsoft and others, that are implementing Gen Ai into their products to truly understand how the models can be trained effectively, and securely deployed to their specific use case.