Can we trust GenAI in telecoms?

Everybody talks about Generative AI (GenAI), and everybody uses or has played with ChatGPT. Yet, everybody seems uncertain about what GenAI is, and what it means for society and, more specifically, for telecoms. A survey illustrates this lack of clarity: AI is the top investment priority in 2024 for 44% of operators, and yet 43% consider GenAI one of the most hyped technologies (followed by AI at 38%).

GenAI’s sudden rise to prominence coincided with the public availability of ChatGPT, a startling showcase of the capabilities of the technology. This has shifted much of the focus from AI and ML to GenAI, and often GenAI is taken to be the latest and most successful incarnation of AI. This is unfortunate because it does not do justice to any of them and misrepresents their potential contributions and risks to telecom.  

What is GenAI – and what is it not

GenAI models use massively large amounts of data to learn to generate plausible new text, graphics, or other content, typically in response to a request in a prompt. Their power and success depend on the availability of extensive and cheap computational resources to train GenAI models. These models can parse more data than a single person can but are not as subtle as most of us in deciding what is accurate, relevant or true. This is what makes them a useful complement to what we do. But this is also what makes it dangerous to rely too much on them because they are likely to generate some plausible but factually false content.

AI has a wider remit, which includes GenAI, but also, more importantly, models to understand and learn from data and to act in response. AI has been around for decades, but only in recent years it has become affordable and reliable enough to be used in telecoms and other industries. Unfortunately, because of the hype, many incorrectly label traditional deterministic algorithms or more mature AI and ML tools as GenAI, and this contributes to confusion on what we should expect from – or worry about – in GenAI.

In telecoms, AI and ML contribute a large set of tools that can be used, from network planning and operations, to service assurance and subscriber services. GenAI can play a valuable but niche role within this context, for instance in generating content or making network and subscriber data available to staff and subscribers. It would not be an appropriate tool to optimize a massive MIMO radio, however.

Understand opportunities, understand challenges

GenAI can be a powerful technology that provides a general-purpose foundation that can be customized to support various automation and human-augmentation tasks, such as customer support, or document parsing and summarization.

The use of GenAI for network operations requires more scrutiny and guardrails than the behavior of generic GenAI foundation models, because of the criticality of the telecom infrastructure.

We also need GenAI models that are specifically tailored to the telecom infrastructure. All-purpose GenAI models are not well suited to efficiently use domain-specific expertise and are bound to generate misleading or irrelevant findings. They may generate insights on a specific aspect of network data provided in a natural language prompt that may not be pertinent to the network state or fail to provide relevant or actionable information without observability over network operations, access to human expertise and continuous feedback to fine-tune GenAI models.

Still needed: Experience, good data

GenAI is not an off-the-shelf solution or tool that operators can plug into any network and get to work right away. They still need knowledge of their network and experience in running it to guide the model and to select the right data to feed. GenAI models thrive on large data sets, but larger data sets may be noisier and, hence, lead to less accurate results. Running models on reliable and high-quality data, and having well-defined targets are crucial to successful outcomes.

While GenAI, AI and ML have the potential to fundamentally change the way we build, run, and optimize networks, a solid data strategy, good visibility in the network, and in-house expertise are traditional capabilities that are still essential and provide the necessary foundation even for disruptive network transformation.

Preserving observability, trust and reliability

GenAI models can process larger amounts of data than a human. As a result, we have less control over the outcome because it is difficult to track how the model got to it. This reduces observability and creates reliability, trust, and security challenges. These issues are not new but are more severe than for previous technologies. We should not underestimate these aspects, but we should not let fear unnecessarily limit the potential of GenAI, AI and ML.

What’s next: Don’t believe the hype, but move forward

Despite the hype and the uncertainty about what we can expect them to do reliably and safely, GenAI, AI and ML will have a large positive impact in driving automation and optimization in telecom networks, as well as in other verticals. They will also introduce profound changes in the culture, work environment, and workforce training – which may require time and effort to adapt to.

Yet, while we need caution to assess the challenges that GenAI and AI expose the networks and our industry to, we should embrace them as tools we need to manage the increased complexity and capabilities of our networks, their performance, cost-effectiveness, and power efficiency. It is not just hype, as long as you have a solid data foundation to use GenAI to assist in planning, running and optimizing your network.

We use cookies to improve your browsing experience. By continuing to use the website you allow Senza Fili to deploy cookies, as detailed in our privacy policy.