The State of AI and LLM Usage in Telecom

Introduction

When evaluating the industries that have thus far adopted AI models at the most intense levels, one often considers the aptitude towards technological innovation, as well as the necessity for the sector itself to compete in continual operational advancements.

‍

Strikingly, telecom is one of such industries that fiercely tackles both.

‍

Telecom and AI Sentiment

As a whole, AI model adoption in telecom is intensifying. Microsoft fully reflects this in an article reviewing the AI in telecom sub-industry, stating the global AI in telecom market is to reach $38.8 billion by 2031. Notably in a 2023 study, Nvidia found that among 400 telecom respondents, 95% confirmed engagement with AI. Infact, over 64% of participants noted being in a Trial Stage or higher (Implementation, Using AI <6mo., Using AI >6mo.). In truth, it comes at little surprise that telecom stands as a strong end user of AI in operational usage due to the plethora of benefits to the sector; these benefits are exemplified by Forbes, where Vodafone implemented virtual assistants that led to a 68% improvement in customer satisfaction. Similarly, Nokia’s virtual assistant tasked with identifying network issues and solutions led to a 20-40% improvement in resolution rates.

‍

In a study by Bariah et al. (June 2023), this focal point centers upon the increase and potential of Generative AI in telecom - especially LLM models. Particularly, the introduction and development of Logical Learning Models (LLMs) in recent months has brought forth new, specialized methods to approach technological innovation as the industry prepares for the AI-powered journey towards 6G integrations.

‍

As enthusiasm for AI implementation uplifts telecom at its core, we will dive deeper into the intricacies that are driving this momentum.

‍

Most Common AI use-cases in Telecom: the rise of LLMs

Presently, AI serves a number of purposes in Telecom with tangible effects. Such includes the following:

‍

- Detection of Unauthorized Network Access: Monitoring network activity and identifying any unusual patterns that may arise, using machine learning (ML) to avoid the restrictions of the usual rule-based system

- Predictive Maintenance: Anticipatory AI that functions by analyzing actual measured usage, operating conditions, and equipment feedback of telecom machinery. Predicts when repairs are needed, allowing for reduced downtime and reduced associated costs

- Detection of SIM-Swapping: Prevention of SIM-swapping fraud by analyzing patterns in the usage of SIM cards, such as sudden changes in location, type of device, and calling behaviors

- Prevention of Bill Fraud: Bill fraud, also known as International Revenue Sharing Fraud (IRSF), poses as one of the largest security challenges to telecom companies. Prevention may include the analysis of abnormal calling and billing data

- Personalized Marketing: AI can analyze individual customer data to create cross-selling packages to respective clients based on usage patterns and preferences

- Automated Decision making: By utilizing deep learning models, decisions regarding network routing, dynamic pricing, and other operational variables can be automated

- Assistive bots: As seen with Vodafone, such digital assistants help telcos streamline and address customer complaints—standardizing the user experience and providing immediate assistance or guidance to clients

‍

At the basis of its implementation, AI guides telcos in solving their persistent pain points. Nvidia notes telecom’s top goals when implementing AI to be the following:

‍

- Optimize operations (60%)

- Reduce costs (44%)

- Enhance customer engagement (35%)

- Meet revenue targets (31%)

‍

To epitomize this with respect to fraud risk - accounting for an estimated $40 billion cumulative loss in revenue for 2021 (2.22% of total revenues) - algorithms detecting abnormal behaviors have become an increasingly pertinent application.

‍

Prospects of LLMs in the Sector

Regarded by Bariah et al., LLMs are envisioned to “open up a new era of autonomous wireless networks, in which a multimodal large model trained over various Telecom data, can be fine-tuned to perform several downstream tasks, eliminating the need for dedicated AI models for each task”. Such revolves around the concept of the “self-organizing network”, where AI can both self-build and self-evolve with respect to the ability to adjust and optimize their functions and parameters according to network conditions and user demands.

‍

The paper continues to expand on this notion, associating LLMs’ ability to train over a vast amount of unlabeled, multimodal data, capture statistical patterns, and host generative capabilities to soon enable wireless networks to hold predictability features and spectrum management even for unseen network scenarios.

‍

While most present AI solutions in wireless networks are tailored to solve dedicated problems, the addition of LLMs aims to solve these unpredicted tasks. The two primary pillars of LLM for Wireless and Wireless for LLM thus arise. As outlined by Bariah et al.:

‍

- LLM for Wireless enables multiple tasks in wireless communications and sensing, where multi-modal LLMs could associate radio signals with image, point-cloud, or sound - enhancing environment reconstruction, positioning, and human pose estimation;

‍

- Wireless for LLM revolves around the future network devices equipped for LLMs, with 6G networks transferring wireless networks from being data & model based to knowledge & reasoning based. As a result, on-device LLMs can interact with each other to accomplish complex tasks while minimizing resources and energy

‍

Moving Forward: Validation and Risk Management

The complexities, and occasionally unpredictable nature, of LLMs can give way to numerous risks; due to their unstructured nature, continuous validation of the model and its application of data is vital to ensure technical, ethical, and regulatory risks are within a company’s governance standards.

‍

As such, a new method of validation and audit must be adopted in order to adapt to these growing concerns across the industry.

‍

At Calvin, we look forward to guiding the potential of AI in this technical field - optimizing and alleviating the concerns of risk management through our modularized range of assessments amidst firms’ dynamic AI portfolios. With our quantitative, holistic approach, our goal is to guide your firm towards AI excellence, ensuring safety both on the internal and external stages.

‍

Interested in increasing your models’ fairness and explainability for your client base? Book a demo with us, and let us show you how we can enhance your AI systems' effectiveness and trustworthiness in today’s ever-changing landscape.