Telecom Review has successfully hosted its latest webinar session entitled, “Riding the Wave of Generative AI,” co-sponsored by Nokia, Umniah, B-Yond, Salam, and Azerconnect Group, covering all aspects of generative AI and providing insight into what's beyond the hype and current real business applications.
Discussing the trends, advancements, best practices, challenges, telecom use cases, and predictions related to Generative AI (GenAI) were Abhay Savargaonkar, Vice President and Head of Technology, Cloud & Network Services, Nokia; Alaa Ibrahim, Chief Technical Officer, Umniah; Rikard Kjellberg, Chief Product Officer, B-Yond; Ayman AlFadhel, VP of Cybersecurity, Salam; and Mushfig Aliyev, Chief Commercial Officer, Azerconnect Group.
Hamza Jadouane, Manager at the Datalab of PMP Strategy, served as the moderator of the session.
Welcoming the audience, Toni Eid, Founder of Telecom Review Group and CEO of Trace Media International, emphasized that this panel was organized to assess the readiness of mobile operators and vendors, as well as their objectives in deploying GenAI.
Telecom Review Analysis: Technology Advancements Are Upon Us: What Will the World Become?
The State of Generative AI: Current Trends and Advancements
When asked about his perspective on why there is suddenly so much hype around generative AI, Savargaonkar cited a creative example wherein different ingredients to make a custard are on the table. Past AI models will be trained to automatically create a custard, but with generative AI, five different recipes could emerge using the same ingredients.
This showcases the “fundamental difference” between previous AI and generative AI, which focuses on generating new content.
Acknowledging a notable trend, the Nokia executive mentioned that NVIDIA’s share prices increased by 150% in the first half of 2024. There has also been a “reality check, as everybody now knows what generative AI, or AI in general, can do, and what it can and cannot be used for.”
Highlighting Nokia’s breakthrough innovations, Savargaonkar articulated that their operations revolve around productivity enhancement, which is an aspect they have bifurcated on. For example, the Nokia Cybersecurity Dome is an overarching solution for threat identification, detection, and verification. Using AI, Nokia has replaced the expert security monitoring team needed to monitor, run, and handle any security products.
In the realm of telecom network complexity, Nokia has utilized the concept of creating a digital twin, incorporating AI to handle software upgrades, threat prediction analysis, and other real-life network tests digitally. Moreover, Nokia’s digital system has become intent-based to eliminate complexities in overall product scenarios.
On the productivity side, Nokia has curated its large database through LLM processing and extracted useful information, making quicker network diagnostics of similar level 3 problems.
Ibrahim concurred that predictive maintenance has been very useful to Umniah. “Self-optimizing networks (SONs) are already part of the 5G standard but when you enable AI on top of a SON, you get predictive capabilities,” he continued. This not only provides root-cause analysis, but also forecasts what can happen due to the evolution of the traffic and provide more concise recommendations to proactively catch a problem before it happens. He mentioned that greater traffic mobility has been observed as a result.
For the past year, Umniah’s CTO has observed a very good progress as “AI models are becoming more efficient, and that's very good news for us, because that reduces the computational power needed to run these models.” This is a positive sign that “we can continue to depend on AI to do what we are looking for.”
Highlighting commercialization, Umniah’s CTO thinks that generative AI is “playing a big role in the personalization of services,” bringing entities closer to customers, understanding their needs, and tailoring services to meet their requirements. He also believes that AI hardware is becoming more resilient.
Using LLMs, chatbots are also becoming more capable when dealing with customer needs. Umniah is capitalizing on this intelligence in its operations.
Salam’s VP verbalized how AI will be helpful in monetizing services, reducing costs, and enhancing customer experience. Notably, AI can reduce the number of people in call centers, and can even suggest more services and products by understanding and tracking the historical data of a customer.
Mushfig Aliyev, Chief Commercial Officer, Azerconnect Group, further highlighted the transformative impact of GenAI on customer care. He noted that within just six months of implementing GenAI, there was a 10% decrease in the number of calls forwarded to customer care, showcasing improved efficiency in customer interactions. Additionally, the workforce employed in customer care saw a 20% reduction, indicating significant cost savings and operational streamlining. GenAI also contributed to a 10% decrease in the workforce needed for marketing and communications by providing valuable advisory services, enhancing customer interactions, and generating AI-driven content.
From a technologist’s perspective, B-Yond’s CPO sees a “technology stack emerging.” These include LLM technologies, retrieval augmentation, knowledge graphs, and large-action models. He mentioned that B-Yond is entering the space of large-action models with automated remediation.
B-Yond’s core product is a 4G and 5G network diagnostic solution, Kjellberg emphasized, “We use generative AI in combination with other AI and ML techniques to create a whole chain of actions, which create useful functionalities.”
Some of the scenarios he mentioned include AI-based image recognition technology to conduct real-time pass/fail tests of network traffic at-scale as well as symbolic ML models to conduct deep-root cause diagnostics of failures. The diagnotics results are combined with generative AI (LLMs) and other information sources, including their own proprietary tribal knowledge and 3GPP documentation, to provide a more readable result to users.
From a technological front, the existence of small language models (SLMs) amidst large language models (LLMs) has also been observed. Kjellberg noted that this facilitates AI at the edge, making it suitable for telco networks and mobile devices. B-Yond has created SLMs that are as small as 10 GB that can reside on the device itself.
“I think there's also a lot of activity at the inference layer where you're leveraging LLMs for niche- and domain-specific applications,” he added, mentioning that there are few players working at the foundational level on the LLMs themselves, such as start-ups (OpenAI, Anthropic, and Perplexity).
Additionally, Aliyev stated the recent trend of applications becoming available in local languages, which are mostly free. In the rebranding of one of the telco players that Azerconnect Group manages, they have utilized generative AI for content generation. He also took the opportunity to highlight the convergence of GenAI and personalized marketing, emphasizing that telcos stand to gain from hyper-personalized localization.
Multi-modal AI models are also among the trends mentioned which involves combining images, audio, and other elements together.
Solutions and Best Practices for Implementing GenAI
Regarding the key considerations and challenges businesses must consider when looking to implement generative AI and the best practices that should be followed to ensure successful GenAI deployment, Rikard Kjellberg, Chief Product Officer (CPO), B-Yond, noted that this is a really interesting space. In working with many of the largest telcos in the world, a crucial question is often encountered: should they go into it alone, or should experts in the field be partnered with?
When the decision is made by telcos to go into GenAI implementation alone, it must be understood that merely having all the data is not enough. A well-integrated team of domain experts, including those who understand the telco domain, data scientists, and expert developers, are needed, all of whom must share a common understanding and work cohesively.
Having the data is just the beginning; it is crucial for the data to be automatically curated into high-quality datasets that can be effectively utilized. Employing various machine learning and AI tools, including generative AI, in iterative fashions is vital. This represents a new way of working for many telcos and requires a robust infrastructure strategy. Decisions must be made about whether to use in-house data centers or leverage public clouds. It is observed that many objections to public clouds are based on security concerns. However, Kjellberg personally believes that more security issues are encountered by cloud providers than any other enterprise, and some of the best security measures globally have been implemented by them. “Proprietary data can be exceptionally well protected, and more telcos are starting to recognize this, though adoption could be faster, in my opinion,” commented Kjellberg.
Another critical consideration is the required response speeds of the AI itself. For instance, in network operations, real-time performance is key. Large language models are not yet suitable for generating real-time results. However, on the support side, they are perfectly suitable. Therefore, a multi-AI technology approach is highly beneficial for network operations.
It is believed that if partners are chosen, adopting a build-operate-transfer model is a good approach. This allows telcos to eventually take ownership of the results as the necessary expertise is gained over time.
As per Mushfig Aliyev, Chief Commercial Officer, Azerconnect Group, the execution of the roadmap often presents significant challenges, particularly in terms of planning milestones accordingly and executing them in a timely manner. Unlike software development, AI planning cannot be exact and precise on a day-to-day basis. This is a critical difference that we have learned through experience.
Aliyev explained that, internally, two teams have been maintained, and are referred to as the digital lab, which comprises 300 people. In addition, partnerships with global experts based in India, Estonia, Poland, and other locations have been leveraged. A common observation, whether it be regarding internal resources or external partners, is that time is required to learn and apply new knowledge. This differs significantly from software development, where quarterly-based planning is more straightforward. In AI initiatives, planning often shifts due to the nature of the work, which demands continuous machine learning and adaptation.
To address these challenges, an AI unit was launched under the commercial division to ensure easier interaction with customers. This unit is enabled by internal resources and aims to leverage global partnerships effectively. This approach has been beneficial over the last three months, allowing for corrections and adjustments in Azerconnect Group’s roadmaps.
According to Ayman AlFadhel, VP of Cybersecurity, Salam, a clear strategy and objective should be established for using AI. The reasons for adopting AI and the areas where it will be applied must be clearly understood. A specific department should be created to manage the large volumes of data that AI will generate and use. The tools and expertise necessary to handle and analyze the data effectively must be provided to this department.
AlFadhel mentioned that data protection is another crucial point. The risks of data leakage posed by AI should be mitigated by using proper consideration and control over data classification. For instance, sensitive or classified data should not be inputted into AI systems like ChatGPT, as the information might be used to answer queries from other users.
Lastly, AlFadhel reiterated that investment in training is essential. The benefits of AI and related tools can only be maximized if employees are well-trained. This is a new field, and without proper training, the full potential of AI technology cannot be realized.
Alaa Ibrahim, Chief Technical Officer (CTO), Umniah noted, “One point from my side, based on practical implementation, is the significant challenge of implementing the closed-loop concept, where AI is trusted to handle the full loop until the implementation stage. This challenge highlights the importance of explainable AI.”
A lot of work needs to be done by vendors to explain AI and its models to the people who will be using these tools. Ensuring that AI is understood, and its processes are transparent will ease its adoption into daily operations.
“The challenge that needs to be addressed is making sure AI is explained to everyone and understood by all, so it can be seamlessly integrated into daily operations.”
Abhay Savargaonkar, Vice President and Head of Technology, Cloud & Network Services, Nokia added, “I would like to add one critical point, which is managing data bias or hallucination, and it is indeed very critical. An example illustrates this well. Consider a chatbot named Rock and an NBA player named Clay Thompson who uses a basketball slang ‘shooting bricks.’ Due to this slang, the chatbot, Rock, mistakenly accused him of vandalizing homes. This incident highlights how detrimental data bias or hallucination can be if not properly addressed when curating and integrating data.”
Regulatory Compliance
Concerning the steps that companies should take to ensure compliance with AI-related regulations and how should businesses address transparency and accountability in generative AI, in Ibrahim’s opinion the need for a dedicated regulatory team for AI compliance is critical.
The establishment of a dedicated team for AI compliance is crucial for progress in AI initiatives while adhering to regulatory requirements and safeguarding data integrity.
Expanding on the regulatory realm, AlFadhel explained that AI revolves around data and its criticality extends to cybersecurity. Therefore, data protection through effective data governance is essential. When considering data governance, AlFadhel highlighted four key domains to focus on:
- Data Assessment: This involves cataloging data, ensuring data quality, managing documents, and overseeing data operations.
- Data Utilization: This domain covers using data for business intelligence, establishing data sharing policies, and ensuring compliance with data sharing rules and regulations, including open data initiatives.
- Data Classification and Availability: It is crucial to classify data appropriately to manage and restrict its sharing effectively.
- Data Protection: Adherence to multiple regulations and frameworks, such as those from the Saudi National Cybersecurity Authority and the Saudi Data & AI Authority (SDAIA), is necessary to ensure robust data protection.
Managing and protecting data, including ensuring data privacy, is critically important in AI applications. Data sensitivity can have profound impacts not only on organizations but also on national security, especially in sectors like telecommunications, which are vital infrastructures. Therefore, comprehensive data management and governance are paramount to mitigate risks associated with potential data breaches.
Azerconnect Group’s Aliyev noted that there have not been any new regulations regarding AI compliance. However, he emphasized that businesses should remain vigilant about customer data privacy and use it as a guiding principle in their operations and implementations. He also highlighted the importance of ethical considerations in AI deployment. Aliyev suggested that implementing use case team analysis can aid in shaping future strategies, ensuring that ethical and privacy standards are upheld while leveraging AI's potential.
GenAI and Telecom Use Cases
Sharing B-Yond’s GenAI use cases, Kjellberg said the technology is being utilized to enhance customer support operations by reducing support staff in handling call volumes. B-Yond has used chatbots to increase customer interaction and has implemented automated voice interaction in sales and marketing.
In B-Yond’s operation and network management sector, he said GenAI is being used for diagnostic purposes, real-time monitoring, and analyzing SLAs. “We're working with the telcos right now all over the world for continuous sampling and diagnostics to monitor VIP customers or specific enterprises.”
He highlighted that B-Yond is utilizing an operational triage model in its cyber incident response to investigate network alerts. “Reducing mean time to resolve and restore service is another area that there's a lot of activity and work going on,” he noted.
He also mentioned that B-Yond is leveraging GenAI in anomaly detection, automation and closed-loop operations. Importantly, he stressed the need to educate staff about the risks of using GenAI.
Concurring with Kjelberg’s response, Savargaonkar said Nokia heavily utilizes GenAI’s capabilities to enhance its customer experience and network performance operations. He also stressed the importance of implementing digital twins to enhance the network’s performance.
The moderator of the session, Hamza Jadouane, Manager at the Datalab of PMP Strategy, interjected, affirming that action models are at the forefront of GenAI implementation.
Ibrahim highlighted that Umniah has three successful GenAI use cases for network optimization and management. Firstly, Umniah has deployed a cognitive AI traffic tool that forecasts network traffic and diagnoses network issues. Secondly, Umniah’s GenAI-enabled power control has introduced a smart deep-sleep cycle, resulting in a 10-30% power saving. Thirdly, Umniah has implemented an abnormal network behaviour identifier, which enhances root-cause analysis, improves latency, and increases network throughput. Ibrahim emphasized the importance of proactively detecting and understanding the root causes of network issues to take appropriate measures for resolution.
To enhance customer experience, AlFadhel emphasized that Generative AI (GenAI) is ideal for network optimization, which could eventually lead to a better customer experience. He also highlighted the integration of all-bandwidth analytics and noted the importance of feedback analysis, bolstered by GenAI capabilities. Additionally, AlFadhel mentioned that GenAI can be utilized in marketing for scalability and competitor research, showcasing its versatile applications in enhancing both operational efficiency and strategic planning.
Highlighing the impact of GenAI on telecom revenue and operational efficiency, Aliyev reiterated that a 20% reduction in customer care and marketing staff could be achieved. He also noted that knowledge-sharing of GenAI insights with staff could help overall customer interaction operations. He noted that the new trends in marketing and communication, driven by Generative AI (GenAI), can provide personalized services, resulting in a 5% revenue increase. He also mentioned that 20% of his company's sales volume was derived from AI-backed telesales.
Commenting on identifying and mitigating cybersecurity threats, AlFadhel said that GenAI can be used for threat detection, predictive maintenance and can analyze networks, providing mitigation solutions. He also stressed the importance of cybersecurity personnel matching the expertise of hackers who use GenAI for sophisticated attacks on networks. “We should protect ourselves by using the same technology [GenAI] used by the attackers and enhance our way of protecting the organization.”
Predictions for the Future of GenAI
Predicting the most exciting future trends in GenAI, influencing industries and business models, Kjellberg said the speed of GenAI applicability will grow exponentially. He pointed out that the human-to-machine interaction will have a tremendous impact on businesses. He also noted the proliferation of multi-modal LLMs, GPU architecture, powerful AI-enabled APIs and memory architectures.
He also emphasized the importance of innovation in data centers and the recent capabilities of hydropower in data center operational enhancement. From a telecom perspective, he highlighted the compatibility and interoperability of GenAI applications and SDN architecture, including network slicing, purpose-built networks, software-defined 5G networks and intent-based solutions.
Aliyev foresees a tremendous growth in content generation applications while AlFadhel pointed out that the combination of ubiquitous AI, quantum computing and machine to human interaction capabilities would have far-reaching consequences.
Audience Standpoint
One of the questions asked by the audience during the panel addressed the developments in the filtering process of datasets in the training of generative AI models. Savargaonkar affirmed that while typical techniques used for regular analytics must be considered, it is critical to employ datasets from other systems to avoid creating bias.
Kjellberg explained how B-Yond utilizes packet captured data, which is often confusing to LLMs. In this case, filtering is required as you have to understand the domain and the specific technology that you are working with.
Relating to this, AI will not be able to recognize a specific type of data without the help of a human feeding it with high-quality data. Hence, responding to the query regarding the balance between workforce development and the AI revolution, Salam’s executive stated that “AI will not replace humans because AI, at the end, will not think as a human.” AI will eventually change the workforce from “doing a repeated job to a smart job.”
Debunking the question about the profitability of telcos relying solely on energy consumption when using AI, Azerconnect Group’s Chief Commercial Officer highlighted various aspects such as staff reduction, reduced time-to-market, customer value management, and marketing.
Nokia’s executive also pointed out that the data centers of today are not “very well designed” for the kind of racks used for GenAI GPUs. “A lot of changes would be needed, and it’s not an easy thing to do.” Use cases and the type of infrastructure—cloud or hyperscale—moving forward will be a big consideration.
Poll questions were also launched to gauge the outlook of the audience regarding generative AI. The majority of the audience (66%) views data privacy and security concerns as the biggest challenge in implementing generative AI in businesses, while 28% believe that a lack of use case understanding and well-defined purpose hinders GenAI implementation.
When the audience was asked which aspect of telecom operations would benefit most from AI-driven processes, 44% answered that customer experience and service personalization would see the greatest benefit.
Inevitably, 63% of participants believe that the telecom, media, and technology (TMT) industry is expected to see the biggest impact as a result of generative AI over the next five years.
