AI for networking – the time is now

05 April 2023

AI is being lauded as the solution to modern network challenges, but is this the decade where every network gains a level of intelligence? Amy Saunders asks the experts

The first AI can be traced back to 1951, when Christopher Strachey developed a programme that could play checkers on the Ferranti Mark I computer. With the best part of a century under its belt, why is now finally the time for AI to join the network?

“Communications service providers (CSPs) have been under big pressure to cut operational costs, but until now they could get away with the traditional business and operational support systems (OSS/BSS) they use to manage the performance of their networks, while driving organisational and process transformations to improve efficiency,” says Sasa Crnojevic, network AI & machine learning business principal, SAS.

Jamie Pitchforth, head of strategic business, UK & Ireland, Juniper Networks, agrees that businesses are now under increasing pressure to perform at a higher level. “As such, IT decision makers can use AI networks to reduce increasing cybersecurity threats and other organisational challenges like hybrid working,” explains Pitchforth. “AI-based solutions can help to resolve the IT skills gap, allowing teams to work more efficiently with limited resources, and streamline processes to allow these resources to be allocated in the most effective areas. These solutions also facilitate the optimal user experience as issues are quickly resolved before working practices are impacted.”

The rollout of 5G is bringing more complexity and greater operational challenges which will be impossible to manage without AI and network automation. Those challenges, according to Crnojevic, include network cloudification; network automation; network monetisation (B2B2X); network slicing; and transforming from multi-play operator to full digital service provider.

“Imagine a simple 5G service like network slicing, which ensures a B2B customer has dedicated bandwidth with a required service level agreement (SLA), based on which the customer will be charged. It is simply not possible to manage dozens or even hundreds of network slices without automation across all layers (radio, core, and transport network),” says Crnojevic. “That’s why AI is a must. Not only for 5G, but also for legacy technologies to optimise operational costs and drive targeted capex investments.”

“With the current economic pressure, AI is being used to boost efficiency, future proofing an organisation by saving time and money with faster problem resolution, fewer onsite technician visits, and more streamlined network deployment models,” outlines Pitchforth. “To keep up with the development curve, IT decision makers will be required to implement AI technologies to maintain a steady pace with competitors.”

Highlighting the boom in cybersecurity challenges for today’s network users, Roman Tobe, product marketing director, global trust at RingCentral, states that “for security and risk management (SRM) leaders, the time to integrate AI for securing networks was yesterday.”

But before a leader can use AI to counteract risks, whether they are from malicious AI or more traditional techniques, understanding the overall team competency needed to assess these solutions is critical. “This could mean having a team of data scientists, analysts, engineers, or developers who can build and maintain the infrastructure needed to support AI initiatives. To do this, there needs to be more experienced security personnel on hand to counteract increasingly sophisticated threats,” explains Tobe.

Human vs machine

The discussion about humans being replaced by robots is widespread, with concerns that many thousands of jobs will be lost to self-scanning tills, driverless vehicles, etc. all making the headlines. Not to mention the inherent problems with today’s technology, which remains in its infancy, causing frustration for end users – ‘unexpected item in bagging area’ anyone?

These fears have been amplified since the November 2022 launch of ChatGTP, which has captured the global imagination with the use of natural language processing. “People have rightly been left amazed by its capabilities,” says Crnojevic. “But imagine adding other AI capabilities like computer vision, machine learning and deep learning, timeseries forecasting, streaming analytics. With all these tools we will be able to drive fully autonomous networks, or as CSPs define them, the ‘holy grail’ – ‘zero touch, zero wait and zero trouble’ services.”

For the networking world, AI will deliver a huge leap forward. AI processing works in a fundamentally different way to a human brain and can process huge amounts of data without getting bored or tired. It works on patterns and quickly learns from mistakes so is better at delivering consistency.

Pitchforth asserts that AI enables better performance in networking; connectivity solutions have the uninterrupted speed and bandwidth to perform at the required level, despite variations in traffic. Moreover, “the repetition involved in AI-powered networks lead to a reduction in human error, avoiding unwanted lapses in security and service quality,” he says. “Beyond the functionality of the network itself, AI solutions can also give better insights into how the network is performing and what it is being used for; increasing IT efficiencies, reducing support tickets, and decreasing main time to resolution, while transforming network operations from reactive troubleshooting to proactive remediation.”

“There’s a level of consistency that can be attained with AI systems that reduces the frequency of errors as well as the ability to introduce predictive analysis that prevents threats as they emerge,” says Tobe. “Sound AI systems are also self-correcting, albeit with human analysts overseeing the model accuracy and maintaining the standards set by the organisation and end users. An example would be an attack on an AI/ML model itself, where malicious code could be embedded into the proprietary model. Machine learning security (MLSEC) would be the best and fastest way to detect this intrusion and correct any further damage from occurring.”

“IT organisations must implement a solution that provides AI-driven operations with end-to-end visibility from the client to the cloud to optimise user experiences,” says Pitchforth. “With AI technologies, enterprises can streamline and automate SD-WAN configuration, event detection, troubleshooting and capacity management. When a network team applies these capabilities to an SD-WAN deployment, it can improve and protect user experience across a distributed enterprise.”

Incorporating AI into the network enables a level of scaling and anomaly prediction that, coupled with the possibility to act in real time, before services are impacted, makes it a no-brainer.

Network incorporation

Rolling out AI onto the network is no easy task.

When updating legacy technology, IT decision makers may face resistance to the implementation of AI, often by teams who have misconceptions around these innovations and how they will be deployed in the business. “Digital transformation also requires investment, which can be difficult to come by in times of economic downturn,” says Pitchforth. “However, many decision-makers now recognise the need to invest in AI for their business in the longer term.”

Meanwhile, Tobe highlights the challenge of maintaining trust: “one of the tenets of good AI practices, for external experiences within the products themselves is explainability, which is making the end user aware of the AI and why it made the decisions it made. Meeting this challenge goes a long way in elevating the trust and confidence in the ethical and unbiased nature of the AI system.”

As all network components must be compatible with each other, open and standardised APIs are key: “that’s where you have organisations like 3GPP, TM Forum and GSMA working to help their members. Proprietary equipment and interfaces were very common in the past, especially among many leading network equipment providers,” says Crnojevic. “Then we come to other things like analytics democratisation, which should allow business users faster access to insights through self-service, freeing data scientists to work on high-value initiatives like analytics development and digital transformation. Operators should also be mindful of GDPR and other privacy regulations.”

Naturally, the advancement of an AI depends upon access to suitable data. “Successful AI systems need a large amount of high-quality data to be trained on, which depending on the endeavour, can be difficult to collect and maintain,” explains Tobe. “Additionally, AI models are complex in nature and difficult for analysts to both interpret and decide on the appropriate response.”

“Enterprises need to responsibly manage AI’s growth with proper governance to stay ahead of regulation and minimise potential negative impacts. In Europe, regulators are starting to classify certain AI use cases as risky and requiring CE certification. AI regulation is changing quickly, and business leaders must make AI governance a strategic priority. Creating new regulations, setting up new governance frameworks or leveraging existing ones for entirely new technologies takes time, people, and capital,” says Pitchforth. “Yet, if there was ever a technology worth the investment, it’s AI. No matter what organisations spend today to build resilient AI governance structures, that investment will be trivial when compared to the resulting upside.”

An automated future

Will we see a day when every network features AI? “Without a doubt the answer is yes - but not at 100% full automation,” says Crnojevic.

“It’s almost a near certainty that by the end of the decade we will see AI in some form integrated into every network security system,” agrees Tobe. “We will also see AI engineering and governance principles guiding these systems to maintain a level of trust and ethical standards.”

It’s widely anticipated that AI will become universal within IT networks. “The well-publicised benefit of AI creating the secure, self-driving, self-healing network is an exciting prospect for business leaders who have lived with the cost of network operations spiralling ever since the advent of wireless devices and cloud applications in the early 2000s,” says Pitchforth. AI, machine learning and data science will underpin automation and optimisation to drive efficient secure networks with improved experiences.

“We are not too far away from this being reality, especially for greenfield networks which are not burdened with legacy architecture,” says Crnojevic. “You will always need people to calibrate and configure the AI in the way we want it, or in a way that is regulated. People will also be needed to ensure full transparency, to explain why a certain ‘AI system’ took a decision to understand whether it was justified to do so.”

The days of legacy on premise systems are limited “because software-defined cloud architectures driven by AI have arrived and their benefits are becoming publicised and realised. The key barometer suggests that the entire networking industry has pivoted towards AI, running on next generation cloud architectures,” says Pitchforth. “AI networking adoption will further increase as IT teams and business leaders demand more for less. Some sectors where policy demands data sovereignty may become late adopters, but in the not-too-distant future, AI will become a new normal.”

Crnojevic reports that he often hears the phrase: ‘It is not AI which will replace people, but it will be a person using AI.’ “It’s therefore important to transform our current workforce into citizen data scientists. Domain knowledge sits with the end business users and not all the tasks will be repetitive and therefore easy to automate, however the level of efficiency needed will be impossible to achieve without the implementation of AI,” says Crnojevic.