日本語
In a Ciena survey of more than 1,500 telecom and IT engineers and managers conducted in March 2024, more than half of the respondents said they expect AI to improve network operational efficiency by 40% or more.
There is no guarantee AI will deliver.
Those implementing AI have concerns that run even deeper. They can’t say for sure how it will behave in the network. They don’t know if they can trust it to make the right decisions. Then there are the privacy and security challenges associated with feeding models vast amounts of sensitive data and ceding control over network security mechanisms.
At the same time, these stakeholders know they cannot miss out on the AI opportunity. They must proceed but need to do so safely, confidently, and quickly.
Spirent sits at the intersection of network testing and AI, where these problems are being addressed with new approaches and tools. I recently shared insights from our early work on this front during the RCR Wireless News “How is AI Being Applied in Network Testing and Assurance?” webinar, which examined the role test and assurance plays in validating AI in networks, as well as AI’s emerging role within testing tools themselves.
Following is a summary of remarks I shared with the audience about challenges encountered to date, testing’s role across the AI lifecycle, and the path appearing toward AI autonomous networking.
Learning How to Trust AI in the Network
Navigating endless buzzwords and boastful claims is just the beginning of clearing AI hurdles. In addition to the concerns mentioned above, operators are struggling with whether AI can be trusted to perform as expected in live networks given AI decision making isn’t always transparent. Operators want to be sure AI models that are implemented will not cause an outage or negatively impact performance. They also want to understand what to expect as it relates to ROI.
The answers to these questions are rarely known at the start of the AI journey. And to be sure, it is a real journey given integrating AI into the network is not a “one and one” event. Rather, it is a continuous process that spans design to deployment and operations. Testing plays a pivotal role in learning how AI will perform at each of the phases in the AI deployment lifecycle, ensuring it is safe, effective, and delivering as expected.
Here is a snapshot of the lifecycle and the testing that occurs at each phase:
Design and development (day zero). The beginning of the AI lifecycle is when models are created and trained to perform specific functions within the network with a focus on ensuring they are realistic and fit for purpose. Testing at this phase focuses on simulating and emulating networks in a realistic risk-free environment to see how AI will perform under varying conditions, before it ever touches a live network.
Deployment (day one). This is the critical moment when AI is introduced into live environments. Non-functional testing for resilience, scalability, security and network performance plays a critical role in ensuring the AI enabled system or network can scale and operate under real world conditions without causing disruption.
Operation and maintenance (day two and beyond). Once deployed, AI models continuously learn and adapt to new data as the network evolves. Testing takes on multiple roles in this phase, proactively monitoring AI’s performance to ensure it is delivering desired outcomes while providing feedback to adjust or retrain the model as network conditions change. This continuous active testing approach gives operators the confidence that the AI is optimized, transparent, and aligned with network objectives.
A Foundation for Autonomous Networks
The work being done in AI today lays the foundation for autonomous, self-managing networks that will fundamentally change how telecom networks are operated.
Today, AI is positioned to support predictive analytics, fault detection, and basic automation, with substantial human intervention still required. In this stage of “partial autonomy,” early AI systems are augmenting—not replacing—manual efforts and oversight by network engineers.
As higher levels of autonomy are pursued, the ultimate, long-term goal is autonomous network management with minimal human supervision.
Progressing toward this goal requires extensive testing and validation that will help stakeholders become confident AI can make decisions in complex, mixed-vendor, and unpredictable cross-domain environments, while still hitting performance requirements.
Continuously testing AI across interconnected domains is a critical linchpin that will help ensure systems can safely handle traffic, security, and performance needs at scale for various network segments. By shifting testing both left (early in development) and right (in live production), operators can ensure AI remains fit for purpose and adapts to changes in the network, data, or evolving AI models. This is especially important in telecom where networks are in a constant state of flux.
AI Ushers in a New Testing Era
AI isn’t just transforming telecom network operations. It is evolving how these networks are tested, making testing tools more efficient, intelligent, and adaptive to network needs, with features like:
Enhanced automation. AI is automating complex and repetitive testing tasks, such as those associated with setting up test scenarios, evaluating results, and making adjustments.
Synthetic test data and traffic generation. AI is drastically improving the ability to generate synthetic test data and traffic, including via digital twins. This is helping operators understand how networks would perform under various scenarios without the need for real-world data, which can often be sensitive or unavailable for certain use cases.
Enhanced intelligence. AI is being embedded into test and assurance tools to incorporate assistants that can manage configurations, analyze results, and make recommendations, conduct root cause analysis, and reference historical data to identify patterns and predict future issues.
Advanced fault detection and predictive testing. AI-powered testing tools can proactively detect anomalies that could result in faults, helping operators address them before they cause outages.
Continuous testing and closed-loop systems. AI is enabling continuous testing in lab and live environments versus legacy approaches that focused on isolated points in time. Real-time feedback loops help AI learn from the network and adjust on the fly for more consistent performance.
AI model testing. AI can be used in testing tools to prevent unwanted behavior like hallucinations or laziness, while verifying AI models are trained correctly.
Recently, Spirent worked with leading operators and smartphone manufacturers to utilize AI-powered testing solutions for enhanced video quality evaluation, using an advanced AI/ML model that provided a video quality measure more accurate than human-based evaluations. The result was 90% correlation with industry metrics like VMAF (Video Multimethod Assessment Fusion).
Additionally, AI/ML models were employed in emergency call failure test and analysis, enabling precise root cause analysis and rapid triage between device and network issues, significantly speeding up troubleshooting and resolution times.
In an industry first, Spirent’s AI workload emulation platform has proven capable of emulating high-density real-world AI traffic patterns at scale, to help identify issues that can lead to network congestion, higher latency and lower throughput and at a fraction of the cost and complexity of building physical xPU systems. The solution helps verify that Ethernet fabrics are equipped to manage the massive AI-driven traffic surges expected in the coming years.
AI is changing the world around us and Spirent is taking a proactive approach with new tools, capabilities, and perspectives to capitalize on the new opportunities that follow. Learn more about our work and the AI-powered path to automation by watching the webinar replay, which also features AT&T Network Analytics and Automation Vice President Raj Savoor. A companion report published by RCR Wireless News takes you further behind the scenes of AI network testing and assurance transformation.
