With AIOps growing in the telecom space, many are wondering just how much autonomy these tools will be given
To keep up with growing network complexity, telcos are applying automation techniques enabled by AI and ML in as many network operations as possible. As a result, questions around just how much autonomy these tools will be given have emerged. Experts seem to agree that it’s not entirely impossible for telcos to truly take their hands off the wheel one day, allowing AI to completely run their operations. However, those that spoke with RCR Wireless News about this question all noted just how far away this future remains.
“For now, using automation and AI/ML is about efficiency and quality,” Javier Antich Romaguera, senior director of product management at Selector AI, which provides operational intelligence of multi-cloud infrastructure and performance-sensitive managed services. Something like making it possible for a task that previously took someone an hour to perform to be done in only a few minutes, for instance. That’s the current priority; not removing people from the picture entirely. “For at least the foreseeable future, there are always going to be people at the wheel, but they will be assisted with workflows that will automate the execution of certain tasks… and by algorithms that will help them make better or faster decisions,” he continued.
For Per Kangru, technologist in the CTO Office at Viavi Solutions, the answer to the question was straight forward — yes, certainly it’s possible for AI to run a network. “But not today or tomorrow,” he said. “Could Viavi today deliver a zero-touch environment with an operator that has the right vision, right data sets and the willingness? Yes. The implementation of it would take some time and it’s probably going to be for a selected domain, but we could deliver it today. Will it be delivered in a larger part of the industry any time soon? No,” he said, citing the lack of inertia at most organizations.
“Some of the algorithms are very powerful,” Romaguera said of AI, “but they are also very opaque, and the explainability in the context of machine learning is very poor.” While he claims this isn’t inherently “bad,” it can be problematic if an algorithm powering an automated workflow makes a mistake. “If that mistake has some cost and no one can explain why that mistake happened, they cannot prevent it from happening in the future,” he said.
Further, while previous barriers around data visibility and accessibility have been reduced in the post-GenAI era, Rahul Kumar, who leads consulting for the global telco industry at IBM, said that algorithm hallucinations, where the models generally available to the public generate false information, continue to be one of the biggest concerns for IBM’s communications service provider (CSP) customers. They wonder how best to prevent hallucinations and any misguided output from reaching their employees or my customers.
“It boils down to what model you are using, what is the source data that the model has been trained on… These GenAI foundational models are trained on massive amounts of data… and there is no governance around that data, it’s just out there,” he said.
And so, the remaining concerns that need to be solved are what is the data set that’s behind these models, and how to tune a model to a real use case and purpose. For its part, IBM is focusing on the data side of the AI conversation just as much as the model side through its Watson X platform, which offers three components to its customers: a platform for development of AI models; the data fabric to go along with the AI; and data governance, which provides tools to monitor and manage models and data for drift, biased and ethics, all of which Kumar claims will work to minimize hallucinations.
And lastly, in a conversation at Google Cloud Next ’23, which took place a few weeks ago, the company’s VP/GM Sachin Gupta told RCR Wireless News that overtime, telcos will leave more and more operations fully in the hands of ML and AI as comfort levels in this technology increases. For example, they will likely soon use ML and AI to automatically detect new threats without human oversight. “But what doesn’t happen, necessarily, is the automatic filtering and firewall rules after the threat is detected,” he continued, suggesting that while they may accept AI telling them there is a problem, it will take much longer for telcos to trust AI to make the decision about how to solve the problem.
Keep an eye out for the complete report about this topic coming soon, called Bringing AIOps to telecom: When will operators take their hands off the wheel?
