With all the excitement swirling around artificial intelligence (AI) as of late, you might be wondering: is there anything AI can't do? It's a question that's unlikely to be answered anytime soon, especially as everyone hops aboard the hype train in search of the AI miracle cure.

But its not just the hyperscalers and cloud providers dancing at the precipice of the technological singularity — telecommunications vendors are betting big on AI to solve some of their toughest problems.

Ericsson and Nokia, two of the biggest telecom vendors, are no exception. The two companies have made numerous investments in AI in an effort to help network operators reduce operational costs and optimize their networks for faster, more reliable communications.

According to a report by STL Partners, which Nokia commissioned in 2019, AI has a very real opportunity to save operators money and improve customer satisfaction. While the report found less than half of telecoms surveyed had invested in AI by mid-2019, STL reported that "telcos are uniquely well positioned to take advantage of AI technology, largely because they are already used to dealing with huge volumes of data on which AI and machine learning [ML] rely."

The report also highlighted five key factors driving AI adoption by telcos. These included network optimization, improved sales and marketing, enhanced customer experiences, monetization of consumer data, and new services.

According to Paul Mclachlan who heads up Ericsson's data science and AI team, AI will be essential to enabling low-latency 5G networks. "5G is all about latency and speed," he said. "As the edge network grows, we need to optimize locations as well as traffic routing. This also requires us to continue to optimize networks using AI such that each packet takes the shortest path to the edge cloud."

And to address those concerns Ericsson added Network Intelligence to its AI portfolio earlier this month. The service uses machine learning to identify and resolve disruptions or outages before they impact network performance.

However, there are numerous problems still facing telecoms as they seek to adopt AI technologies. Mclachlan said one of the biggest challenges facing telcos today is the lack of AI models tailored to their specific use cases. As a result, Ericsson has had to task its AI team with developing these models from scratch.

This is where telcos are facing another problem. To build these AI and ML models, they need high-quality data that is not siloed or fragmented.

STL's report showed most telcos have struggled with inconsistent and fragmented data as a result of poor historic data capture. "At the most basic level, telcos have not always been capturing the data they now realize they need to properly train their ML models," the report reads.

And while STL notes that data lakes may help to eliminate these silos, allowing ML models to connect the dots between otherwise unconnected events or phenomena, they aren't without their own challenges. "Data lakes present their own unique challenges, especially because the data sets they contain are not necessarily well structured," the report reads.

There are also security concerns associated with data lakes since a large amount of data is accessible from one location. STL says without proper governance a data lake can not only become a liability but unwieldy as well.

The other problem is how and where to process this data. Here, latency once again becomes an issue, Mclachlan said. "We also need to be mindful that many 5G use cases are predicated on low latency, so we have the challenge of building models that compute quickly," he explained. And by low latency, Mclachlan means 1 millisecond from one end of the connection to the other.

Today, most AI use cases are designed to run in the cloud without much consideration given to speed or latency. For this reason, Mclachlan says telcos will need to build a mixture of AI models including some that prioritize latency.

Data and latency reach a crossroads when talking about IoT devices, which are anticipated to stream significant amounts of data in real-time over 5G networks. Here, there's an opportunity to start making sense of what is relevant data and what is noise by processing that data at the edge of the network.

This is where several chipmakers and designers like Intel, Nvidia, and Arm have been focusing their AI efforts.

Earlier this month, Arm announced two chips that it said would eliminate the need for cloud-based AI by delivering ML capabilities right on the device.

And while Mclachlan and Ericsson firmly believe AI workloads will be split between the cloud and the edge, Arm is focusing its attention on the edge.

During a keynote at Arm TechCon in October, Arm CEO Simon Segars said while it’s possible to offload AI workloads to the cloud, it isn’t very efficient and doesn't scale well. Instead, he said the next generation of IoT devices will handle AI and ML workloads locally.

Arm isn't the only one thinking about the edge. In late 2019, Nvidia announced a massive deal to deploy its EGX edge supercomputing platform on China Mobile's network. This platform, announced in October alongside a partnership with Ericsson, is powered by the Cuda Tensor Core graphics processing unit (GPU) and can process 15 teraflops of data per second.

China Mobile is also using Nvidia’s technology in its 5G network, which it says equips first responders and doctors with AI-powered tools that can remotely diagnose patients and transmit footage from drones at the scene of an unfolding disaster.

However, Nvidia isn't just looking at AI at the edge. The chipmaker announced a collaboration with Microsoft Azure in November to put its high-end V100 Tensor core GPUS in the cloud.

However, AI remains uncharted territory even for established chipmakers like Intel.

In Early February, Intel announced it was abandoning its Nervana neural networking processors, which it developed in collaboration with Facebook, in favor of Habana Lab's AI chips, which Intel acquired for $2 billion in December 2019.

According to STL, most telecoms have only just begun to scratch the surface of what AI has to offer.

For now, it anticipates the line between policy-based automation and AI will continue to blur with ML beginning to inform automation and eventually evolve into fully autonomous systems.