While people widely discuss “AI” as the cure to most of the problems in the telecoms universe, not all AI is created equal.
While much conversation today is essentially about how we can use AI to do the things we already do better, cheaper or faster, at 6G Forge we heard about two approaches that may well have a profound impact on how the telecoms environment operates. And they’re likely not the ones you’d initially be thinking of. Not least, getting off the terrifying and expensive “G” upgrade cycle.
Mediocre performance? Good show!
The telecoms industry has worked wonders over the past few decades in how we transmit signals. We’ve become very good at sending signals – maybe not flawlessly, but good enough to identify where there are flaws and re-transmitting the data. And doing so faster and faster.
Is there another way to do this, though? Emilio Calvanese Strinati and his colleagues at CEA-LETI believe so.
For Emilio, the problem is with the fact that we are transmitting signals “blindly”, with no awareness of the content, purpose or end-goal of the communication. “It’s like pressing letters randomly on a keyboard and trying to be absolutely precise,” he observed.
It worked fairly well in the past, but with the network densification that 5G required, we are reaching a point where interference starts to eat into the capacity benefits of doing so.
“So we’re pushing the limit of being ultra-precise, while on the other hand, we now have intelligent agents who can do something different.”
By using semantic AI agents, Emilio argued, we could change the system so that, instead of sending every bit of data, we could “share only what cannot be deduced or inferred by an AI.”
By adding intelligence on the semantics into the process, it’s possible to change some of the fundamentals in the system. Instead of absolutely requiring perfect text such as this, we cluod aslo mnagae wtih txet like tihs.
“This is a text which is horrible noise, but you can somehow understand it because you’re an intelligent, reasoning being. So the point is that, by using intelligence at the transmit and receive, you can relax your requirements. You don’t need to have an ultra-good signal because, in the end, we can correct it. This opens up a lot of opportunities,” he said.
Does that sound a bit like hype? Consider how much of telecoms is built around incredibly exacting specifications and demands for performance. What happens if now, rather than perfection, “you can ask the network to be average good, and AI can do the rest,” as Emilio observed.
Some of the answer to that Emilio gave directly:
“It changes the way we deal with imperfections in the hardware – for example, power amplifiers – we have different ways to compensate for the issues behind it. You can have different kinds of waveforms, different kinds of power amplifiers, different kinds of beamforming algorithms, different kinds of small cell locations.”
All of that implies being able to reduce cost, both in electronic parts and in working with prime locations for cellular deployments, because the performance requirements can be relaxed. However, there’s another benefit as far as opening up capacity goes; and that lies in the nature of semantic communication.
AI Semantic Antics
With AI-based semantic communication between a transmitter and receiver, the amount of data that needs to be transmitted can reduce over time in a way which we have never seen before in technology, but which we understand well from personal relations – simply, that two entities can get to know one another.
Emilio gave the example of a couple sharing a piece of bread. On their first date they introduce themselves, ask politely for food with a properly described request, say please and thank-you.
After a few months, they don’t need introductions, and one can say to the other, “Do you have some bread?” and have it understood that, if the answer is yes, they would like some.
Over the years, the couple get to know one another well enough that they don’t even need to say anything for one to know the other will be wanting bread. By gaining a better understanding of context – the typical behaviour of the other, their identity and how they communicate – only novel elements need to be communicated.
Over time, Emilio noted, the vast majority of actual data can be stripped out and still convey the same amount of information, saving several times more energy than the algorithms use and, at a stroke, opening up several times more capacity from the existing network.
“So using AI is not just to optimize the network, but to change the way information can be shared,” he concluded.
What Do We Want With “Intent”?
While this clearly offers some useful opportunities, Adlen Ksentini, Professor at EURECOM, wants to use AI in some very different but related ways. While semantic communication could change how we build networks, intent-based communication could change how we interface with them.
“We wanted to open 5G to verticals, and we saw that verticals aren’t experts in our technology,” Adlen pointed out. “They just want to deploy a service and make it work.”
Intent-based networking isn’t a new concept. Adlen highlighted that it was explored by 3GPP as early as 2015 as a way to abstract the complexity of the 5G system.
“Intent comes back to the forefront because in 6G there are a lot of components that are interacting. We are speaking about sensing, speaking about IoT, speaking about radio, cloud… to a level that requests a new technique in order to simplify this.”
And that solution, Adlen suggested, is using AI/ML to understand the outcomes that are desired and then go execute on it.
“In intent we have a customer – could be an OSS/BSS, could be a person – that defines an intent and then an agent that acts to fulfil it,” he explained.
While in earlier discussions intent might have been expressed through programming in JSON or YAML, today we can use LLMs to express what we need in human language. Adlen gave an example of what that ‘intent’ might look like.
“I need a network composed of three XR applications… Each application requires 4 vCPU and 2 Gigabytes of memory… in the 5G network located in Nice area and tolerating a latency of 5ms.”
Meaning, Not Language
Translating that in ways a network can understand isn’t simple. The LLM needs to break that down into what is required of different parts of the network, and how to deploy it; and then negotiate with the network to understand whether it’s possible to fulfil what is required; and then to activate it and assure it, managing the lifecycle and the service levels.
“LLMs can fit there for natural language processing; to allow LLMs to understand the intent. Also for the negotiation, to use it as a chatbot. If the demand can’t be met, the user needs to know.”
Not only that, but the system can be made to troubleshoot in a closed loop for quality assurance. The industry has had predictive algorithms for years now which can identify potential problems and pre-empt them. As part of managing an SLA, this ties in with being able to deliver on the intent of the user for their service.
Adlen was able to share a demonstration of how they are delivering this at EURECOM, in a way which makes programming or altering network performance as simple as enterprise customers might want.
The next step, according to him, is to improve the translation back from the network to the person for ‘human-readable management’ of what’s happening. In other words, giving a customer more simple visibility into what’s happening and why.
Making Telecoms Better?
It’s almost a trope today that “AI is going to change everything” and most of the time what we see is more a case of trying to automate what people already do. It was refreshing to see work going on that could lead to fundamental changes, providing only that this work comes to commercialisation.
Any and all of these seem to be implied by semantic and intent-based communications, individually or together:
- Instead of chasing spectrum and new capacity in a next generation, we could chase better semantic algorithms. We get a similar result without the license auctions and infrastructure rollout.
- We could finally deliver enterprises of all sizes the simplicity of experience that they need to make telecoms deployments worthwhile.
- We could use ‘good-enough’ equipment and handsets to deliver top-quality services, reducing the costs to purchase and run. Cash-strapped telcos could deliver an upgrade that makes an old network function like a new one (great news for cash-strapped tier 2 or 3 players).
…and so on, with second and third order effects on the ecosystem and customers.
If these resonate with you, you can watch the talks from Emilio and Adlen, as well as more than a dozen other talks and panels, here. And if you’d like to be kept in the loop as we explore how to make the telecoms industry better, consider signing up for our InFlux newsletter and comms.
