AI and Future of Telco 2 of 3: AI is fundamental to 6G—but like all technologies that drive real value in telecom, nothing will really happen here until the standards come on stream
In the first part of this short three-part Summer 2023 look at AI and Future of Telco, we looked at the importance of standards for both 6G and AI’s success at the political level.
Now, let’s look in more depth at why that is true at a more technical one. Our departure point must be that even on the most optimistic timelines, 6G communication technology is not expected to be a real thing in the market until 2030–2040. So, why are we talking about a cellular network that is yet to be released?
Because we all really want to make 6G work. And to get there, we need to talk about Artificial Intelligence… but fear not—this is not another ChatGPT discussion and is instead about AI for grown-ups trying to achieve serious, game-changing industrial and business change.
Indeed, the ITU has specifically said no AI, no 6G (‘The introduction of ML techniques has become a necessity due to the complexity of future networks in order to supply the intended Future services’).
Let’s back up a second and define some terms. A 6G network will be a cellular network that operates in so-far unexplored parts of the radio spectrum. The idea: enable high-speed, low-latency communication at speeds many times faster than 5G—it already has set goals for itself such as a 1 Tbps peak data rate, 1 ms end-to-end latency, and up to 20yr battery life for this next generation of communication networks, and much more along this axis of ambition.
AI in 6G will be integral to specific applications and vertical services, and also to the network itself for everything like Service Problem Management, policy-based management, and orchestration. But 6G’s going to be about more than just doing what 5G can do better and faster for both business and consumer smartphone and mobile network users. It’ll also have its own features for the easy snapping-together of next-generation wireless communication networks for linked devices, allowing us to finally (we hope) see all the promises of smart cities, driverless cars, IoT virtual and augmented reality and other innovations delivered.
‘Huge potential’ in using AI across many parts of the 6G spec
That’s quite a spec, so no surprise that researchers are already completely convinced a lot of this just can’t happen without significant amounts of the heavy lifting being done by machine learning and AI. As Nokia Bell Labs among others has noted, to achieve these amazing faster rates and lower latency performance gains, a very, very efficient network will be required—one that can dynamically allocate resources, change traffic flow, and process signals in an interference-rich environment.
And that can only be delivered by extensive, planned use of machine learning and AI. AI will therefore be central in boosting 6G in terms of high QoS, high data rates, ultra-reliable low-latency communication (URLLC), and all its other highly desirable features. (It’s even been suggested that a new AI/ML-enhanced physical layer might also enhance the energy efficiency of 6G networks by achieving as much as a 50% reduction in transmit power over 5G for the same bandwidth and data rate.)
So, by optimising networks and designing new waveforms, we can see huge potential in using AI across many parts of 6G, mapping AI applications to the standard Open Systems Interconnection network layers model, including physical, data link, network, and applications appear to be early research targets. But—and it’s a very important but—this can only really move the needle by standards.
Let’s see how.
6G and AI: Do they go together like a horse and carriage?
Every operator and vendor have plans to introduce automation. But to do that, as we’ve seen, due to the complexity and scale of building out a sixth generation network, it must be backed by AI—which means AI is not just a tool to use, but is integral to the network, a fact that’s specifically acknowledged by the first wave of specifications.
But when it comes to AI, as we’ve just seen with Generative AI it’s very easy to spin off into orbit, especially with some of the deeper application ideas like use of cognitive AI techniques that emulate the decision-making processes of the human mind like Deep Reinforcement Learning or Network AI/MF Empowerment Method.
In fact, we will probably need a lot of advanced techniques to be applied for automating operator workflows. But we have to align all these great ideas and promising lab research with wider needs like Network 2030, which is all about seeking early possible answers on hard questions on the best network architecture and enabling mechanisms suitable for novel use cases like holographic type communications, extremely fast response in critical situations and high-precision communication demands of emerging market verticals.
We’re already seeing some real work here out of ITU Network 2030, like the Network Logical Architectural Integration of multiple AI/ML methods element of ITU FG-NET2030-Arch, especially (as we’ve seen) around the use of AI/ML in the potential autonomous/self-management and orchestration of a 6G network. But this is where we do have to acknowledge the ongoing ChatGPT hubbub, because while AI is expected to deliver much, it is also a highly problematic space with all kinds of messy ethical / moral and privacy / protection issues.
And quite rightly—tools of this power need to be properly introduced and controlled. AI is thus the focus of a wide set of standardisation efforts, and on multiple fronts: there’s some real progress by IEEE and ISO/ANSI with SC42; the EU already has legislation before policymakers; the US has some proposals for an AI ‘Bill of Rights’, and other initiatives are ongoing in other parts of the world (e.g. Japan) and the UN says regulation of AI is essential in a number of areas like surveillance and disinformation.
In almost all these initiatives, trustworthiness is a key principle. Whether or not we will arrive in 2030 with all this sorted out, or be faced with as bad an unregulated mess with AI as we have with the probably broken beyond repair state of the Internet, remains to be seen.
The nightmare scenario of a 6G-AI conspiracy theory getting out into the wild
As we discussed in the first part of this short series, the good news is that AI standardisation is inevitable and that in our sector in particular, standardisation of AI in the 6G context must happen, because telecoms is inherently standardised.
After all, 5G is the result of a single standardisation effort, and 6G probably won’t be as simple—but it’s a process we must support. To make what is truly an absurd but inescapable prediction, in a world where people see AI as an enemy, the ‘COVID was caused by 5G’ conspiracy could easily mutate into a much worse piece of Luddite backlash against 6G, stifling it at birth.
That’s a scenario none of us wants, believe you me. The time to start thinking about how to head that off by making AI safe and accepted via standards and openness it is now, as the IMT for 6G is likely to result in multiple approaches to standards which will be led by a number of stakeholders, including governments, industry verticals, new groups and more (e.g., we would certainly expect something like a 6G version of the 3rd Generation Partnership Project 2 happen at the very least, for example).
And AI must be brought into this process, as 6G simply cannot deliver all its potential without it. Participants—all of us—need to recognise that the kind of AI we need to make 6G work needs to be regulated and standardised, and the sooner the better.
Bottom line: the global communications industry must recognise this in good time and act to drive standards for telco AI processes which can then quickly input into 6G standards, and at the root.
In the next and concluding part of this three-part AI and Future of Telco series, we will examine the last and perhaps most critical part of all this--automation.