-
CSPs should welcome autonomous AI – and fear it
-
We are in only the first of three stages of AI development
-
Telecom culture is causing it to trail enterprises in deploying AI
I read an interesting research report about telco AI on Friday. “Building AI into the Telco of the Future” is authored by the excellent Chris Smith for MEF and provides a good overview of the opportunity and risks of artificial intelligence (AI) to communications service providers (CSPs).
That said, I disagree with one of the report’s main findings, that:
“The superficial relevance and suitability of AI generated output should always be questioned, and the last word left to the human element.”
It’s true that today’s AI tech is not ready to fly solo on telco networks, but that will change. And given the current levels of investment — driven in large part by Nvidia’s extraordinary success – and the hyper-accelerating pace of AI development, it’s likely we will see the first telco networks run autonomously by AI without human involvement before the end of this decade.
CSPs should both welcome this change — and fear it.
Autonomous AI is inevitable for two reasons. First, the volume of telco data and the speed at which it must be processed demand a pace of decision-making that is far beyond human capabilities; inserting human-incepted checks and balances into 21st century telco network processes is like chaining a jet plane to a pyramid.
Second, telcos need a business model silver bullet to secure their future in the face of competition from both hyperscalers, and also enterprises (Et tu, Brute?), which are using cloud infrastructure to end run CSPs. With its ability to reduce costs, increase operational efficiency and — at some point — manifest new revenue streams, AI will be the difference between the telco’s survival and extinction.
A guide to dummy AI
Before AI technology can run the telco show, it needs to go through some significant upgrades. Despite the name, current AI models are actually too stupid to undertake the autonomous telco management role.
Artificial intelligence technology falls into three classes corresponding to different levels of intelligence. Artificial narrow intelligence (ANI) systems are less intelligent than humans but do simple tasks really well. Think of them as loyal virtual servants – nice but dim.
The AI now being installed in telco networks is ANI-class, and typically operates at the device level. There, it can manage and improve functions like power management and error correction, but only according to pre-set parameters which are determined by human operators. It’s a low-risk, high-benefit approach to AI but, honestly, not very exciting. And no one is suggesting that ANIs should be put in charge of any strategic telco functions.
The second AI class, artificial general intelligence (AGI), is where things get more interesting. AGI’s are about as intelligent as humans and can think in abstract terms, undertake reasoning, and suggest solutions to problems — but at industrial scale and much, much faster than we can.
There’s quite a feisty debate among AI nerds about whether anyone has developed AGI yet (they almost certainly haven’t, even in controlled lab environs). When they do, telcos will start to face a lot more pressure over AI deployment. AGIs absolutely have the ability to operate independently, without human overseers, including making their own moves adds and changes to network operations and policies.
But will telcos let them? CSPs are a risk-averse bunch and extremely wary of AI, to the point that telecom now trails enterprises and vertical industries like transportation and mining in deploying even the more rudimentary (slow-witted) AI agents.
At the same time, a technology that can revolutionize operational efficiency and profitability is exactly what they need as soon as possible, which means they have little choice but to move forward with it. And rather than focusing exclusively on its risk profile, they might want to consider that an AGI might well have headed off an event like AT&T’s horrendous outage last week, which appears to have been the result of hooman error.
There is also something selectively luddite about the way tier 1 telco execs approach AI, with the same people counseling a slow-and-steady approach but being perfectly happy to put their own lives in the virtual hands of, say, Tesla’s AI autopilot.
Ericsson is one company that is looking to increase telcos trust level in AI. Its XAI cognitive code allows cloud network architects to understand the reasoning behind the decisions made by AI agents on telco networks, rather than having to take them on faith.
Ultimately, the benefits achieved by the first CSPs to deploy autonomous AGI are likely to be so mind-blowingly amazing that the rest of the lot will have literally no choice but to jump in so they might as well get used to the idea now.
Cogito ergo sum
Beyond ANI and AGI, there’s a third category of AI, called advanced superintelligence (ASI). This is a form of AI that will exceed human intelligence and be better than us at just about everything — including decision-making, problem-solving, and even designing new technologies.
There is no clear consensus on when ASI will arrive (and possibly not for decades), but when it does, it will bring capabilities that people incorrectly attribute to AI today. It also will bring with it a level of risk that transcends telco business concerns, to put it mildly.
Assuming ASIs follow Descartes’ “I think therefore I am” first principle, it is logical that they will possess a survival instinct. In a worst-case scenario, ASIs embedded throughout industrial society could infer, based on a wealth of quantitative evidence, that humankind is the biggest threat to the Earth’s welfare and their own survival.
A logical remedy — from the AIs’ point of view — would be to “shut down” the human race. That would be surprisingly straightforward…release the bioweapons and chemical weapons from the not-so-secret government laboratories, meltdown the world’s 413 nuclear reactors and shut down the electrical grid (cutting the power alone would cause the death of up to 90% of the population in North America, according to Senate Committee testimony from 2015).
So yes, there is AI risk, but for telcos in the rest of the decade, it’s all about opportunity.
More on that once I leave Mobile World Congress, I hope.
Follow our coverage of Mobile World Congress 2024 in our dedicated channel here.