“The AI roadmap [for telecom] is still being created,” AI expert Jean-Christophe Laneri said.
Slow and steady will win the day in industrial-scale AI-enabled networks
AI Edge Minions are here, but carriers are still leery of the core Borg
A recent overnighter to Sweden to visit Ericsson HQ paid dividends for your correspondent when Jean-Christophe Laneri, the company’s AI maven, shared some key insights into where and how artificial intelligence (AI) is being deployed on carrier networks.
My main takeaway from the briefing: the pace of AI’s uptake in telecom networks will be determined as much by trust in vendors as by the state of the technology itself.
The AI twofer
AI will happen in two parts of the network, in two phases, Laneri told me.
Phase 1 is already happening, and being monetized by Ericsson and others, with AI agents being installed at the network edge to perform tasks like QoS, network slicing, and power management.
“These tasks are undertaken locally and defined by the need for time-sensitive operation, the need for microsecond speeds,” Laneri said.
Phase 2 will see AI installed in the core of networks, where it will be tasked with autonomously performing modifications to network operations on an industrial scale.
“Central AI deals with bigger, less time-sensitive but more complex problems,” said Laneri.
The AI agents at the edge of the network are basically Minions: loyal, eager to do their masters’ bidding. And in AI terms they have an only slightly higher level of intelligence than the titular characters in Despicable Me.
Core AI is a different beast altogether, more like the Borg from Star Trek: hyper-intelligent, thinking at the speed of light, emotionless. And, in telecom terms, potentially omnipotent.
Unsurprisingly, service providers and carriers are extremely hesitant to install this technology (androids may dream of electric sheep, but cloud architects go to sleep worrying that their core AI will turn out to be HAL-9000).
“Lots of operators have done AI, but very few have done it at scale,” said Laneri. “The engineering side of carriers needs trust. The technology has to be explainable. It cannot be a complete black box.”
Who you gonna call?
Trust, of course, is where Ericsson has an advantage over other companies which haven’t been delivering five 9’s solutions to their telecom customers for 147 years.
While there are numerous new companies that specialize in developing AI models for different industries, including telecom, I don't see them as viable options for our industry right now.
Take DataRobot. The company makes AI models (beep!) and claims that telecom is one of its target markets (boop!).
“At DataRobot, we move fast and reward hard work,” the company’s hilariously right on about us page tells us.
Rewarding hard work? Wow. Proper game-changer, that.
But it’s the other half of the statement that will be most bothersome to telcos wrangling a life-or-death transition to 21st Century cloud networking. Carriers actually don’t want their AI providers to move fast (or break things). Slow and steady is the cadence they are looking for.
New AI players don’t understand this. It’s not in their DNA. They call their companies things like Flipflop and Bippetty-Floop. Their Silicon Valley/SF company cultures purport to be all about the feels and inclusivity but in practice they seem to be exclusively run by arrogant self-important wankers.
As an arrogant self-important wanker, myself, I can tell you that these are literally the last people in the world you should entrust with creating an autonomous, self-managing core telecom network.
Ericsson has another advantage over the newcomers: it manages networks that serve more than 1 billion subscribers worldwide. That gives it an absolutely immense pool of quantitative data which it can feed into its core AI, which can then sharpen its virtual HB pencil to look through all of that information for patterns to model in its own telecom decision-making. The more real-world data, the more patterns, the better the decisions (probably... no one really knows yet).
So, when will Phase 2 – autonomous AI in the telecom core – happen? Laneri estimated five years for wide-scale adoption, which feels about right to me. Tier 1 service providers will move first, with Tier 2 and 3 coming later. Of course, some carriers will move faster than the curve, Laneri said, pointing to DoCoMo as one such trail blazer.
Do AI right, not first
It has been a busy week for AI. Nokia today cited AI at one of the three technologies that will have the most profound effect on communications networks through the next decade (you can read about its thoughts in its Technology Strategy 2030 manifesto/ thingumabob).
And President Biden unsteadily signed an executive order to strengthen safeguards around the technology (note: not being well-known for demonstrating a high level of natural intelligence, the U.S. government perhaps hopes it will have better luck with the artificial varietal).
Carriers are right to proceed with caution. Greenwashing, cloud-washing, and now AI-washing...We’re already in the latest inevitable phase of the tech hype spin cycle and carriers won’t be falling for that nonsense again, thank you very much. From the 90’s Internet bubble, to the implosion of optical networking in 2000, to the collapse of NFV (my bad), carriers have lived through the hype cycle of multiple comms technologies and know today’s AI market has all the hallmarks of hubris – that most peculiarly American tech trait.
“Carriers will use AI when and where it makes sense,” Ericsson’s Laneri told me as I was leaving Ericsson’s impressive new technology innovation center in Kista.
I would add: “and from whom” to his list of qualifiers. Core AI is an area where the Euro telecom incumbents – Ericsson and also Nokia – can and should lead our industry.
At least we’ll be in safe hands.
Do you have a cloud innovation that is alive and kicking? Submit for our Cloud Innovation Awards by November 3, 2023 here.