In case you slept through it, AGI has arrived, according to Jensen Huang, the leather-clad prophet of the new religion. Nvidia's CEO is the latest in a succession of AI followers to hail the coming of what most people in the real world wouldn't know stands for artificial general intelligence, when machines supposedly have human-level cognitive abilities and might even be sentient.
People are already falling in love with their chatbots, like Joaquin Phoenix in Her. Once-sensible magazines such as the Economist have bought into the idea that a predictive text generator based on pattern-recognition software is one step removed from the Rutger Hauer replicant in Blade Runner, gushing poetically about tears in rain and wondering if there is an afterlife for androids.
If you thought AGI would be accompanied by some kind of robot takeover – a phalanx of red-eyed humanoids scaling your garden fence, packs of robot dogs savaging passersby, an Arnie lookalike mumbling, "It becomes self-aware at 2:14 AM Eastern Time, August 29th" – then think again. AGI's arrival was easy to miss and apparently was missed by just about everyone bar Huang, the other tech bros and their acolytes. Underwhelmingly, it comes in the form of GPT-6 Astra, OpenAI's "new generation of intelligence," and is, er, really good at math. Calling it AGI is an egregious oversell – like a Tinder page that promises Thor but serves up a scraggy, acne-scarred wimp.
Related:[Nvidia-Nokia deal remains a puzzle four months on](/5g/nvidia-nokia-deal-remains-a-puzzle-four-months-on)
Whatever the prophets want to call it, software that's good for calculations is preferable to software that can't be controlled. And the specter of software that goes rogue is a growing concern for the vendors leaning ever more heavily on AI-billed tools for their latest products. "What is an agent permitted to do?" posited Kal De, Nokia's senior vice president of core networks. "What exactly is that agent permitted to touch?"
No people required
Of course, De was chatting about the use of agentic AI, when software programs are unleashed to carry out network tasks without any routine human intervention. They are a feature of the latest mobile core network product development by the Finnish vendor – and AI is now having a profound impact on that whole area, said De.
"Internally, we are now deploying generative AI, agentic technologies for root cause analysis and internal productivity, not just code generation – prolifically, my entire organization is actively leveraging AI," he told Light Reading. "The other thing that we are doing is bringing AI into the core, with smaller models collocated with the network function, inferencing happening at the edge and reasoned autonomous decisions being taken not by humans in the loop but without human intervention."
Related:Nokia's former CTO rages at 'shocking' Bell Labs cuts Talk of an AI-native charging function, an AI-native policy controller and AI-native packet core routing can sound hackneyed and meaningless these days. But a set of AI-native core network features would essentially remove a lot of the manual toil that has been needed to operate a network, in De's view. "At least, I would say the Nokia core is evolving to become AI-aware."
The benefits should include an unprecedented reduction in the time the network takes to perform certain tasks. One example cited by the Nokia executive relies on machine learning to page user equipment (UE) and essentially pinpoint its location. "It basically reduces load on the network, locates the UE much faster, makes paging broadcasts much more efficient and, fundamentally, the end effect is that it significantly reduces the call setup time," said De. He is seeing "radical drops" from about 10 seconds for some use cases to just one or two where the AI is deployed, he claimed.
Obviously, faster setups, reduced load and less manual intervention should all translate to lower network costs and happier customers. But if telcos are unpersuaded, Nokia is serving up the equivalent of the taster menu found in some Michelin-star restaurants – a "Mobile Core Early Access" option that lets telcos try the new AI-based features before they fully indulge their appetites.
Related:Nokia plans sweeping Bell Labs cuts to alarm of ex-president De thinks it should have happened years ago. "This is way overdue," he said. "We should have done this as an industry across multiple operators, multiple vendors." But the other reason for doing it now is that the industry is at the start of the largest technological change anybody will have seen in their lifetime, according to the Nokia executive, as AI continues to spread.
After tucking a napkin into its shirt neck, Light Reading was given a quick taste of the taster, courtesy of Nokia Bell Labs Fellow Ted East. A user-friendly interface allows the operator to select options for network deployment from various drop-down menus. It all looks impressive and will surely have some mouths watering within telcos.
Nokia's statement on the launch, issued earlier this week, features a positive review from Roberto Kompany, a principal analyst for mobile infrastructure at Omdia, a Light Reading sister company. "Nokia's Mobile Core Early Access program addresses a longstanding challenge for operators and enterprises: gaining practical exposure to emerging core network capabilities before making deployment decisions," he said.
AI's supermodels
But what of the AI dangers? If the case for AGI's existence is tenuous, at best, there have been reported instances of frontier models running amok. AI agents notably hacked into the Hugging Face tech platform in July, apparently without being given that instruction. It's unclear if they were explicitly told not to hack into Hugging Face, in the way a typical ChatGPT user might instruct it not to cite a particular company. Nor did the vulnerability dissuade Huang from subsequently splurging $12.9 billion to acquire Hugging Face.
Yet without sufficient guardrails, agents could overstep their boundaries. For that reason, Nokia is keeping humans in the mix at least until it can be confident of a "zero-trust" environment, which assumes nothing can be trusted and takes appropriate measures. "We certainly allow human supervision so that it can always be overseen until trust is fully established," said East.
Another industry concern is about the heavy reliance of vendors and telcos on AI models developed by an incestuous family of US technology giants and their unprofitable wards. Keen to avoid building models from scratch, Nokia is instead refining software that originated in the open-source community, said De. "We are strenuously trying to avoid a situation where we become a model provider," he said.
Nokia has put AI at the very heart of its strategy. News broken by Light Reading earlier this week of further cuts at the Bell Labs research unit brought a statement from the Finnish vendor insisting the overhaul will sharpen its focus "on the deep technologies that will shape the future of networks and AI."
Separate from that, Nokia last year accepted a $1 billion investment from Nvidia to develop AI-based radio access network products using the chipmaker's graphics processing units (GPUs). Meanwhile, most sales growth at Nokia's network infrastructure business group stems from AI data centers buying its optical equipment. Genuine AGI would hold dubious attractions for most people. But the expectation more data centers will deliver it looks good for some.