Vodafone tests genAI waters, getting Microsoft tech to write code

Not many telcos sound as gung-ho about generative artificial intelligence (genAI) as Vodafone. Slides in a recent presentation it gave to analysts suggest the new-fangled technology is already gushing through all manner of Vodafone systems, from the chatbots used to communicate with customers to the critical software-writing part of the business. If software really is eating the world, as we're often told, the lines of code that control important infrastructure should presumably be treated something like the Coca-Cola recipe or last unpublished chapter in the Harry Potter saga – locked in a vault and guarded by fierce dogs. Letting genAI in on the secret sounds dangerous.

Vodafone has not gone this far, of course. It's mainly running trials and proofs of concept (PoCs) to figure out where genAI could productively and safely be used. But they are wide-ranging. As many as ten involve chatbots – whether the friendly named TOBi, Vodafone's customer-facing chatbot; ASKHR, its less imaginatively christened chatbot for employees; or systems deployed for supply-chain management and other activities. If genAI is fed with domain-specific knowledge, it could theoretically provide insightful answers to natural-language queries.

There are apparently another seven PoCs and trials targeting content generation across numerous Vodafone divisions, including VBPS (Vodafone Business product and solutions), network ops, commercial ops and Learning Portal, a Vodafone training resource. Vodafone notes in its presentation that members of its Internet of Things team used ChatGPT – genAI's most famous progeny – to work up a bid response based on earlier successful submissions. It's the approach some college students are probably taking to coursework.

And then there are the five PoCs and trials aimed at boosting developer productivity. Perhaps the biggest revelation in the slides is that Vodafone has now exposed some developers in its CyberHub team to GitHub Copilot, best thought of as a code-writing version of ChatGPT. Where the latter will write you an essay on the causes of the First World War, GitHub Copilot can churn out code when asked in natural language to solve a programming problem. It can also translate from Python to JavaScript and between various other programming languages.

Going deeper into the cloud

GitHub Copilot was born more than two years ago, before genAI had entered the mainstream lexicon, and fathered by Microsoft and OpenAI, the same lab behind ChatGPT. Heavier reliance on it would, by implication, make Vodafone much more dependent on a big public cloud at a time when operators are nervous about dependency on big public clouds. If nothing else, Vodafone is bound to worry that GitHub Copilot could expose company secrets to other telcos. It is unlikely to be the only operator exploring a relationship.

Vodafone's answer to that, conveyed at a press briefing last month, is to seal any confidential data into private containers in the Microsoft data centers where GitHub Copilot is trained. It is doing a similar thing with Vertex AI, a machine-learning platform built and hosted by Google. The alternative would be to use its own facilities, but Scott Petty, Vodafone's chief technology officer, has said there is currently no business case for investment in on-premises large language models.

During trials involving about 250 developers, Vodafone claims to have seen a productivity gain of between 30% and 45%. "The productivity increases are actually higher the more senior the developer is and lower the more junior the developer," said Petty. "The reason is that junior developers are typically writing very specific pieces of code. The more complex stuff is written by the more senior developers, and they spend a huge amount of their time just writing the commentary, embedding all the things that they'd forgotten from ten years ago in the development environment. It's really helpful for us because the more senior they are the more expensive."

Yikes. Watch out, senior developers. Petty dismissed the suggestion that genAI would claim jobs, but telcos initially tend to play down the impact of new technologies on the workforce. Headcount has tumbled across the industry, as Light Reading has been reporting for years, and technology is partly responsible. Back in 2019, when he was chief technology officer for Vodafone's UK business, Petty attributed many previous job cuts to the automation of network operations centers (NOCs) and said this would continue. "Five to seven years from now, there will be a very small number of people that run the NOC infrastructure," he said. "The way we run networks is fundamentally changing."

Not so fast, AI

But substituting genAI for skilled human coders, or even the best people in the less glamorous-sounding role of customer service assistant, would be as risky as bringing a Boston Dynamics robot on for Lionel Messi in the dying minutes of a World Cup final. Those robots look increasingly acrobatic, and might one day out-dribble the Argentinean superstar, but fans will be even less forgiving when a robot, rather than a human, misses a sitter by a country mile. Hallucinations, where genAI runs amok and churns out gibberish, still happen. Nobody wants to be the telco whose network blew up after the humans were taken off the mission.

Today, it is only in the relatively safe zone of deriving insights and data from customer calls that genAI seems to have advanced beyond the PoCs and trials stage at Vodafone. The operator has been feeding about 60,000 call transcripts a day into a summarization engine that subsequently spits out informed analysis. It can be used to improve customer care, still largely handled by humans.

GenAI is very obviously the buzzword of 2023 for a telecom industry that loves buzzwords. If history is any guide, it will do little to reinvigorate telcos and eventually make way for something else – even if that is just an updated version (recall virtual network functions being superseded by cloud-native network functions). Long before that happens, every telco vendor and its dog will be attaching the genAI label to as much as possible.