Vodafone Group CTO Scott Petty pointed to positive progress with generative AI (genAI) and believes the technology could transform the economic model of the internet, but warned those with inflated expectations today were in for an unpleasant surprise.

The company has been developing genAI applications across a number of areas including to assist employees, simplify business processes and for its TOBi chatbot.

It has used what it describes as a partnership approach, working with hyperscalers and producers of large language models (LLM) so it can focus its efforts on fine tuning models and capabilities.  

During a press briefing Petty (pictured) said it had dismissed the option of using standard publicly available LLMs on data privacy grounds, while building infrastructure from scratch was deemed to take so long the hardware would be outdated by service launch. This is due in part to the current speed of GPU evolution.  

Although positive on the outlook of genAI at Vodafone the executive branded it generally as the “most overhyped technology for many, many years in our industry”.

“Hopefully, we’re reaching the peak of those inflated expectations because we’re about to drop into a trough of disillusionment as companies really struggle to go from POCs [proofs of concept] around genAI and create real scaled benefits across large organisations,” he added.

Accuracy
Petty noted genAI has been “actually quite difficult to get accurate” with the quality of the data available and its management key.

Among the measures put in place by Vodafone are guardrails for its use, formation of an AI council and tools to ensure privacy of data in-line with local regulations. Human experts are used to check for accuracy and deal with sensitive areas high on its internal “risk grid”.

Petty believes its framework and policies have “enabled us to avoid many of the impacts of hallucinations” and “some of the brand-impacting genAI experiences” suffered by those who “took humans out of the loop too quickly and made [applications] directly available to end users and customers”.

He noted the “real value” of genAI “comes from rewiring your business processes, being able to take functions in your organisation” including within improving customer service, performance and cost.

Looking forward the executive claimed the use of genAI applications using small language models on mobile devices could transform the economic model of the internet and put network performance and latency under the microscope.   

Fundamental changes
He added capabilities in the coming years could herald a “fundamental change in the internet and the way we use it to serve customers and as a channel to market”.

“The economic model of the internet today is really simple,” Petty explained. “You go to a search box, you type something in and depending on who paid the most money or did the best search engine optimisation you’re presented with a set of results that dictate what you do”.

“In a new world where genAI is running on my device and my agent on that device is now my primary interface with the internet … that creates a totally different digital channel”. He added in this scenario brands would need to engage with the LLM or bot on the device for marketing.

He also indicated applications engaging though images, video or voice rather than text would place a new focus on latency which was less of an issue in the text search era.  

Network performance in this on-device genAI era is anticipated by Petty to be impacted by small language models interfacing with LLMs in the cloud, meaning “the latency performance between my handset and that cloud is going to be really important to whether I think that’s a great experience or not”.