Spotlight: The AI answer to your broadband problems

The AI uprising

There’s no doubt that AI has become the technology of 2023. ChatGPT has practically become a household name, generative art services are delivering images of unbelievable quality, and AI is being rapidly integrated into everything from search engines to operating systems to everyday apps.

It might feel like AI has come out of nowhere to dominate the tech industry in 2023, but Richard Windsor says the key breakthrough in generative AI happened several years ago. “The most important innovation was that made by Google in 2017, which was the discovery of the transformer neural network architecture,” he said.

“The main difference it made was, instead of a neural network translating word by word, it allowed the neural network to consider the position of words in the sentence; it allowed the neural network to consider the whole sentence and also the context within which certain words were being used. And what that meant was a substantial improvement to the quality of translation, and that – in my opinion – is the genesis of GPT-based systems and generative AI.”

SamKnows’ data might soon be enhanced by AI

There are other factors that have supercharged the development of AI over the past year. “What has happened more recently is the advent of new silicon from Google that’s more capable, more silicon from Nvidia that’s much more capable, a ton of money from Microsoft, and a ton of data.”

All of these have combined to create “the phenomenon that most people know as ChatGPT,” said Windsor.

Putting AI to work

As the vast majority of people will have seen by now, ChatGPT and its contemporaries are capable of truly humanlike dialogue. And it’s not only the ability to converse in a way that’s difficult to tell apart from a human that is extraordinary, but the AI’s ability to converse ‘in character’. For example, ChatGPT can be told to write sonnets like Shakespeare, or write speeches in the style of Martin Luther King, or explain something like it’s talking to a seven-year-old – and it can do a convincing job of each.

That degree of sophistication and stylistic flexibility has naturally led to businesses – SamKnows among them – exploring how ChatGPT or its like could work for them. “We started to look at ChatGPT as soon as it was released and, as a technology company, we’re very interested in it,” said Alex Salter.

At the time, the company was on the lookout for a technical writer, and it began experimenting with feeding information to the AI to see what it could deliver. “What we found is that it was very good at taking the information that we’d give it and structuring it in a way that a technical writer would for a first draft,” said Alex.

ChatGPT was an instant phenomenon

“You still need to edit it, you still need to go through it, you still need to look for imperfections in the writing or information that’s presented as a fact that’s actually incorrect. But we realised very quickly that this is something that could have a profound impact on the business.”

Providing support

But it’s not technical writing where Alex thinks AI will come into its own, but in providing customer support. “So much has been written about AI and how powerful it is, and the impact that it can have on almost anything; but for AI to be powerful, it needs data,” said Alex.

“One of the issues with internet performance is that, until SamKnows, there hasn’t been a way of effectively measuring everything that could be causing problems with people’s internet connections. As we integrate AI, we see our role as one of collecting as much data as we can to make the AI as powerful as possible.”

“The role of SamKnows is to measure internet performance, and we’re always looking for insights because that’s what people really want from us,” Alex added. “AI gives us an opportunity to have another way of looking at data and then generating insights. We think the way that we can integrate that into our products is very straightforward. We can use it to help an individual at home who might be having a problem with their internet connection to identify quickly – looking at all the data that we collect – what that problem might be.”

It isn’t only the end consumers that this blend of SamKnows data and AI could help, but the human support teams, too. The majority of customer care teams currently follow scripts, hoping to eventually find a solution that resolves
a customer’s problem. “The opportunity that we have with AI is not to replace any of those people, but certainly to make those calls much faster,” said Alex.

The AI challenges

AI’s rapid rise has naturally raised concerns along the way. One big fear from both consumers and businesses is that personal data entered into AI systems might leak or be used to train future AI models, but Alex Salter is confident that risk can be managed. “I don’t think that AI on its own amplifies that risk,” he said. “Obviously, when you’re collecting a lot of data as we do, you must have a significant number of controls to ensure that the data that you’re collecting is protected and secure, and that it’s only being used for the purpose for which it’s being collected.”

By offering AI assistance to support staff or field engineers, “we’re not making the data available to anyone other than those who would have access to the data anyway,” said Alex. “We’re just making their lives faster and easier.”

There’s another big problem with AI, which Alex touched upon earlier – its tendency to make stuff up. Not only does AI often generate these so- called “hallucinations”, it does so with conviction, asserting with supreme confidence that something it has invented is fact. That could be a big issue in a situation where AI is dealing directly with customers. So can it be fixed?

“This is the $64,000 question,” said Richard Windsor. “If you solve that question, you’re basically well on your way to fixing the artificial general intelligence problem, and the super intelligence, and then the robots might actually come over the hill!”

AI hallucinations are a cross- industry problem

In other AI systems, such as self-driving vehicles, Richard Windsor says hallucinations are regarded as a “catastrophic failure” – for example, if a car starts accelerating wildly because it’s misread a speed limit sign. “I don’t know how it will be fixed. My view currently is that it won’t be until we find a new way of creating artificial intelligence systems that do not rely on the statistical neural networks that professor [Geoffrey] Hinton and his team created in the 1970s and the 1980s... that this problem will fundamentally be addressed.”