Interview: Phil Rumens and Mat Scalpello of West Berkshire Council talk of its early initiatives with the technology and plans for a ‘show and chat’ event
West Berkshire Council is set to provide an example of how some local authorities are feeling increasingly confident in exploring the use of AI.
On 14 December it is running a ‘show and chat’ session on what it has done and for the sharing of ideas. The event is for public sector employees only – sign-up details here – and has been organised with the possibility that it could lead to the formation of a specialist group for using the technology in local government.
Phil Rumens, West Berkshire’s digital services manager, says it was initially planned for the council’s own staff but has been expanded for wider sharing.
“I wanted to show the examples to colleagues and say ‘Where are the processes where you think this could fit into your working lives?’” he says. “Then I thought there was no harm in showing it to other councils. They might use a different stack of technology but it’s about sparking those ideas.
“It’s about putting together the concepts of the technology and possibilities of its use.”
Boost for thinking
The event comes around a year after the launch of ChatGPT made AI seem more accessible to a lot of people and provided a boost for thinking about new applications, with some in local government proving eager to use it to solve business problems. Rumens’ colleague Mat Scalpello, the council’s application development manager, says it has influenced their thinking.
“It is a step change above and beyond everything else that is out there. Maybe Open AI has caught everyone else on the hop a little. When we purchased a chatbot at beginning of lockdown it was state of the art, but it’s definitely not now and we thought about how we might do things differently using something based around GPT.”
He makes the point that other AI platforms build on large language models (LLMs) are available, such as Microsoft Copilot and Google Bard, which also have immense potential; and there is a possibility of the council developing an in-house platform for building applications, which would make it possible to ring fence sensitive information.
As an example of what can be done they demonstrate West Berkshire’s ChatGPT based tool for drafting job advertisements. It takes information from a form filled by a council officer and shapes it into a draft of an ad and a post for LinkedIn.
Rubens says that when they tested it with recruiting managers it received ratings of 7.5 and 8 out of 10 and that it has now been in use for over three months, providing major cost savings.
“We used the Chat GPT API to plug into the process,” he says. “We’re not using it as standalone but plugged into process for creating a job ad. We just slotted it into that process rather than doing something standalone.”
Text projects
Another project, currently in prototype stage, involved redacting text such as personal details of individuals from a document as part of the responses to freedom of information requests. There is also an initiative to plug an application into incoming emails to identify the type of enquiry, such as for help with council tax, extract information and place it into the relevant case management system.
The council has also used an AWS service for machine translation of documents for asylum seekers, reducing the cost per document from £160 to seven pence.
“Again it’s plugging AI into a process,” Rumens says. “The real value is when you start plugging it into other things.”
While talking up the potential of the technology they also emphasise the importance of human review of the outputs.
“You should be using it in the same way you use any tools,” Scalpello says. “The same way we have dead man’s handles on railway trains is you want some human to intervene at some point, because at the moment we don’t have a clear idea of what it is going to put out.”
He is working on an AI policy for the council, partly reflecting the need to privacy and security to the fore in developing new solutions, and to ensure that staff do not start doing their own thing in working with LLMs. Rumens speaks in terms of giving them secure methods and tools to ensure they do things properly, and Scalpello says there could be a high level policy for senior leaders and a more basic guide for other staff.
“The challenge is that it needs to be something everybody understands, but it’s not something that is easy to understand,” he says.
He also refers to a plan for a corporate dataset to which machine learning and AI could be applied to provide new insights.
Data rich, information poor
“We are data rich and information poor at the moment,” he says. “We’re awash in numbers but don’t do enough with them.
“One of the ways we can help with that is to plug it into things like ‘What if?’ models and being able to ask questions of the datasets we have. Why do we spend X in this quarter on a particular thing.”
As it stands the developments are taking place on an ad hoc basis combined with other work on digital services – “juggling and spinning plates at the same time” as Rumens puts it – and comes from working with service delivery teams to understand what users need. It has developed some momentum through putting the examples out there to get people thinking about how they might use AI – which could be based on platforms from different providers – and the show and chat is intended to encourage further ideas.
“For the ideas we’ve come up with, our colleagues will come up with loads more,” Rumens says. “The secret to having good ideas is to have lots of ideas.”
