Generative AI and Large Language Models

Welcome to the third instalment of the AI for Local Government blog.? In this blog, we’re going to take a look at Generative AI and Large Language models, exploring some of the use cases we are seeing in Local Government.

Most of you reading this will by now have played with Chat GPT or the new Bing Search to ask a question and get an answer.? One of my favourite early use cases of this was to ask Chat GPT to create a brief story for my children’s bed time, where they would ask something like ‘write me a story where I would go back in time and play with a dinosaur’ and Chat GPT would create me a short story to read them.? Other people would ask it to create a poem or song, or to help plan holiday ideas.?

Some of you may also have used something like Dall-e, where you can ask a generative AI engine to create a picture for you, with some description which will then get an output.? As an example, for my Festive Email to my team last year, I asked Bing search for enterprise, which uses Dall-e to create an image with some specific prompts to get the representation in my team appear in the image, and it created this fantastic image:

?

?

All of these are examples of Generative AI, which is a branch of AI that focuses on creating new content from existing data sets. ?Generative AI can produce text, images, audio, video and even help with both Coding and creating Low Code applications based on the prompts provided to it.?

Much of the excitement at the moment around Generative AI, especially in Local Authorities come from solutions built on Large Language Models (LLMs).? LLMs are a type of generative AI that can generate natural language text. They are trained on massive amounts of text data, such as books, articles, websites, social media posts, and learn to predict the next word or sentence given a prompt.

The opportunities these capabilities present are significant for Local Authorities.? When we start discussing the capabilities of a LLM, this is where we see ideas spawn to life in a Local Authority.? Showing the ability to generate coherent documents in different styles, and tones, to show a LLM answer questions based on Council data, to show a tool summarize texts like cabinet papers, to translate languages, or create a simple application or workflow brings these tools to life, and generally will leave our audience in stunned silence as they contemplate the broad use cases that could be enabled with this technology.

In terms of Microsoft solutions that we are seeing councils leverage, these LLMs and Generative AI capabilities flow through in several different ways, which I will provide examples of to bring some ideas to life.

The first area we are quickly seeing councils stand up is Bing Chat for Enterprise.? This is free to any council using Microsoft365, and allows councils to access Chat GPT Language models in a secure, council owned environment where any data the Council uses to ask questions does not leave the councils environment, and is not used to train the language model.? This creates a safe space for employees to begin to use Generative AI, and we are seeing councils stand up policy to prevent staff using Chat GPT natively in favour of Bing Chat for Enterprise.

The second capability we are seeing Councils leverage is through Microsoft Copilot.? Microsoft Copilot brings the capabilities of LLMs and Generative AI into the Microsoft365 suite of products, with some incredible capabilities.? For example, in a Teams meeting, Copilot will provide a summary of the meeting, what was discussed and distribute actions.?

In both outlook and teams, you are able to ask Copilot to summarise missed messages or email threads, provide highlights from them, and any actions you own.? You are able to ask Co-Pilot for a summary of your interactions with a colleague, or to summarise discussions around a topic or case, and it will provide a summary of this.

In Word, we are able to rapidly speed up content creation or summarise documents, and already we are seeing Councils use this capability to speed up the creation of policy, to summarise document packs to go to councillors for cabinet meetings, and create Easy Read letters, which with Offices Translation capabilities can then be translated into other languages – including Welsh for any authorities with a statutory requirement to communicate in Welsh.

In our Low Code toolsets, we are able to use Co-Pilot to quickly create forms and applications which can be rolled out at a hugely increased pace, and there is a lot of excitement from councils starting to leverage this capability to rapidly cut their application request backlog.

With these capabilities built into the Microsoft365 tools, it is understandable that Local Authorities are rapidly looking to experiment with these capabilities, and we now have 15 councils who are using Co-Pilot.? In order to help with sharing of learnings and business cases, we have stood up a sub group to the Innovation and Collaboration forum where we have over 30 councils now collaborating, sharing ideas and looking at policy creation, chaired by 2 councils.

Finally, there is also the capability to develop applications, infused with these Large Language Models and Generative AI.? Through Azure Open AI, we are able to allow councils and partners to securely connect to services from the likes of Chat GPT and DALL-E and others for their own use cases, and again, we are seeing some really interesting use cases begin to bubble up from this.

We currently have projects going live across the UK with a wide range of examples in Local government using these capabilities, including large scale document ingestion, document translation services where the cost of translating a document has decreased from ￡120 for a human to complete to 20p.? We are seeing use cases for FOI handling, automation of Invoice processing, AI for Contact Centre enhancement, from using bots to deflect calls to transcription services to automate case notes creation, and image recognition services being used to identify potholes, graffiti, fly tipping and even being used to try and identify cases of RAAC or flammable cladding in buildings.

These use cases have really only come to the fore over the past 6 months as the technology has come along rapidly, and I am excited to continue to explore the incredible opportunities these technologies can provide councils in future blogs.