Beyond the hype (and the fear): How AI can deliver value for local authority planners – and how you can trust it

Dr Peter Lawrence, Research and Innovation Lead, Digital Urban

Artificial intelligence will transform business and society in remarkable ways, but – understandably – there’s a lot of fear, uncertainty and doubt about both its effectiveness and the dangers it presents.

When ChatGPT was first released roughly two years ago, it unleashed a huge wave of hype as people were stunned at its capabilities in handling a wide range of text-based tasks, from writing computer code or creating business plans, to composing poetry or recommending the most interesting sights to see in Paris.

But pretty soon people also realised that large language models (LLMs) like this also suffer from ‘hallucinations’ – that is, on occasions they will respond to questions with seemingly credible answers that are entirely false.

A recent example was when Google’s AI Overviews was asked: “How many rocks shall I eat?” Its response was: “According to geologists at UC Berkeley, you should eat at least one small rock per day.” It has also recommended using glue to stick cheese to a pizza and asserted that astronauts have met cats on the Moon.

These may be amusing and relatively harmless, but when it comes to business use, such hallucinations can be damaging. This has caused many organisations to be wary of deploying AI tools – but with the result that they’re potentially missing out on the enormous business value that AI can generate.

So how to create an AI-powered business tool that you know you can really trust and will deliver genuine value for your organisation?

It all boils down to the old adage “knowledge in, knowledge out”. In other words, an AI system’s accuracy relies on the source documents you allow it to access when generating an answer to your prompt. General usage LLMs such as ChatGPT or AI Overviews are using the entire internet, and have no way of truly understanding which information can and cannot be trusted. And as we all know in this age of misinformation, there’s a lot of very unreliable material out there in cyberspace.

But what if the tool was using only 100-per-cent trustworthy knowledge to give you the answers to your questions? And what if it could show you exactly where it had sourced those answers, so that everything it tells you is fully transparent and explainable?

Technology like this could potentially help many organisations to improve their working practices in all manner of ways.

And, with Digital Urban’s Place Co-Pilot, that is exactly what we have achieved by creating a highly effective means for local authorities to speed up and improve adherence to design codes and other similar regulatory documents.

These documents can be very long and full of detail, meaning those who rely on them – such as local authorities and urban developers – may find it challenging to search through them to find answers to specific queries. These challenges can be exacerbated if more than one document needs to be consulted for a project. It may either become too time-consuming, or all too easy to miss the relevant information through human error, often resulting in poor outcomes for all stakeholders.

To solve this problem in an effective and reliable manner, Place Co-Pilot enables users to pose queries in natural, human language, and rapidly receive clear answers.

The key to this tool is that it works in a very different way to, say, a Google internet search or a simple text search of a PDF. Searches of this kind will only give results that are either a list of documents or words that match the text a user inputs – but they won’t necessarily answer a question.

We should also contrast Place Co-Pilot to a general-purpose LLM such as ChatGPT, which, if it doesn’t know the answer to a prompt, can have the habit of making something up –?hence the hallucinations.

To actually answer a question truthfully and precisely, Place Co-Pilot uses the trusted documents we give it –?for example, a local authority’s design codes – and only those documents. It then breaks the information they contain into fragments and synthesises it. So, if a tree is given its English name in one place, and its Latin name in another place, it will know we’re talking about the same kind of tree. It will also know that trees may be relevant to a query about something tangential, such as regulations concerning street design.

Then, and only then, the AI uses all this interconnected information to rapidly create an answer in language that can be easily understood, while allowing the user to check the sources it has used.

We like to think it’s a bit like Stephen Fry. Someone who has access to, and is capable of assimilating, a large amount of knowledge from trustworthy sources, and can then present it in a useful and reliable way that’s valuable for those receiving it.

Place Co-Pilot may not be as witty and entertaining as Stephen, but it’s even smarter and more reliable. And we’re convinced it will lead to vastly improved outcomes for local authorities, those with whom they work, and the general public they serve. Watch this space.

To find out more about how Place Co-Pilot could help your organisation, contact us now for a chat with one of our expert humans.