Naturvårdsverket
Simplifying environmental guidance with AI
Tags
The Swedish Environmental Protection Agency (Naturvårdsverket) is dealing with an increasing number of questions from supervisory authorities looking to understand complex environmental regulation. We worked with the agency to test AI’s potential to ease the burden on staff and speed up answers.
In just two weeks, our data scientists and AI experts developed a prototype tool that proves AI’s ability to search and interpret highly detailed regulatory text and generate meaningful guidance. During further studies with Naturvårdsverket, we confirmed AI’s potential and how it paves the way for future initiatives to improve services and strengthen inspection practice. This would in turn contribute to making it easier for companies to comply with complex environmental regulation.
Improving services with a faster response
To combat global warming, loss of biodiversity, and the spread of persistent pollutants, among other environmental threats, environmental regulation is increasing on both EU and national levels. As a result, the supervisory authorities – subject to regulation – often need help to understand the rules.
For the Swedish Environmental Protection Agency (Naturvårdsverket), this is a growing challenge. The agency is receiving more and more queries from organisations seeking advice on how to comply. The team of case management officers at Naturvårdsverket also provides guidance to more than 1,000 municipal and county level inspectors working in the field.
Starting small to kick-start innovation
Working closely with the agency, we launched a small project to test whether AI could search and interpret regulatory text and generate meaningful guidance. We used a set of 100 documents already in the public domain to train a prototype AI tool. This approach minimised security risks and enabled us to move fast. Beyond this prototype, the tool is capable of ingesting significantly more data and of identifying where guidance is different across documents or has changed based on new information.
Using our custom-built Generative AI platform, our data scientists and AI experts developed the first version of the new AI tool within just two weeks. The platform utilises large language models (LLMs) to analyse and learn from Naturvårdsverket’s knowledge stores. Staff are then able to query the tool using natural language to receive contextually meaningful results in a form that is easy to understand. The tool also provides refences to the data used to construct the basis of its answer to enable staff to verify results and learn as part of the process. We followed up with live testing with the Naturvårdsverket team that enabled refinements and ensured the AI-generated guidance was clear and accurate. Over two months, we continued to develop the prototype, including enabling the tool to interpret regulations and respond to queries in Swedish.
This work builds on our expertise in AI, having partnered with Unilever to empower its R&D teams to address changing consumer needs and PostNord to build a generative AI chatbot to support its logistics teams find more efficient transport routes.
Thanks to our long-standing relationship with the agency, we were trusted to address concerns around AI’s impact in the workplace.
Proving future potential for the public sector
By the time the project concluded, the tool was producing impressive results. So Naturvårdsverket invited several government agencies and municipalities to view a demonstration and consider AI’s potential to help generate guidance on complex rules and procedures.
For our part, this was a great opportunity to demonstrate how a ‘start small, scale fast’ approach is an effective way to begin innovating with AI. “We’re proud to be working with the agency to help them innovate with AI to realise their mission,” adds PA delivery expert, Erik Mossberg. He concluded, “Naturvårdsverket’s environmental focus is closely aligned with our own purpose – to build a positive human future and contribute to a sustainable world.”