AI for Metsänhoidon suositukset – AI solution that answers complex topics in an understandable way

TapioOpens in a new tab provides independent forest and nature-related information, expert services and solutions for the sustainable use of forests to government, private sector operators and individuals. The services bring climate and biodiversity considerations alongside economic values.

Metsänhoidon suositukset (the Best Practices for Sustainable Forest Management) is a service commissioned by the Ministry of Agriculture and Forestry, and is maintained and coordinated by Tapio. The recommendations are prepared together with forest, environmental and climate actors. The service is aimed at Finnish forest owners and forestry operators. It will be widely used as it provides scientifically based options and best practices for sustainable forest management.

From 2020 onwards, the information will be distributed in digital format. The recommendations are openly available free of charge to all, including commercial operators, at www.metsanhoidonsuositukset.fiOpens in a new tab. Through the open interface of the information service, the recommendation data can be transferred to other information systems in the desired format and scope, thus enriching e.g. forestry e-services, applications or learning materials.

The aim of the meeting was to pilot the use of artificial intelligence on the Metsänhoidon suositukset website. Historically, one of the main challenges in using the website has been the scale of the information and the diversity of user groups. It has been a challenge to find, assimilate and understand the information, especially for people who don't have previous knowledge on these complex topics. For example, information on climate resilience is difficult to find when the subject has been covered in many different articles and the range and complexity of the topic is wide.

In recent years, the recommendations on forest management have been thoroughly updated from the perspectives of climate, nature management and water protection, among others. In parallel to building the AI solution, Tapio was engaged in a broader design of the service and wanted to involve users in the evaluation of the AI solution concept.

Before the AI search, the website already had a subject search. The search can be used with a single term and the result is a whole series of articles where the term is mentioned at some point. The search requires knowledge of the correct term in order to return relevant search results. For the user, the number of results is usually too large and browsing directories and headings is preferred to the search function.

In order to make the information easier to find and understand, AI-assisted search was implemented. The search will help new users in particular. It is now possible to find the answer to questions that are much broader than just a single sentence or a single paragraph of text. This makes it easier to assimilate and make use of the extensive recommendation data.
The key to the technical development of the AI solution was that the language model does not form an answer at all if there is not enough information.

"The hallucinations of the AI has been successfully limited. It doesn't make up answers out of thin air. If there is not enough information, it will inform you and guide you to reformulate your question," says Tarja Ollas, Marketing and Communications Specialist at Tapio.

The newly released AI solution searches only the content of the recommendations, so the answers are based on the best research and practical experience of sustainable forest management. The AI not only formulates the answer, but also provides links to relevant articles.
Service Area Manager Kalle Vanhatalo from Tapio encourages people to try and use the new search:

"AI understands the meanings of questions, which traditional word searches cannot. It helps find relevant content to read more about in the original, expert-generated form."

Technical implementation

The project, delivered by Wunder, involved:

1. Expansion of the existing web service (Elasticsearch) search into a semantic search. The advantage of semantic search is that it helps to find information based on the meaning and contextual relevance of the search query instead of the hit of search terms. Elasticsearch semantic search functionality is based on the natural language model (NLP).
2. Integration into the interface of the selected AI language model (OpenAI). The natural language model needs a prompt that is provided with each query offered to the language model and based on which the answers are generated. If the generation of answers is not fit for purpose as such, the answers can act as a "helper" for at least inexperienced users, who can be guided through the site by links.
3. Design and interface implementation for entering the questions and presenting the answers received from the AI.

OpenAI was chosen as the provider of the tool and the service currently uses the GPT 4o-mini language model.

The AI solution has been developed under the leadership of Tapio and funded by the Ministry of Agriculture and Forestry through the Carbon Capture Climate Programme.

Augmented search was launched in early October 2024, so in just a few months, a collaborative effort was made to build an AI search function that meets the needs of Tapio and service users and leverages Wunder's solutions and innovations – that doesn't hallucinate! This will enable the exploitation of very broad and reliable data for a wider audience. In particular, the AI search will benefit all those who are not yet familiar with forestry topics.