Smart Interfaces for biodiversity models

Research team

.

Zdeněk Plesěk – University of Exeter

Professor Hywel Williams – University of Exeter

Dr Oscar Rodriguez de Rivera Ortega – University of Exeter

Aims

Smart Agentic Interfaces (SIs) have the potential to get Computational Models ready for use more quickly and allow non-expert stakeholders to make use of them. This PhD research will explore the use of Smart Agentic Interfaces for biodiversity models.  

This work will: 

• Develop an agentic framework (see glossary or footnote) that expedites use of Computational Models (CMs) for an expert stakeholder.  

• Expand the framework to allow a non-expert stakeholder to effectively use CMs to answer questions. 

• Study the impacts SIs have on human skill acquisition and retention. 

Approach

Computational Models are a staple of scientific research. They model potential solutions to represent inherent trade-offs between them or to gain insights about different outcomes without the need for large-scale experiments. Effective use of Computational Models requires technical expertise as well as subject-specific expert knowledge.  

Within RENEW, we are developing a computational model named Ebrel to support land managers’ decision-making. Ebrel focuses on spatial prioritisation, identifying where and what types of habitat should be created or restored to support viable species populations across a landscape. The model has been developed with technical expertise from our environmental intelligence team and subject-specific expert knowledge from our land management team.  

In the context of CM, Smart Interfaces (SI) represent a promising application of Large Language Models (see glossary). Large Language Models (LLMs) generate coherent, human-like solutions to cognitive tasks. SIs empower a non-expert stakeholder by delivering just-in-time contextual expertise required for effective use of CMs – in other words, the initial SIs can convert the natural language of the user into the technical terms needed for the CM to action the task. The next stage of SIs will achieve this reasoning and provide answers to full or partial questions of the stakeholder. 

The full impacts of this technology are unknown. While SIs have the potential to speed up deployment of CMs, shorten the learning phase, and democratise access to CMs for non-expert users, recent research has also cautioned about the negative effects LLM use can have on skill acquisition and retention – could this harm people’s ability to learn and retain skills? 

This PhD research will employ LLMs and to develop a Smart Interface to Computational Models. The research will also study skill acquisition and retention among the users through formal evaluation such as examinations. 

Next Steps

• This work is in its early stages. The next step is development of a simple agentic interface, which will then be used for the initial interrogation of skill acquisition and retention. 

Glossary

Smart Interfaces

A smart interface understands what a user is looking for and brings it forward faster – helpfully anticipating what you need.

Agentic Framework  

An agentic framework enables autonomous systems, such as AI agents, to make decisions, adapt to changes and execute actions without human intervention. 

Large Language Models 

Machine learning understanding of human language – the machine is trained for text generation, translation, summarization, and question answering. LLMs use deep learning techniques and are typically based on transformer architecture, which excels at handling sequential data like text input.