Generations: Understanding Deep Geological Repository Stakeholders through Large Language Model Personas

This report provides a technical description of Generations, an Interactive.li-based serious game aimed at exploring and understanding stakeholder roles in Deep Geological Repository (DGR) siting with Large Language Model (LLM) personas.

In the game, participants assume the role of a policymaker, making decisions at key stages of the programme and receiving feedback from simulated stakeholder groups. Stakeholder responses are generated by an LLM trained on contextual information, government policies and strategies, as well as each group’s specific interests, expectations, and legal responsibilities.

The aim of the game is not to identify optimal stakeholder engagement strategies, but to raise awareness of the complexity involved in long-term governance, the need for trade-offs, and the importance of aligning diverse stakeholder perspectives.

Generations was designed to spark dialogue, promote critical thinking, and encourage reflection on how decisions are framed and negotiated. To enhance the realism and relevance of the simulation, the game is grounded in the Swiss context and publicly available sources. Switzerland’s participatory approach in radioactive waste governance provides a compelling example for exploring stakeholder dynamics in depth.

The learning experience is also intended to be playful and engaging. As they progress through different stages of the game, players collect points based on how well their decisions satisfy the expectations of different stakeholders. The winner is the player with the highest score at the end. However, players must also avoid receiving a stakeholder veto, which results in an immediate “game over”.