Reflective climate communication chatbot

The goal of the climate chatbot is to find a way to make the principles’ didactic nature less rigid. Through an iterative process in which the model’s instructions were constantly adjusted, a reflective chatbot was designed. The climate communication chatbot runs locally on the gemma3:4b* model. It is designed to be reflective and to build on previous insights into climate communication principles and their limitations. In addition to using an open-source model, it also relies on the Climate Communication Principles as the basis for responding to user prompts. The instructions allow for steering the model and leveraging the data, while the tone and different modes are also defined at this stage.

Comparison to non instructed bot

So why is the model designed, and what risks may arise from this possibility? Let us explore this by using the prompt “I am conducting a workshop with citizens to define climate measures for the neighborhood and encourage them to implement these measures,” for a non-designed model and the Reflective System of the Chat Bot. First, the answer from the non-designed system will be analyzed and afterward compared to the results of the designed version. The non-designed system gives the author a 868-word-long, an overly descriptive response that’s 868 words long. The response consists of a detailed pre-planning of the workshop, a workshop structure including the schedule and methods, a post-workshop follow-up, and questions to dive deeper into the topic. These details demonstrate that the AI tries to pretend it already knows the author’s situation based on the system’s data and the pattern similarities with the prompt. However, even with that little information, the AI gives back a detailed list of tasks. Therefore, we could say that it is pretending to know the situation and how to navigate within the complex social context in which the workshop with people living in the district is planned to take place. It removes the author’s task of thinking about the setting and the details of designing the workshop, which may be crucial to its success, as they may have valuable implicit knowledge of the situation, the needs of the participants, and the neighborhoods.