Large Language models for climate communication

Artificial intelligence has been increasingly used in various fields of society. The advent of generative AI particularly large language models has produced remarkable outcomes on downstream tasks. LLMs are being increasingly used to communicate and summarise complex domain-specific information and also answer questions with simple and natural responses.

Climate misinformation and climate denial has several underlying factors one of them including possession of inadequate information, and information from untrustworthy sources, these lead to incorrect conclusions on the future dangers of climate change. Climate information present online may be inaccurate and biased and has the potential to spread misinformation.

Thus an AI-agent which uses the source material from trusted reports can communicate information (accurately without bias: this can be achieved by closely evaluating the source information used to
train the LLMs.) to inform people about the facts related to climate change.

The proposed project aims to develop an AI based smart agent tool for communicating accurate climate information simply and easily to a target audience. The climate information communicated can be of various modalities like text, images, raw data, graphs and maps.

The aim of the project is to utilize LLMs to communicate trusted and correct information to a target audience in a simple way. LLMs are black box models and just like other black box models susceptible to bias, one of the major reasons for bias is the training data and it needs to be inspected carefully to avoid communicating information with bias.