3.10 Case study: AI for science communication

Effectively communicating complex environmental science to diverse audiences – from policymakers to the general public – is a critical component of driving change. This case study provides a practical guide to using large language models (LLMs) like ChatGPT, Google Gemini, or Claude as a “digital scribe” to make this communication more accessible and impactful, while navigating the significant ethical risks involved.

LLMs offer several clear benefits for science communication. They can rapidly summarise long, dense research papers, making the key findings accessible to non-experts. They are powerful tools for improving the clarity and readability of drafts, which is particularly helpful for researchers who are non-native English speakers. One study experimentally demonstrated that AI-generated summaries of scientific articles were perceived as clearer and easier to understand by the public than the original summaries written by the scientists themselves.

However, the pitfalls are equally significant and demand a vigilant, “human-in-the-loop” approach. The most prominent risk is that of “hallucinations,” where an LLM confidently fabricates information, including plausible but non-existent citations. One analysis found that 47% of citations generated by a recent version of ChatGPT were fake. LLMs can also perpetuate biases present in their training data and can oversimplify complex topics, stripping them of crucial nuance. The unreliability of information access, as evidenced by firewalled or unavailable web pages, underscores that one cannot blindly trust the sources an AI might claim to use.

Therefore, the ethical and effective use of LLMs in this context requires adherence to a strict set of best practices. The LLM should be used as a tool to assist, not replace, the human communicator. It can be used to brainstorm ideas, generate a first draft, or suggest alternative phrasing, but the final content must be the product of human intellect and judgement. Every factual claim, statistic, or citation generated by an LLM must be rigorously verified against original, reliable sources. Finally, transparency is paramount. When an LLM has been used significantly in the writing process, its contribution should be explicitly acknowledged, just as one would cite any other tool or collaborator. By following these guidelines, communicators can leverage the efficiency of LLMs while upholding the scientific and ethical integrity of their work.