Saffron O’Neill, Veronica White, Tristan J. B. Cann, Felix M. Simon, Alastair Johnstone-Hack, Simon Puttock, Sylvia Hayes, Oliver Blewett, Chico Q. Camargo, Travis Coan, Francisco Gonzalez Espinosa, Ranadheer Malla, Sarwat Qureshi, Ned Westwood
Visualising climate change is a challenge for journalists. Climate images can be clichéd, disconnected and even divergent from information communicated by other modalities (such as text); and can represent a missed opportunity to engage audiences. Generative AI could potentially provide transformative opportunities to visually illustrate climate news. However, risks may include reinforcing stereotyped, inaccurate or unethical visual content, with knock-on effects on audience trust and public informedness. Working as a transdisciplinary team of social scientists, computer scientists and journalists, we co-produced three studies designed to represent a spectrum of potential future engagement with generative AI technology. We prompted five generative AI tools (ChatGPT, Stable Diffusion, Midjourney, Microsoft Copilot, Generative AI by Getty Images) with seven typical climate news topics (climate change, flooding, heatwave, migration, Net Zero, COP26, nature) via three prompt styles (simple, detailed, story). These results indicate that there are substantial barriers for generative AI tools to create meaningful images for climate-related stories. We reflect on the implications of these results for both researchers and journalists; in terms of the evolving eye-witnessing role of photojournalism in an age of AI and misinformation, and in how such images may act to constrain transformative climate action.