In a recent article, two notable chemists, Audrey Moores from McGill University and Vânia Zuin Zeidler from Leuphana University, have expressed apprehensions regarding the application of generative artificial intelligence (GAI) for creating chemistry-related visuals. This encompasses molecular structures, where inaccuracies could adversely affect science education. Their collaborative piece, published in *Nature Reviews Chemistry*, calls for prohibiting GAI in the generation of molecule and element illustrations due to the prevalence of observed discrepancies.
Moores was especially motivated to voice her concerns upon noticing incorrect chemical structures on the cover of a journal that displayed AI-generated images. She shares her experiences with tools such as ChatGPT and Copilot, as she tried to produce visuals for her inorganic chemistry course. The outcomes, she notes, were unsatisfactory, with the AI failing to accurately depict a periodic table and chemical structures despite significant efforts in adjusting prompts.
Moores stresses that although GAI can have beneficial applications, the problem arises from its misuse without appropriate oversight, resulting in misleading chemical representations. Her worry is that this could mislead audiences, as initial impressions from GAI outputs might seem accurate but necessitate further examination to reveal flaws.
The article contends that unregulated use of GAI in chemistry could transform educational methods and the community’s perception of itself, since AI-generated material does not originate from well-informed scientific evaluations. Moores provides instances where GAI generated incorrect representations, like incomplete benzene rings, advocating for a suspension of AI-generated molecular illustrations until enhancements are implemented.
Major publishing entities, including Nature, the Royal Society of Chemistry, Elsevier, and Wiley, have implemented guidelines or restrictions on the use of GAI for scientific imagery, highlighting concerns related to accuracy and licensing. These policies differ in their rigor, most stressing the importance of transparency and proper attribution when AI is utilized.
Moores and Zeidler promote the need for more stringent actions than those offered by current policies, calling for a ban on GAI in the production of molecular and elemental visual content. They caution that permitting inaccurate AI to represent chemical ideas could distort science, emphasizing the obligation scientists have in accurately portraying reality and educating upcoming generations.