Will Artificial Intelligence be Useful (or Misused) in Environmental Toxicology and Chemistry?

Like some of you, I was introduced to computers in the early 1970s by learning “Formula Translation” or FORTRAN. It was an old programming language that required typing punch cards that ran on a mainframe computer. I still recall checking my program for “do-loops,” while also worrying about dropping the cards and having to resort them. I prepared this Points of Reference after reading about artificial intelligence (AI) and the many recent improvements in its capabilities. I was intrigued by learning how some college students were using the improved AI tools to complete their writing assignments with less time and effort. That led me to the question posed in the title and to consider what it might mean for the Society of Environmental Toxicology and Chemistry (SETAC). What is “artificial intelligence” or AI? Simply put, “AI is a computer system that is able to perform tasks that ordinarily require human intelligence…” (Achin, 2017). Moviegoers might consider Skynet as an example—the computer/program that becomes self-aware and manufactures Terminators that harm humans. But today, examples of AI include self-driving cars and so-called chatbots, which can write documents nearly indistinguishable from those written by a human. There are other definitions, more comprehensive than this simple one; but this one will suffice for this discussion. Has SETAC been involved with AI? Clearly, it and other scientific societies have. For example, using the search term artificial intelligence from 1989 to 2023 some 41 papers were identified in Environmental Toxicology and Chemistry and from 2009 to 2023 another 36 in Integrated Environmental Assessment and Management. As one might expect, most of these publications focused on gaining an understanding of the underlying patterns in large, complex data sets, including developing structure–activity relationships among chemicals and constructing models to predict the environmental fate or toxicity of various chemicals. Similarly, a recent call for papers on AI and “machine learning” was found, appearing in the journal Environmental Science and Technology (EST; Lowry et al., 2022). The call for papers included a wide variety of AI subtopics that could be of interest for a special series of papers, including several focused specifically on environmental toxicology and chemistry. Based on the SETAC publications and the call by ES&T for a special series on the topic, using AI or AI-type tools has been ongoing in environmental toxicology and chemistry for some time. Maybe it was not identified as AI per se, and the software/program did not become self-aware or drive a car; but it meets the simple definition provided earlier. The above applications for analyzing large data sets or developing fate and effects models do not appear alarming, yet the more recent developments and improvements in AI, especially in the form of chatbots that can write documents, should give us some concern. What is a chatbot? A simple example is the little pop-up “chat” window that appears when accessing information or asking a question online from an organization. However, since November 2022 a much more powerful chatbot for writing and other tasks was posted as an open-access/free program—it is called “ChatGPT” (Vanian, 2022). Over the past few months, ChatGPT has become popular among college students, seeking a way to satisfy writing assignments in an expedited manner. This led me to posit a question: Could one take ChatGPT to the point of writing a manuscript on data that were not developed by the author, whether from the laboratory or from field studies? In the past this was called “dry-lab” work—in simpler terms, it would be “made up.” How would editors and the editorial staff at our journals know if the manuscript that came in was not derived from a chatbot? Carrying this a little further, how good is our ability to ascertain whether abstracts submitted for our annual SETAC meetings are “real” or not. In other words, could someone develop a well-written abstract citing data that they did not generate? Maybe this is not a new concern because an individual could do this without the help of a chatbot. Although our two SETAC journals have safeguards in place to check for plagiarism in submitted manuscripts, it is not clear whether these safeguards will also detect chatbot-derived manuscripts. Given the increasing power of the chatbots to mimic human writing, it is an issue that is not likely to disappear until there are tools in place to address it. This is something that SETAC should consider, if it has not already done so. Perhaps chatbot-detecting programs should be explored and evaluated for their capabilities to address these concerns (“Best AI Content Detection Tools,” 2023). In whatever form it might take (see Sills, 2017), AI has considerable potential to both help and harm various aspects of human endeavors. It can expedite drug development and medical diagnoses, improve climate change models, and perhaps do great good in many other areas (Gill et al., 2014). On the downside, it could be a tool to spread disinformation or misinformation and, specific to SETAC, used inappropriately in preparing materials for our annual meetings or manuscripts for our journals. This is an issue that could challenge the SETAC leadership, the editors, and the editorial staff of our journals and our membership. Now is the time to address these concerns if they have not been addressed already. The author declares no conflict of interest. Ralph G. Stahl, Jr.: Conceptualization; Data curation; Writing—original draft; Writing—review & editing.