2023-07-31 11:28:22
researchers of the CREATE and the CSIC have revealed that ChatGPT does not reach human creativity in research on plants, although it can speed up the work. In a new study, the results of which are published in the journal Trends in Plant Scienceit is shown that the artificial intelligence (AI) chat helps to formulate scientific questions about plants, although the proposals are limited and do not consider aspects such as the climatic urgency, the political context, more recent studies or the need to raise multidisciplinary investigations.
In view of these results, the team of authors warns that the list must always be enriched with the human gaze. “We have detected a bias towards the most abundant content. So, if we only rely on this technology, we risk narrowing the scientific perspective,” he says. Josep PeñuelasCSIC researcher at CREAF and one of the authors.
The list must always be enriched with the human gaze since we have detected a bias towards the most abundant content
Josep Peñuelas, CSIC researcher at CREAF
An example of this bias is that, according to ChatGPT, the number one priority for research should be to develop sustainable products with plants, “we believe that the reason is that there is a lot of content on the web that relates plants and phytosanitary products, medicines, clothing, etc.”, says Peñuelas.
The 100 Critical Issues
To carry out the research, the authors asked ChatGPT the 100 critical issues that plant science should focus on in the coming years. The answers were compared with an international scientific study in which these hundred questions had already been agreed upon.
“Our goal was to assess the differences and find out to what extent the results were similar,” explains the co-author.
When they compared the two lists, they found that both included topics such as plant-pollinator interactions, nitrogen fixation, or using plants to make sustainable products. However, the ChatGPT did not include questions that the researchers consider highly relevant. Between them, issues related to climate change and carbon sequestration“very present in the social and political agenda”, says Peñuelas.
ChatGPT did not propose cutting-edge topics such as the study of the immune system of plants, which can contribute to developing ‘plant vaccines’ against pathogens, insects and other stressors
Nor did he propose cutting-edge topics such as the study of the immune system of plants or primingwhich can help develop ‘plant vaccines’ against pathogens, insects and other stressors or investigate the potential of seaweed to reduce pollution “Probably because there is not much information published yet,” Peñuelas clarifies.
On the other hand, they also consulted with the ChatGPT the 10 most relevant topics in plant science and where research should focus in the second half of the 21st century and the 22nd century. He responded differently to each question, “this shows that he is capable of giving different answers depending on the scenario,” says Peñuelas.
The limitations of basing science on AI
The research results also reveal that the ChatGPT does not take into account issues that require a deeper understanding of the world “something essential for science to advance.”
The application does not mention the importance of transferring scientific results to politics or maintaining genetic biodiversity among vegetation, essential in a context of climate crisis.
For example, it does not mention that it is important to transfer scientific results to politics; strengthen the link between science and society; maintain genetic biodiversity among vegetation, essential in a context of climate and biodiversity crisis, or the importance of science collaborating with other disciplines, including architecture or engineering to green cities or the agricultural sector to address food security. “These nuances, for the moment, he still cannot appreciate,” Peñuelas points out.
In addition to CREAF and the CSIC, the University of Nanjing, the University of Berlin and the Institute for Advanced Research on Biodiversity of Berlin-Brandenburg (BBIB) have participated in the research. The authors conclude that ChatGPT can save time when reviewing and analyzing the bibliography, but warn that should be used with great caution and always counting on the vision of the scientific community.
Reference:
Agathokleous, E., Rillig, M.C., Peñuelas, J., Yu, Z. “One hundred important questions facing plant science derived using a large language model”. Trends in Plant Science (2023).
Rights: Creative Commons.
#ChatGPT #underestimates #climate #urgency #plant #inquiries