According to Nature, scientists have produced a research paper in less than 1 hour with the help of ChatGPT (a tool guided by artificial intelligence that can understand and produce human-like text). The paper was smooth and structured in the expected structure of a scientific paper, but the researchers say there are many hurdles to overcome before the tool can be truly useful.
For this research, scientists used a data set that included information on diabetes status, fruit and vegetable consumption, and physical activity on 250,000 people.
The researchers asked ChatGPT to write code that they could use to discover patterns in the data so they could perform the analysis. On its first try, the chatbot generated code that was riddled with errors and didn’t work. But when scientists sent error messages and asked it to correct the mistakes, it eventually produced code that could be used to explore the dataset.
With a more structured data set in hand, the scientists then asked ChatGPT to help them develop a study objective. The tool suggested how physical activity and diet affect diabetes risk. When ChatGPT generated more code, it produced these results: Eating more fruits and vegetables and exercising are associated with a lower risk of diabetes.
ChatGPT was then asked to summarize the key findings in a table and write the entire results section. Step by step, they asked ChatGPT to write the abstract, introduction, methods and different sections.
Finally, they asked chatgpt to edit the text. Apparently, this article was excellent in the form of a scientific article.
One of the problems the researchers encountered was ChatGPT’s tendency to fill in the gaps with fake things, a phenomenon known as hallucinations for the bot. This chatbot generated fake citations and false information in this case.
Scientists also worry that such tools could make it easier for researchers to engage in dishonest practices.
Another concern is that the ease of producing articles with artificial intelligence generating tools could lead to journals being filled with poor quality articles.
Vitomir Kovanovic, who develops AI technologies for education at the University of South Australia in Adelaide, said AI tools should be seen more often in research papers. Otherwise, it will be difficult to evaluate the validity of the findings of a study.
Massachusetts scientist Shantanu Singh says generative AI tools have the potential to speed up the research process by performing simple but time-consuming tasks (such as summarizing and generating code). Artificial intelligence can be used to generate articles from datasets or to develop hypotheses, he says. But since it is difficult for researchers to detect illusions and biases, it is still not a reasonable task to write full articles.
Leave a Reply