AI in dentistry – from fake research to treatment tool
Authors: Asst. Prof. Ilze Maldupa, Lead Researcher at Rīga Stradiņš University (RSU) Institute of Stomatology
Assoc. Prof. Sergio Uribe, Lead Researcher at RSU Department of Conservative Dentistry and Oral Health
Artificial intelligence (AI) is entering medicine. We conducted a study on how to bring effective, research-driven methods to dentistry¹ and looked in more detail at the opportunities that AI offers. Having studied the use of AI in more depth, we noticed how it has influenced the of scientific publication system. As with everything involving AI, these tools can be supportive in a scientific environment, but there are also risks associated with their use. What are the risks, and where can we benefit from AI?
We analysed nearly 300,000 research abstracts published between 2018 to 2024. As we know, AI is “fed with information”- a lot of data needs to be invested, and this requires manpower. Part of this manpower comes from Africa where there are many countries that were once British colonies. English is therefore the native language for many of the people working on AI, but it is a different English with terms that are not commonly used in other English-speaking countries.
Photo: Shutterstock
By using a so-called bibliometric approach, we were able to identify what content was generated by ChatGPT. For example, there are words that were hardly ever used in publications before AI software was released. Now their proportion has grown. For instance, the usage of the word “delve” has increased 17 times in frequency since the more widespread release of AI than before. Other words (e.g. “transformative”, “realm” and “revolutionize”) have also become much more widespread. These results show that almost a tenth of publications contain specific words, indicating a significant rise in the use of AI in producing scientific research.
It should be noted that there were more errors in the first years of ChatGPT, many of which have now been corrected. Even last year when this study was carried out, the keywords mentioned were still present in the AI tool responses.
Our study does not allow us to conclude whether the use of AI makes the research in question inferior or untrustworthy, or whether AI was used only to correct language.
The use of AI to improve grammar is normal practice in scientific research. In a sense, there is even unfair competition between a researcher who, has English as their native language and a researcher for whom English is not their first language, but the publication has to be written in English.
It is then acceptable to use AI to correct language, preferably with an honest disclosure. However, publications translated entirely with AI translation tools may sound strange or even illogical, raising doubts about the reliability (validity) of the research.
Unfortunately, we know that nowadays there is a lot of poor quality writing² and that AI is used in bad faith. Many thousands of scientific papers have been written and published with AI tools since they were introduced not so long ago.
It is true that one of the most important tasks in research is to communicate your findings and conclusions to the public. A single scientific publication consumes a lot of time and energy, however, for example, clinical or cohort studies can take several years to be published, while an AI-produced "scientific article" can be created almost instantly.
The main aim of research used to be to produce new knowledge, whereas currently there are researchers who are motivated to "produce" publications to be competitive in the labour market, or to climb the academic career ladder, or to get research funding. Such motivations do not align with public interest. The number of journals has spiked in geometric progression. The number of universities and scientists is also growing, but there is a limited number of quality reviewers who can professionally assess the quality of a potential publication. And we are not just talking about journals of minor importance!
There are even comical cases where a scientific publication has not deleted the AI-generated text (a feature of earlier versions of AI software), which allowed choosing a different response: 'regenerate response'.
In the history of science, there have also been cases where publications have been retracted, for example, because of plagiarism or fraud. For instance, there was a popular study about vaccines causing autism. The publication was retracted³ because it was revealed that the findings were selected in a scientifically incorrect manner, but the harm done to society is no longer retractable.
Even nowadays, there are publications that are retracted for various reasons (website with list of retracted scientific papers). This is most often the case for publications created in China, Russia, the Middle East and the USA. Certainly, many of these studies do not attract any attention and are not read anyway, but the credibility of the industry as a whole is diminishing.
Source: LSM.lv
Read the full article on LSM.lv
¹ Fundamental applied research project Implementation of the Evidence-Based Paediatric Caries Management Strategies in Latvian Clinical Practice - an Evidence Transfer Study
² Lex Bouter. 2024. "Fake academic papers are on the rise: why they’re a danger and how to stop them. The Conversation" and Rizqy Amelia Zein. 2024. "Paper mills: the ‘cartel-like’ companies behind fraudulent scientific journals." The Conversation
³ Dr AJ Wakefield. 1998. Retracted: "Ileal-lymphoid-nodular hyperplasia, non-specific colitis, and pervasive developmental disorder in children" and The Editors of The Lancet. 2010. Retraction: "Ileal-lymphoid-nodular hyperplasia, non-specific colitis, and pervasive developmental disorder in children"