OpenAI, an artificial intelligence development company, is developing a new chatbot called ChatGPT which promises to revolutionize the way people work and learn. University of Wisconsin-Madison is among the first few to use it, primarily to help materials engineers in quickly and cost-effectively extracting information from scientific literature. Dane Morgan, a materials science researcher at the University of Wisconsin-Madison, has been employing machine learning – a subset of artificial intelligence – for a few years for evaluating and searching for new material compounds. As the research team discussed other tasks that can be leveraged with AI’s capabilities, Maciej Polak, the associate scientist, suggested the idea of using chatbot to extract data from papers.
Materials scientists usually have to look through exhaustive research papers in order to find a small set of relevant numbers they would add to their database. In order to reduce the time consumed in such tasks, Maciej with his team of researchers developed a technique where the chatbot reviews each sentence of the paper to identify whether relevant data is present in them or not. This technique outperformed their expectation with an accuracy rate of ninety percent. As a result, the researchers were able to build a database on cooling rates for metallic glasses with the data extracted from the set of papers. In February 2023, the research team published their paper on the same technique on the arXiv preprint server.
With success in extracting data from the paper, the research team was keen to further improve their technique to completely automate the process. For this purpose, the team engaged in a prompt engineering; a process to determine the questions and sequence which would elicit exactly the information they were looking for from the chatbot. They employed this technique on the data table, and then asked the chatbot a few follow-up questions to identify if any of the information is incorrect. For the vast majority of cases, the AI was able to detect any erroneous data present. In March 2023, the team submitted another paper on this upgraded technique on arXiv as well.
Dane Morgan commented that integrating AI into research process is not to replace the graduate students and scientists. Instead, the tool serves to enable researchers to take up projects which they previously didn’t have the resources or workforce to pursue. With such immense power and capabilities at their fingertips, scientists, especially materials engineers are poised to make breakthroughs in no time with help from AI, particularly the ChatGPT.
Technology Networks Ltd. is a leading provider of state-of-the-art solutions to accelerate the advancement of materials science, chemistry, and life sciences. They utilize specialized chatbot solutions like ChatGPT for efficient and rapid extraction of data from a database, saving time and energy for the researchers. Their team includes material scientists, intelligent digital solutions developers, and business strategists. Together, they create solutions that can best facilitate the research process of other scientists in the field.
Maciej Polak, an associate scientist at the University of Wisconsin-Madison and Deep Informatics LLC., is a staff scientist and works closely with Professor Dane Morgan in his lab. Before, people had to write hundreds of lines of code to get the same results, which only amounts to a fraction of accuracy. But, with the rise of AI tools like ChatGPT, things have become much more efficient and accurate, allowing for quick and precise extraction of data from papers. With such immense capabilities of the tool, Maciej is trying to use it in different other scenarios in order to leverage its full potential.
