The rapid pace of development in artificial intelligence has caught the attention of academics and industry alike, with innovative ways of harnessing AI for the benefit of both. ChatGPT, a large language model from OpenAI, is no exception. Researchers at the University of Wisconsin-Madison are already leveraging its power for materials engineering and have reported an impressive 99% reduction in the time needed to read papers to extract data and tabulate it.
Dane Morgan, a professor of materials science and engineering at the university, has long used machine learning models with considerable success. His staff scientist, Maciej Polak, wondered if AI could be used to off-load even more labor-intensive activities, such as sifting through scientific papers for data. After developing a technique to do this, Polak found that the chatbot could review sentence by sentence and decide whether each contained relevant data or not; when asked to present the information in a table, the accuracy rate was around 90%. As reported in a February 2023 paper posted on the arXiv open access preprint server, Polak and colleagues used the machine learning models to construct a database of critical cooling rates of newly developed metallic glasses.
To further optimize the process, the team introduced follow-up questions in sequence to introduce the possibility of data errors. Surprisingly, the AI was able to recognize and flag mistakes in the vast majority of cases. This improvement was detailed in a March 2023 paper on the same journal.
As of now, the research team is finding ways to enlarge the scope and accuracy of this technique with ChatGPT. While this technology is not intended to replace researchers, it can potentially save them a great deal of time while pursuing projects they would not have enough resources, time or money to otherwise pursue. Similar to the way Google and other search tools revolutionized research, ChatGPT could potentially change the way we do research yet again.
OpenAI is a San Francisco-based artificial intelligence and machine learning research company founded in December 2015 by Elon Musk and Sam Altman. It has quickly become synonymous with the development of powerful AI models, including GPT-3, which powers numerous applications and services, and its research has been widely cited and utilized by organizations, universities, and companies around the globe.
Maciej Polak is a Staff Scientist who works closely with Professor Dane Morgan at the University of Wisconsin-Madison, and specializes in using machine learning to assess and search for new materials. He has researched and developed techniques to make chatbot technologies, like ChatGPT, more efficient in collecting and verifying data by utilizing language prompts. Furthermore, he has recently published papers on the arXiv journal both alone and in collaboration with researchers. His work is particularly interesting as it promises to further automate tedious and labor-intensive research tasks.
